SciencePediaWhat is information? While we often think of it as abstract data—the ones and zeros of computer code—this view misses a more profound truth. In the physical universe, information is not separate from reality; it is reality. It is the structure in a crystal, the sequence in a DNA strand, and the pattern that distinguishes order from chaos. This article challenges the abstract notion of information, revealing it as a tangible, physical quantity. It addresses the fundamental question of how matter and energy embody and process information to create the complex world around us. First, in "Principles and Mechanisms," we will explore the foundational concept of information made physical, how it builds complexity through hierarchies, and how scientists use it to construct and validate models of reality. Following this, "Applications and Interdisciplinary Connections" will take you on a journey through physics, biology, and computer science, demonstrating how this single concept unifies our understanding of atoms, ecosystems, and even the internet. We begin by examining the core mechanisms that tether the abstract idea of a message to the physical substance of the world.
What is information? The word conjures images of computer code, of zeros and ones flashing across a screen. But that's just one of its many disguises. In the physical world, information is not an abstract ghost in the machine; it is the machine itself. It is structure, pattern, and the very thing that separates a chaotic soup of atoms from a star, a crystal, or a living cell. Information is the answer to the question, "What makes this thing what it is?" It is the blueprint, the recipe, the set of constraints that carves order out of the infinite space of possibility.
Let's step inside a biologist's laboratory to see this principle in action. Imagine an engineered bacterium, a tiny chemical factory designed to perform a specific task. Inside, we have constructed a simple circuit: Device A produces a molecule, Protein A, which in turn switches on Device B, causing it to produce a fluorescent Protein B. When the cell glows, we know the circuit is working. The question is, what flowed from Device A to Device B?
On one level, the answer is simple: a stream of Protein A molecules diffused through the cell's cytoplasm. This is a flow of material. But something else flowed, too: a message. The message was "Turn on!" This is a flow of information. Here, in this microscopic world, we find a profound truth: the material and the information are one and the same (``). The information is not carried by the protein in the way a letter is carried by a mail carrier; the concentration and presence of the protein is the information. The physical substance embodies the abstract message. This is the cornerstone of physical information: it is always written in the ink of matter and energy.
This principle echoes throughout nature. The shape of a key carries the information needed to open a lock. The sequence of nucleotide bases in a strand of DNA carries the information to build an organism. The frequency of a light wave carries the information about the color of a distant star. In every case, the message is inseparable from its physical medium.
If information is physical, then how does nature use it to build the breathtaking complexity we see all around us, from a single cell to a functioning ecosystem? It uses a strategy familiar to any engineer or computer programmer: abstraction hierarchies. Simple components, each with a defined function, are assembled into more complex modules. These modules are then combined into even more sophisticated systems.
Consider the leap from a single cell to a plant. A plant is not just a jumbled bag of cells. It is a masterpiece of organization. Groups of cells with a shared developmental origin and function form a tissue, like the protective epidermis of a leaf. Multiple tissues, such as the epidermis, the photosynthetic ground tissue, and the vascular transport tissue, are arranged in a precise, stereotyped way to form an organ, like the leaf itself—a structure capable of an emergent function (photosynthesis) that no single tissue could perform on its own. Finally, multiple organs, like leaves, stems, and roots, are integrated into an organ system, like the shoot system, linked by conduits of material and information (water, nutrients, hormones) to serve the entire organism (``).
Each step up in this hierarchy—from cell to tissue to organ to system—represents a new layer of informational control. The information is not just in the DNA of a single cell; it is in the spatial arrangement, the communication protocols, and the integrated functions of the collective. This nested structure allows for incredible complexity to arise from a limited set of basic building blocks. The information that defines a "leaf" is a higher-level abstraction than the information that defines a "cell."
The same logic applies to an ecosystem. An ecologist might describe a species' niche as its potential "profession" in the environment—the full range of conditions and resources within which it can survive and reproduce, a concept defined by the species' own biology (``). This is an informational concept, an abstract data space. The habitat, in contrast, is the physical "address"—a specific location with measurable abiotic properties. The community is the list of all species that actually live at that address and interact. The "realized niche" that a species occupies is the result of the intersection of these three things: the species' potential (its fundamental niche), the available physical environment (the habitat), and the influence of its neighbors (the community). The complex web of life is structured by these layers of physical and informational constraints.
Having seen how nature uses information, let's turn the tables. How do we, as scientists, extract information to understand nature? We build models. A scientific model is, at its heart, a compressed bundle of information that we hope captures the essence of a phenomenon. It's a story we tell about the world. A simple model for a quantum system might propose that the probabilities of finding it in three different energy states are , , and , all dependent on a single parameter (``).
But how do we know if our story is any good? We compare it to reality. We go into the lab, perform many measurements, and get an empirical distribution of outcomes, . Now we have two distributions: our model's story, , and reality's story, . We need a way to measure the "distance" between them.
This is where a powerful idea from information theory comes in: the Kullback-Leibler (KL) divergence. The KL divergence, , quantifies the "information lost" when we use our model to approximate the true data distribution . You can think of it as a measure of "surprise." If our model is a good fit, then when we observe the real data, we shouldn't be very surprised. If our model is poor, the real data will seem highly improbable and our surprise will be large. Finding the best model parameter is then a matter of tuning to minimize this surprise. Amazingly, it turns out that minimizing the KL divergence is mathematically equivalent to one of the most fundamental procedures in all of statistics: maximizing the likelihood of the data given the model. The search for the "most likely" model is the same as the search for the model that is "least surprised" by reality—a beautiful and profound link between statistics and information theory (``).
When we have several competing models, say two different weather prediction algorithms, we need a fair way to decide which one tells a better story about the historical weather data (``). The KL divergence is asymmetric, which can be awkward for direct comparison. A related tool, the Jensen-Shannon Divergence (JSD), provides a symmetric, well-behaved "ruler" to measure the distance between probability distributions. By calculating the JSD between each model and the empirical data, we can quantitatively declare a winner—the model whose informational story lies closest to the story told by reality itself.
The world is complicated, and we almost never have all the information. We gather scraps of data and try to build the most reasonable model we can. But what does "reasonable" mean? It means being honest about what we don't know. This is the guiding wisdom of the Principle of Maximum Entropy. It states: among all possible models that are consistent with the data you do have, choose the one that is maximally noncommittal about the data you don't have. In other words, choose the model that has the highest entropy (is the most random or uniform) subject to your known constraints.
Imagine trying to understand the "grammar" of RNA splicing—the rules that tell a cell's machinery where to cut and paste genetic information. We can analyze thousands of known splice-site sequences and easily calculate the frequency of each nucleotide (A, C, G, U) at each position. A simple model, a Positional Weight Matrix (PWM), assumes each position is independent and builds a probability distribution from these single-position frequencies. This is a maximum entropy model if our only information is those single-position frequencies.
But what if we dig deeper and find that certain positions are correlated? For instance, maybe having a 'G' at position 3 makes it much more likely to have a 'C' at position 7, perhaps due to the way the RNA molecule folds to interact with the splicing machinery. The PWM model, assuming independence, is blind to this crucial piece of information (``). The Principle of Maximum Entropy tells us how to proceed: we must find the new model of maximum entropy that is consistent not only with the single-position frequencies but also with these newly discovered pairwise correlations. The resulting model will have terms that explicitly link those positions, capturing the dependency. It is the most honest model because it incorporates all the information we have, but fabricates nothing further.
When we model a complex system like a protein, we face a dizzying amount of information. A simulation that tracks every single atom is called an "all-atom" model. It's incredibly detailed, but also computationally monstrous. Often, to study large-scale processes like protein folding, we must simplify. We must throw information away. This is the art of coarse-graining.
But what information do we discard, and what do we keep? There are two main philosophies (``). The "bottom-up" approach starts with the high-fidelity, all-atom simulation. It then tries to derive an effective force field for a simplified model (where groups of atoms are lumped into single "beads") that best reproduces the structural distributions seen in the all-atom reference. It attempts to preserve the essential physical information flowing up from the finest scale.
The "top-down" approach, in contrast, doesn't look at a more detailed simulation. It looks at the real world. It aims to build a coarse-grained model whose parameters are tuned to reproduce macroscopic, experimentally measured properties, like the density of a liquid or the partitioning of a molecule between water and oil. It only cares about keeping the information necessary to match these large-scale observables. Neither approach is inherently "better"; they are different strategies for choosing which information is most relevant to the question at hand.
This choice about what information to include in a model leads to one of the deepest questions in computational science. What does it mean for a model to be truly predictive, or *ab initio*—"from the beginning"? An ideal ab initio model would use only the fundamental laws of physics (the Schrödinger equation) and a list of the atoms involved, and from that alone, predict the properties of a molecule. It uses no information from experiment.
In practice, this is extraordinarily difficult. Consider the workhorse of modern chemistry, Density Functional Theory (DFT). The theory is formally exact, but its practical application requires an approximation for a component called the exchange-correlation functional. Many popular functionals, like the famous B3LYP, are hybrids (``). They are a cocktail of ingredients: some parts are derived from pure theory (like a fraction of Hartree-Fock exchange, which helps correct for an electron incorrectly interacting with itself), while other parts and the mixing coefficients are empirically fitted to match experimental data for a set of reference molecules. This injected experimental information is why B3LYP works so well for many problems, but it also means it is not strictly ab initio.
This creates a fascinating debate. Some modern techniques involve "tuning" a parameter in a functional for each specific molecule being studied (``). If you tune the parameter to match an experimental measurement for that molecule, you've clearly made the model empirical. But what if you tune the parameter to enforce a known exact theoretical condition that the perfect functional ought to satisfy? For example, forcing the energy of the highest occupied molecular orbital to match the calculated ionization potential. In this case, no new experimental information is used. The information guiding the model comes from the internal consistency of the theory itself. One can strongly argue that such a procedure, while system-specific, remains in the true spirit of an ab initio method. It is a profound distinction, forcing us to ask: what is the ultimate source of our model's information?
The goal of a model is to simplify, but it is possible to simplify too much, to throw away the very information that matters most. This is the classic pitfall of essentialism—the idea that a complex, variable group can be understood by a single "ideal type" or "essence."
Imagine a theoretical model of a fish species that predicts a single, optimal age for reproduction, calculated by balancing the probability of survival against the increase in fecundity with age (``). The model might churn out an answer: the optimal age is 4.73 years. This is an essentialist prediction. It contains information about a hypothetical, average fish.
But then we go out to the real world and find something much richer. In a lake with few predators and low population density, the fish reproduce at a wide range of ages, centered around 4 years. In another lake with high density, the distribution of reproductive ages is just as broad, but the average is now 6 years. The model of a single "optimal" type has failed completely. It missed the most important parts of the story. The crucial information wasn't in the average; it was in the variation within each population and the context-dependence of the outcome on the environment (population density).
This is a powerful closing lesson. Population thinking, the cornerstone of modern biology, teaches us that variation is not noise to be ignored; it is the reality to be explained. It is the raw material for evolution. A model that discards this information in search of a single, Platonic ideal is not just wrong; it is profoundly misleading. The journey to understand physical information is a journey to appreciate nuance, context, and complexity. It teaches us to be precise about what we know, to be honest about what we don't, and to never forget that sometimes, the most important information lies not in the signal, but in what we might have first dismissed as noise.
In the previous chapter, we explored a rather lovely and powerful idea: that at its heart, physics isn't just about forces and particles, but about information. It's about building conceptual models of the world and then, like a detective, gathering clues—empirical data—to see how good our models are. The relationship between the model and the data is the physical information. This might sound a bit abstract, but the moment you start to look for it, you see this principle at work everywhere, unifying vast and seemingly disconnected fields of human inquiry. It's the engine of discovery, and in this chapter, we're going to take a tour to see it in action.
Let's start at the smallest scales. How do we know anything about the frantic, ceaseless dance of atoms in a crystal? We can't see them directly with a simple microscope. Instead, we do something clever: we scatter particles, like neutrons, off the crystal and look at the pattern they make. This diffraction pattern is, in essence, a coded message. The crystal's structure and the motion of its atoms encode information into the paths of the scattered neutrons. Our job is to decode it.
Modern experiments can do this with astonishing precision. By measuring the intensities of scattered neutrons over a wide range of angles, we can go beyond just finding the average position of an atom. We can reconstruct its "anisotropic displacement parameters"—a fancy term for a very beautiful piece of information. This tells us the shape and orientation of the little cloud of probability where the atom jiggles and vibrates. We learn that a hydrogen atom in a molecule doesn't just buzz randomly; it might move more freely along a hydrogen bond, or trace a tiny arc as the molecule it's attached to librates back and forth. Even more wonderfully, by performing these experiments at very low temperatures, we can see that the atoms never stand still. They retain a residual jiggle, a zero-point motion dictated by the uncertainty principle. This is purely quantum mechanical information, read directly from the physical world. We are, in a very real sense, observing the physical consequences of quantum information.
But what's truly remarkable is that we can often extract powerful information about a physical system even without a complete microscopic theory. Imagine you're bombarding a solid surface with ions, a process called sputtering that's crucial for making microchips. You want to know how many atoms get knocked off for each incoming ion. This depends on the ion's energy (), its mass (), the target atom's mass (), and how strongly the atoms are bound to the surface (). You could try to build a complex simulation of all the atomic collisions, but there's a more direct path. The laws of physics have a certain grammar, a consistency we call dimensional analysis. An energy must always be an energy; a mass, a mass. This simple fact provides an enormous constraint. It tells you that these variables can only be combined in very specific ways to produce a dimensionless number for the sputtering yield. By simply demanding that the units on both sides of our equation match up, and adding a couple of key physical insights about the process, we can deduce the form of the governing law. We find that the information was there all along, hidden not in the messy details of the collisions, but in the fundamental structure of physical quantities themselves.
This idea of extracting information from a system's structure scales up beautifully. Consider the World Wide Web—a sprawling, complex network of billions of pages and links. Is there any meaningful information hidden in this tangled mess? A physicist, looking at this, might see an analogy to a random process. Imagine a surfer clicking links at random. Some pages, by virtue of being linked to by many other important pages, will be visited more often. The "importance" of a page, then, can be defined as the long-term probability of finding our random surfer on it.
This is precisely the core idea behind Google's PageRank algorithm. The link structure of the web is our empirical data. We build a mathematical model of it, a "Google matrix," which describes the probabilities of hopping from any page to any other. The most important piece of information in this entire system—the relative rank of every single webpage—turns out to be the dominant eigenvector of this matrix. The same mathematical tools we use in quantum mechanics to find the ground state energy of an atom can be used to extract the informational hierarchy of the entire internet. It's a stunning example of how a physical way of thinking, modeling a system's structure and dynamics, can reveal profound information in a completely non-physical domain.
The connection between physics and information becomes most explicit, of course, in the field of quantum computing. Here, the information itself—the qubit—is a physical quantum system. The great challenge is that this physical information is incredibly fragile, constantly being battered by noise from the environment. The solution is a masterpiece of physical information theory: quantum error correction. We don't store our information in a single physical qubit. Instead, we encode a single logical unit of information across many physical qubits, using an intricate structure like the "surface code."
This encoding is a physical system designed for one purpose: to protect information. A random physical error, like a single faulty operation on one tiny part of the computer, doesn't immediately destroy the logical information. Instead, it creates a subtle signature within the code. The code is designed so that most local physical errors correspond to detectable signatures that don't flip the logical state. However, the system isn't perfect. A single physical fault, occurring in just the wrong place, can mimic the signature of a logical operator, causing the logical information to flip without us realizing it. Calculating the probability of such an event shows us the deep interplay between the physical layout of the computer, the nature of physical noise, and the integrity of the abstract information being processed. We are literally building physical universes whose laws are engineered to preserve knowledge.
Nowhere is the concept of physical information more manifest than in biology. A living organism is a whirlpool of information, processing it, storing it, and passing it on. And we find that this biological information is often shaped and constrained by the laws of physics and geometry.
Consider a simple question: why don't we have mice the size of elephants, or elephants the size of mice? For centuries, naturalists have observed that an animal's physiology changes with its size in predictable ways. This is the science of allometry. For example, for a vast range of mammals, the metabolic rate () doesn't scale in direct proportion to body mass (), which you might expect if an organism were just a big bag of identical cells. Instead, we find a relationship closer to . This exponent, , is a piece of physical information, a summary of a deep truth about biological design, extracted from measurements across hundreds of species.
Why this particular number? The explanation lies in physical constraints. An organism isn't a simple bag of cells; it's a complex machine that needs to transport oxygen and nutrients to every one of those cells. One beautiful theory proposes that the architecture of life is dominated by fractal-like, space-filling transport networks—our circulatory systems, our respiratory tracts. Optimizing the physics of fluid flow through such a network to minimize energy dissipation predicts a scaling exponent of . Other models focus on how the relative composition of different organs changes with size, or how the metabolic activity of individual cells might decrease in larger animals. The observed exponent of is likely a result of several of these physical and compositional effects combined. The key insight is that a simple number, an empirical piece of information, acts as a powerful clue, pointing us toward the fundamental physical principles that govern the form and function of all life.
This "information-first" perspective can even illuminate the grand process of evolution. We can model the change in a trait—say, the average beak size in a population of birds—over thousands of generations. The mean trait value is a piece of information. Each generation, this information is updated by two competing processes. First, natural selection pushes the trait in a certain direction; this is the environment providing directional information (the "selection gradient"). Second, in any finite population, random chance—genetic drift—causes the trait to jitter unpredictably. By combining the breeder's equation from quantitative genetics with a model for stochastic drift, we can create a physical model for the evolution of information. We can then ask precise questions, such as "After a thousand generations, what is the expected beak size, and what is the variance, or uncertainty, in that expectation?" This framework treats evolution itself as a stochastic process acting on information, a beautiful synthesis of biology and statistical physics.
This perspective becomes a powerful tool in the urgent field of conservation genetics. When a species is critically endangered, its genetic diversity plummets. This can harm the population in two ways: deleterious mutations can become "fixed" (present in every individual), creating a "fixed load," while other harmful recessive alleles can be exposed through inbreeding, creating a "segregation load." How can we tell which is the bigger problem? We can turn to the organism's own physical information. By sequencing the genomes of many individuals and correlating their fitness with their genome-wide heterozygosity (a measure of genetic diversity), we can build a simple linear model. This model, grounded in population genetics theory, allows us to take the slope and intercept from a statistical regression and partition the total fitness decline into the parts attributable to fixed load versus segregation load. It's a remarkable feat: we are reading the population's genetic code to diagnose the source of its ailment, providing crucial information for its survival.
This modern view of science as an information-gathering enterprise is not, in fact, so modern. It has been the implicit engine of science for centuries. In the 18th century, Carolus Linnaeus set out to create a system to classify all of life, his Systema Naturae. This was, in essence, an information architecture, a vast filing system for the living world. But in its early days, it was based almost entirely on European specimens. To make it a truly global system, Linnaeus dispatched his students—his "apostles"—on perilous journeys across the globe. Their mission was to collect physical information in the form of specimens of plants and animals. This flood of new data from around the world was what transformed his system from a regional curiosity into the foundation of modern taxonomy. It was a massive, organized effort to gather physical information to populate, test, and validate a theoretical model of the world.
Even earlier, in the 17th century, Antony van Leeuwenhoek peered through his handmade microscopes at drops of water and saw a world of "animalcules" no one knew existed. The dominant philosophy of his time was essentialism—the idea that every species was defined by a single, perfect, unchanging "type." Any variation seen in real individuals was just an imperfect deviation from this ideal form. In this view, variation is noise. But Leeuwenhoek was a meticulous observer. When he described a "sort" of animalcule, his drawings didn't show one ideal form; they showed a range of forms—individuals differing in size, shape, and structure. By carefully recording this variation instead of dismissing it as imperfection, Leeuwenhoek, without realizing it, was making a revolutionary philosophical shift. He was treating variation as signal, as part of the essential information content of a species. This act of recording the full spectrum of physical information was a necessary precursor to Darwin's theory of evolution, which recognizes variation not as noise, but as the indispensable raw material for natural selection.
And so we come full circle, back to the relationship between measurement and model. In modern computational science, this interplay is more vibrant than ever. Suppose an experiment gives us a set of measurements, which we summarize in a histogram—a set of counts in different bins. This histogram is our physical information about the process. We can then build a simple model that approximates the underlying probability distribution, for instance by assuming it's uniform within each bin. The final, magical step is to use this model to run a simulation. Using a technique called inverse transform sampling, we can turn a stream of random numbers into a stream of simulated events that, statistically, look just like the real process. We have used the information extracted from the physical world to create a virtual world that obeys the same rules, a world we can explore to our heart's content.
From the quantum jiggle of a single atom to the evolutionary trajectory of an entire species, from the structure of the internet to the very process of scientific discovery itself, the unifying thread is this quest for physical information. It is the process of observing the world, building models to contain our understanding, and using the tension between the two to reveal the deep and often hidden connections that weave the fabric of reality. It is a journey of endless fascination, and there is always another layer to explore.