Claude Shannon worked out what information is. Not what it means, which he deliberately set aside, but how much of it there is, how tightly it can be squeezed, and how fast it can be pushed through a noisy wire without errors. He did this largely in one paper in 1948, and every digital system built since has run on the answer.
The reason a quantum publication should care is more specific than general reverence. Quantum information theory is not a new subject that happens to share a word with Shannon’s. It is his framework rebuilt piece by piece for a world where states cannot be copied and reading destroys what you read, and the rebuild took fifty years because several pieces came out genuinely different.
His master’s thesis made digital circuits possible. Written in 1937, it showed that Boolean algebra describes relay switching exactly, which is the idea every logic circuit since has been built on.
The 1948 paper created information theory in one go. It defined entropy as the true size of a message, defined channel capacity as the true speed of a channel, and proved what each one limits.
Noise limits speed, not accuracy. Shannon proved you can transmit below capacity with an error rate as close to zero as you like, which contradicted what engineers assumed and made error correction a subject.
Quantum information theory is his framework rebuilt. Entropy became von Neumann entropy, the bit became the qubit, and source coding became Schumacher compression, each needing a new proof.
Some pieces came out genuinely different. Classical channel capacity is a single number that adds up. Quantum capacity is several numbers, and in 2008 one of them was shown not to add up at all.
He proved the one-time pad is unbreakable. That 1949 result is the standard quantum key distribution aims at, which is why QKD is judged against a theorem Shannon wrote.
- Who Claude Shannon was
- The thesis that built the digital age
- What the 1948 paper actually says
- Entropy, and the naming story
- Redundancy and compression
- The result nobody expected
- Naming the bit
- Secrecy, and the one unbreakable cipher
- Why quantum had to start again
- Von Neumann entropy came first
- Schumacher and the birth of the qubit
- The Holevo limit
- Where the analogy breaks
- Error correction, classical and quantum
- From the one-time pad to QKD
- What Shannon limits rule out
- Shannon and Turing
- The juggling, the mouse and the useless machine
- Information as a physical quantity
- Reading the 1948 paper today
- What he actually left behind
- Frequently asked questions
Who Claude Shannon was
Claude Shannon, born Claude Elwood Shannon on 30 April 1916 in Petoskey, Michigan, and died on 24 February 2001. He studied at the University of Michigan and then at MIT, where he produced a master’s thesis in 1937 and a doctorate in 1940, the latter on an algebra for theoretical genetics that has little to do with what he is remembered for.
He spent his most productive years at Bell Telephone Laboratories, then a research institution of extraordinary density, before returning to MIT as a professor. He was by most accounts uninterested in fame, and spent a good deal of his later life on unicycles, juggling and mechanical toys.
What he did in between is the reason the phrase information age is not a metaphor. Shannon took a word everyone used loosely and gave it a definition precise enough to calculate with.
The thesis that built the digital age
In 1937, as a master’s student, Claude Shannon wrote A Symbolic Analysis of Relay and Switching Circuits. The observation is simple enough to state in a sentence and was not obvious to anyone before him. The behaviour of electrical relays, which are either open or closed, is described exactly by Boolean algebra, the nineteenth-century logic of true and false.
The consequence is that circuit design becomes algebra. Rather than building a switching network and testing it, you write down an expression, simplify it by the rules of logic, and build the simplified version, knowing it behaves identically. Every logic gate, every processor and every quantum control system that switches on classical logic descends from that observation.
It is routinely described as the most influential master’s thesis ever written, and the claim is hard to argue with. Shannon was 21.
What the 1948 paper actually says
A Mathematical Theory of Communication by Claude Shannon appeared in the Bell System Technical Journal in 1948. It opens by discarding the thing most people assume communication is about, stating that the semantic aspects of a message are irrelevant to the engineering problem. Whether a message means anything does not affect what it costs to send.
What remains once meaning is stripped out is a system with five parts. A source produces a message, an encoder turns it into a signal, a channel carries the signal and corrupts it with noise, a decoder reconstructs the message, and a destination receives it. That decomposition is now so standard it feels like a description of reality rather than a modelling choice.
From it Shannon derived two theorems that bound what any communication system can do. The first sets a floor on compression and the second sets a ceiling on reliable speed, and between them they define the whole subject.
Entropy, and the naming story
Claude Shannon needed a measure of how much information a source produces, and the quantity he arrived at is now written H and called entropy. It depends only on the probabilities of the possible messages, and it is largest when every message is equally likely, which is the situation where you learn the most by finding out which one arrived.
A fair coin flip carries one bit, because there are two equally likely outcomes. A coin that lands heads 99 times in a hundred carries far less, because you can usually guess it. This is why text compresses and random noise does not, and Shannon’s source coding theorem makes the point exact. No lossless compression scheme can get below the entropy of the source on average.
A famous anecdote has John von Neumann advising Shannon to use the word entropy because it already existed in statistical mechanics and because nobody really understood it, giving him the advantage in any argument. The story is widely repeated and probably apocryphal, since it traces to a retelling published more than twenty years after the supposed conversation. The connection to physical entropy, however, is real rather than a pun.
Redundancy, and why English compresses
One idea from the 1948 paper reaches ordinary life more directly than the theorems. Shannon measured the redundancy of English, meaning the degree to which its letters are predictable from what came before, and estimated that natural text carries far less information per character than its alphabet size suggests.
He tested this with a memorable experiment, asking people to guess a text one letter at a time and recording how often they were right. Their success rate is a measurement of the language’s redundancy, and it explains both why text files compress well and why you can read a sentence with most of its vowels removed.
Redundancy is also the resource error correction spends. A code protects a message by adding structured redundancy, so that a corrupted version still points unambiguously at the original. Shannon’s insight was that natural redundancy is haphazard while engineered redundancy can be optimal, and the entire field of coding theory is the work of engineering it well.
The result nobody expected
The second theorem is the one that changed engineering practice. Before 1948 the working assumption was that noise put a floor under the error rate, so that a noisier line meant more mistakes and the only remedies were more power or slower transmission.
Claude Shannon proved otherwise. Every channel has a capacity, a number he called C, and for any transmission rate below C there exist codes that make the probability of error as small as you like. Noise limits how fast you can send information, not how accurately. Above capacity reliable communication is impossible, and below it, arbitrarily reliable communication is available to anyone who can find the right code.
The proof is famously non-constructive. Shannon showed that good codes exist by demonstrating that a randomly chosen code is usually good, without providing any method to build one. The following sixty years of coding theory were substantially an effort to construct what he had proved must be there, and the same pattern recurs in quantum error correction.
Naming the bit
The 1948 paper introduced the bit as the standard unit of information, crediting the contraction of binary digit to his Bell Labs colleague John Tukey. It is a small piece of terminology that did a great deal of work, because giving information a unit is what makes it measurable and therefore tradeable.
Once information has a unit, questions that were previously rhetorical become arithmetic. How much does this file weigh, how much can this line carry, how much does this cipher leak. The qubit is the direct descendant of that decision, and it was named for the same reason nearly fifty years later.
Secrecy, and the one unbreakable cipher
In 1949 Shannon published Communication Theory of Secrecy Systems, applying the same machinery to cryptography. The paper had circulated as classified wartime work, and it did for ciphers what the 1948 paper did for channels, which was to replace intuition with proof.
Its most quoted result concerns the one-time pad, a cipher using a random key as long as the message and never reused. Shannon proved it offers perfect secrecy, meaning the ciphertext reveals literally nothing about the plaintext beyond its length, and that this property comes at the price of a key as large as everything you will ever send.
That result matters here because it is the standard quantum key distribution is aiming at. QKD does not invent unbreakable encryption. It provides a physical method for distributing the enormous keys that Shannon proved are necessary, which is a solution to the delivery problem rather than to the cipher problem.
Why quantum had to start again
The theory Claude Shannon built assumes things about information that are false for quantum systems, and the assumptions are so basic that they are almost invisible. He assumes a message can be read without changing it, that it can be copied freely, and that a signal has a definite value whether or not anyone looks.
None of that survives quantum mechanics. An unknown quantum state cannot be copied, a consequence known as the no-cloning theorem. Measuring a state generally disturbs it. And a system in superposition has no definite value to read in the first place, which means the very first step of Shannon’s model, observing what the source produced, is not available.
So the whole apparatus had to be rebuilt. The remarkable part is how much of it survived in recognisable form, and the instructive part is which pieces did not.
Von Neumann entropy came first
The quantum measure of information is called von Neumann entropy, written S, and there is a chronological oddity worth noting. Von Neumann defined it in 1927, more than two decades before Shannon’s paper, as part of quantum statistical mechanics rather than as a theory of communication.
It generalises Shannon’s quantity in a precise sense. For a system in a definite classical mixture the two agree, and for a quantum state in superposition von Neumann entropy captures something Shannon’s cannot, which is that a pure entangled state can have zero total entropy while each of its parts has a great deal. That fact has no classical counterpart at all, and it is the mathematical signature of entanglement.
Schumacher and the birth of the qubit
The quantum analogue of Shannon’s source coding theorem arrived in 1995, when Benjamin Schumacher proved what quantum compression can achieve. The result says that a quantum source can be compressed down to its von Neumann entropy and no further, exactly parallel to Shannon’s statement about classical entropy.
In the same work Schumacher introduced the word qubit for the unit being counted. The naming follows Shannon’s logic precisely, since a compression theorem needs a unit to express its answer in, and the answer to how many two-level quantum systems you need turned out to be the natural unit of quantum information.
The correspondence is close enough to be startling. Two theorems separated by forty-seven years, one about classical messages and one about quantum states, have the same shape, the same role and an entropy on the right-hand side.
The Holevo limit
Then the differences begin. A natural hope is that quantum states, holding continuous amplitudes, might carry far more classical information than classical systems of the same size. Alexander Holevo proved in 1973 that they do not.
The Holevo bound limits how much classical information can be extracted from a quantum system by measurement. Whatever you encode into n qubits, no measurement strategy recovers more than n classical bits. The amplitudes are real and they are not readable, which is the same wall that quantum algorithms run into when they cannot simply read the superposition they have built.
This is one of the most useful results to know when assessing claims about quantum storage or communication capacity. Quantum systems are not roomier than classical ones for the purpose of holding classical data.
Where the analogy breaks
Channel capacity is where the rebuild stopped resembling the original. A classical channel has one capacity, a single number that answers every question about it, and if you have two channels their capacities add.
A quantum channel has several. There is a capacity for sending classical information, another for sending quantum information, another when the sender and receiver share entanglement in advance, and they are different numbers with different formulas. Worse, in 2008 the quantum capacity was shown to be non-additive, meaning two channels used together can carry more than the sum of what each carries alone, and there are channels with zero capacity individually that have positive capacity when combined.
Nothing like that happens classically, and it is not a technicality. It means there is no single number summarising a quantum channel, and that the neat accounting Shannon established does not carry over. Fifty years of work turned up a genuine structural difference rather than a translation.
Error correction, classical and quantum
The Claude Shannon noisy channel theorem is the direct ancestor of quantum error correction, and the intellectual debt is explicit. He established that reliable transmission through an unreliable medium is possible in principle, which turned error correction from wishful thinking into engineering.
The quantum version faced an obstacle that looked fatal. Classical error correction works by copying, and no-cloning forbids that. Measurement is how you detect an error, and measurement destroys the state you are protecting. For some years it was reasonable to believe quantum error correction was impossible.
The resolution, found in the mid-1990s, was to encode information across several physical systems so that errors can be detected without measuring the encoded data itself, by measuring only relationships between the parts. That is what a stabiliser code does, and it is why the surface code and the codes emerging around it work. The threshold theorem that followed is a direct quantum descendant of Shannon’s capacity result, saying that below a certain physical error rate, arbitrarily reliable quantum computation is achievable.
From the one-time pad to QKD
The 1949 secrecy paper closes the loop. Shannon proved the one-time pad is perfectly secure and that its security requires a key as long as the message, shared in advance and never reused. The difficulty was never the cipher but the logistics of distributing that much key material.
Quantum key distribution attacks exactly that problem. Protocols such as BB84 use the fact that measurement disturbs a quantum state to detect eavesdropping while a key is being established, giving two parties a shared random string with a physical guarantee about who else could have seen it.
Read in Shannon’s terms, QKD is a key delivery mechanism for a cipher he analysed in 1949, and this framing is a useful corrective to the marketing. It also explains why post-quantum cryptography is a separate and largely more practical response to the same threat, since it changes the cipher rather than the delivery.
What Shannon limits rule out
The practical use of information theory is often negative, which is to say it tells you what cannot be done and saves the effort of trying. Several recurring claims in technology marketing are ruled out by results in these two papers, and recognising them is a quick filter.
A compression scheme that shrinks any input is impossible, since if every file got smaller two files would eventually compress to the same output and could not be distinguished. A channel cannot carry more than its capacity reliably, regardless of clever encoding. And a cipher with a key shorter than its message cannot offer perfect secrecy, which is why every practical cryptosystem trades provable security for computational difficulty.
The quantum versions of these limits are equally useful. The Holevo bound rules out packing more classical data into qubits than into bits. No-cloning rules out amplifying a quantum signal the way a repeater amplifies a classical one, which is precisely why entanglement-based quantum repeaters are hard and why long-distance QKD remains an engineering problem rather than a solved one.
A claim that violates one of these is not ambitious, it is wrong. Knowing which limit applies is usually faster than evaluating the claim on its own terms, and it is the reason information theory remains a practical tool rather than a historical one.
Shannon and Turing
During the war Claude Shannon worked on cryptography and on fire-control systems at Bell Labs, and in 1943 Alan Turing spent time there on a related mission concerning secure speech. The two met over tea, and by both later accounts they talked about the possibility of machines that think, without being able to discuss the classified work each was actually doing.
The pairing is worth noting because their contributions are complementary in a way that is easy to miss. Turing established what can be computed at all, defining the limits of mechanical procedure. Shannon established what can be communicated and stored, defining the limits of representation and transmission. Between them they bounded the two axes on which every computing system is measured.
Quantum computing has needed both bounds redrawn. The quantum extension of Turing’s work asks what a quantum machine can compute efficiently, which is the complexity class BQP and the subject of quantum algorithms. The quantum extension of Shannon’s asks what can be stored and sent, which is quantum information theory. The two programmes have run in parallel since the 1990s and they meet in quantum error correction, which is a communication result that computation depends on.
The juggling, the mouse and the useless machine
The other output of Claude Shannon resists summary and is worth recording. He built Theseus, a mechanical mouse that learned to solve a maze and remembered the solution, which is among the earliest demonstrations of machine learning in any recognisable sense.
He built what he called the Ultimate Machine, a box with a single switch, which when flipped opened a lid so a mechanical hand could emerge and switch itself off. He rode a unicycle through the corridors at Bell Labs, sometimes while juggling, and wrote a scientific paper on the mathematics of juggling. He also built machines for playing chess and, with Edward Thorp, a wearable device intended to beat roulette.
The temptation is to file this as eccentricity. A better reading is that the same instinct produced both the toys and the theorems, which is a habit of taking a system apart to find the rule that governs it, and not much caring whether the system was important.
Information as a physical quantity
There is a deeper thread running from Shannon into modern physics that deserves a mention, because it is the reason the entropy connection is not a coincidence. Rolf Landauer argued in 1961 that erasing information has an unavoidable thermodynamic cost, dissipating a minimum amount of heat per bit destroyed.
If that is right, and experiments since have supported it, then information is not an abstraction sitting above physics but a physical quantity subject to physical law. Charles Bennett later used this to resolve the puzzle of Maxwell’s demon, showing that the demon’s bookkeeping must eventually be erased and that the erasure pays back exactly the entropy the demon appeared to have saved.
This matters for quantum computing in a concrete way. Reversible computation, which avoids erasing information and therefore avoids the Landauer cost, is not an efficiency curiosity in the quantum setting but a requirement, since quantum evolution is reversible by construction. The line from Shannon’s entropy through Landauer’s principle to the reversible circuits that quantum algorithms are built from is continuous.
Reading the 1948 paper today
The Claude Shannon paper is unusually approachable for a founding document, and anyone working in quantum information will find the effort repaid. It is written in plain language with the mathematics separated out, and the opening pages explain the model with no formalism at all.
Two things strike a modern reader. The first is how completely the framework arrived, since Shannon states the model, defines the measures, proves both theorems and discusses applications in a single work, where most fields assemble their foundations over decades. The second is the tone, which is careful about what has been proved and what has merely been made plausible, and notably unexcited about the implications.
That restraint is instructive given what followed. Information theory became fashionable in the 1950s and was applied enthusiastically to biology, psychology, linguistics and economics, often by people who had taken the vocabulary without the mathematics. Shannon wrote a short piece asking for the bandwagon to slow down, on the grounds that the theory was a technical result about communication rather than a general theory of everything. The parallel with the current enthusiasm for applying quantum vocabulary to unrelated fields hardly needs drawing.
What he actually left behind
The legacy of Claude Shannon is usually stated as the digital age, which is true and vague. The sharper version is that he established that information is a physical quantity with laws, and that those laws set limits no amount of engineering can cross.
That is why his framework survived the transition to quantum mechanics rather than being replaced by it. The specific theorems needed rebuilding because the physical substrate changed, but the questions turned out to be the right ones. How much information is here, how far can it be compressed, how fast can it be sent reliably, how much can an eavesdropper learn. Quantum mechanics gives different answers to all four, and it did not produce a fifth question.
For anyone assessing quantum computing and communication claims, that is the practical inheritance. When a proposal promises unlimited capacity, unbreakable secrecy without key distribution, or storage beyond what the qubit count allows, the relevant objection is usually a Shannon-style limit, or the quantum result that replaced it.
Frequently asked questions
Who was Claude Shannon?
Claude Shannon was an American mathematician and engineer, born in 1916 and died in 2001, who founded information theory. His 1948 paper A Mathematical Theory of Communication defined how information is measured and what limits apply to compressing and transmitting it, and his 1937 master’s thesis established the logical basis of digital circuit design.
What is Shannon entropy?
It is a measure of how much information a source produces, depending only on the probabilities of its possible messages. It is highest when all outcomes are equally likely, and it sets the limit below which lossless compression is impossible.
What is Shannon’s channel capacity theorem?
It states that every communication channel has a maximum rate, called its capacity, below which information can be sent with an error probability as close to zero as desired. The surprising part is that noise limits speed rather than accuracy, which is the basis of all error correction.
Did Claude Shannon invent the bit?
He introduced the bit as the standard unit of information in his 1948 paper, crediting his Bell Labs colleague John Tukey with coining the contraction of binary digit. The formal role of the bit as the unit information is measured in is Shannon’s.
How does Shannon’s work relate to quantum computing?
Quantum information theory is his framework rebuilt for quantum systems. Shannon entropy becomes von Neumann entropy, the bit becomes the qubit, his source coding theorem becomes Schumacher compression, and his noisy channel theorem is the ancestor of quantum error correction and the threshold theorem.
What is the difference between Shannon entropy and von Neumann entropy?
Von Neumann entropy is the quantum generalisation, and it agrees with Shannon entropy for classical probability mixtures. It differs for quantum states, most strikingly in that an entangled pure state can have zero total entropy while its individual parts each have a large amount, which has no classical counterpart.
Did Shannon prove the one-time pad is unbreakable?
Yes. In his 1949 paper on secrecy systems he proved that a one-time pad, using a truly random key as long as the message and never reused, offers perfect secrecy, meaning the ciphertext reveals nothing about the plaintext except its length.
What was Shannon’s master’s thesis about?
It showed that Boolean algebra, the logic of true and false, describes the behaviour of electrical relay circuits exactly. That turned circuit design into an algebraic exercise and provided the theoretical basis for all digital logic, and it is frequently called the most influential master’s thesis ever written.
What is the Holevo bound?
Proved by Alexander Holevo in 1973, it limits how much classical information can be recovered from a quantum system by measurement. However you encode data into n qubits, no measurement strategy extracts more than n classical bits, so quantum systems are not roomier than classical ones for storing classical data.
Why is quantum channel capacity harder than classical?
A classical channel is described by one capacity number, and capacities of separate channels add. A quantum channel has several distinct capacities depending on what is being sent and what resources are shared, and in 2008 the quantum capacity was shown to be non-additive, so two channels together can carry more than the sum of their individual capacities.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
