Superdense Coding

Superdense coding is the cleanest proof that entanglement is not mystical decoration but a countable resource for communication. Introduced by Charles Bennett and Stephen Wiesner in 1992, the protocol does something that sounds impossible: it transmits two classical bits from a sender Alice to a receiver Bob while only one physical qubit ever travels down the quantum channel. A lone qubit cannot do this — the Holevo bound caps the accessible information of a single qubit at one bit — so the extra bit has to be bought somewhere. It is bought in advance, with a shared Bell pair. The setup is an accounting trick made physical. Before Alice knows what she wants to say, the two parties are each handed one half of a maximally entangled pair, $|\Phi^+\rangle = (|00\rangle + |11\rangle)/\sqrt2$ — one ebit. When Alice is ready, she applies one of four local Pauli operations, $I$ , $X$ , $Z$ , or $ZX$ , to her qubit alone. Those four operations rotate the shared pair into the four mutually orthogonal Bell states. She then sends her single qubit to Bob, who — now holding both halves — performs a Bell-state measurement and reads off the two-bit message with certainty. Nothing about the message ever lives in the qubit she sends: its reduced state is maximally mixed, the same $\tfrac12 I$ for all four messages. The information rides the correlation, which only becomes readable once the two halves are reunited. The bookkeeping is exact and elegant: one qubit plus one ebit is worth two classical bits, the resource inequality $1\text{ qubit} + 1\text{ ebit} \geq 2\text{ cbits}$ . Read backwards in time, the same picture is quantum teleportation — the two are mirror images, together the founding primitives of entanglement-assisted communication.

Key takeaways

Superdense coding sends two classical bits by transmitting a single qubit, provided Alice and Bob pre-share one half each of a Bell pair (one ebit).
Alice encodes her two-bit message $ab$ with one of four local Pauli operations $\{I,X,Z,ZX\}$ , mapping the shared $|\Phi^+\rangle$ onto one of the four orthogonal Bell states.
Bob reunites Alice's transmitted qubit with his half and performs a Bell measurement — a CNOT followed by a Hadamard and readout — recovering $ab$ deterministically.
The transmitted qubit alone leaks nothing: its marginal is the maximally mixed $\tfrac12 I$ for every message, so the Holevo bound is respected and an eavesdropper on the channel learns zero.
With a partially entangled pair the capacity drops smoothly to $1 + S(\rho_A)$ bits; with linear-optics Bell measurement it is capped at $\log_2 3 \approx 1.58$ bits.
Key idea: the four Pauli operators applied locally to one half of a Bell pair generate the four orthogonal Bell states; one qubit transmission plus a Bell measurement then conveys two perfectly distinguishable classical bits.

The core insight

The four Pauli operators applied locally to one half of a Bell pair generate the four orthogonal Bell states; a single qubit transmission plus a Bell measurement then conveys two perfectly distinguishable classical messages.

Where this lives today

Superdense coding is no longer just a textbook parable; it is the prototype for a whole branch of quantum Shannon theory and a recurring engineering target. Its direct descendant is the entanglement-assisted classical capacity $C_E$ of a noisy channel, the Bennett–Shor–Smolin–Thapliyal theorem, which is the one capacity in quantum information with a clean single-letter formula — superdense coding is the noiseless corner of that landscape. In photonic experiments the protocol runs headfirst into a famous obstacle: a linear-optics Bell-state analyzer can unambiguously resolve only two of the four Bell states, so the channel is capped at $\log_2 3 \approx 1.58$ bits rather than the full two. Getting past that ceiling has driven real technology — hyperentanglement (extra qubits encoded in a photon's momentum or orbital angular momentum, pioneered by Kwiat's group) and atom–photon interfaces both restore a complete Bell measurement, and trapped-ion experiments with deterministic entangling gates reached the full channel capacity as early as 2004. The same primitive reappears in continuous-variable optics and in quantum-network architectures that trade pre-shared ebits between teleportation and dense-coding uses depending on whether the payload is quantum or classical.

The physical picture

A single qubit is a stingy messenger. Its state is one point on the Bloch sphere, and any measurement pulls out at most one bit — the Holevo bound for a two-dimensional system. So when Alice conveys two bits by sending one qubit, the second bit cannot be sitting inside that qubit. It is sitting in the correlation between her qubit and Bob's, a correlation established before Alice touched anything. Think of the shared Bell pair as a codebook physically split between the two parties: Alice's four local operations turn to one of four pages, but no page number is written on her single qubit. Bob, holding both halves, simply reads which page the reunited pair is on. The first figure walks the whole protocol from the shared ebit to the two recovered bits, and carries the resource count that names the entire subject.

Figure

Superdense coding end to end : a Bell source pre-shares the ebit Φ⁺ with Alice and Bob, Alice applies one of the four Pauli operators I, X, Z, ZX to encode her two bits ab, sends her single qubit down the channel, and Bob's Bell measurement reads ab with certainty — the resource count is 1 qubit + 1 ebit ≥ 2 cbits

The magic is entirely in the four-way branching of the encoding. Because $|\Phi^+\rangle$ is maximally entangled, acting on Alice's qubit with a Pauli is enough to walk the global state around a set of four orthogonal — hence perfectly distinguishable — states. The deep symmetry to keep in mind is that superdense coding is the exact time-reverse of teleportation: teleportation spends one ebit and two classical bits to move one qubit; superdense coding spends one ebit and one qubit to move two classical bits. Reverse the arrow of time and swap the quantum and classical channels, and one becomes the other. The second figure is the dictionary at the heart of the encoding — which Pauli produces which Bell state, and why the sent qubit stays mixed throughout.

Figure

The encoding dictionary : each two-bit message ab picks a Pauli that rotates the shared Φ⁺ into one of the four orthogonal Bell states, 00→I→Φ⁺, 01→X→Ψ⁺, 10→Z→Φ⁻, 11→ZX→Ψ⁻ — yet every one has the same maximally mixed marginal ρ_A = ½I on the transmitted qubit

The core equations

The starting resource is one shared Bell pair, one ebit, distributed before any message exists. It is the maximally entangled state of two qubits, and everything downstream is a consequence of its structure. Alice holds qubit $A$ , Bob holds qubit $B$ .

|\Phi^+\rangle_{AB} = \frac{1}{\sqrt{2}}\big(|00\rangle + |11\rangle\big) \qquad (1)

where: $|\Phi^+\rangle$ — the standard Bell state shared before the protocol; $A,B$ — Alice's and Bob's qubits respectively. Reading: the pair is perfectly correlated in the computational basis and maximally entangled, so tracing out either qubit leaves the other maximally mixed. The four Bell states are the target of the encoding. They are an orthonormal basis of the two-qubit Hilbert space, which is exactly why a measurement can name them without error. Alice never sees more than one qubit of this basis, but her operation moves the joint state between its four members.

\begin{aligned}|\Phi^+\rangle &= \tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)\\ |\Phi^-\rangle &= \tfrac{1}{\sqrt2}(|00\rangle-|11\rangle)\\ |\Psi^+\rangle &= \tfrac{1}{\sqrt2}(|01\rangle+|10\rangle)\\ |\Psi^-\rangle &= \tfrac{1}{\sqrt2}(|01\rangle-|10\rangle)\end{aligned} \qquad (2)

where: the four states are mutually orthogonal, $\langle\Phi^\pm|\Phi^\mp\rangle = \langle\Phi^\pm|\Psi^\pm\rangle = 0$ , forming a basis of $\mathbb{C}^2\otimes\mathbb{C}^2$ . Reading: because they span the space and are orthogonal, a single projective measurement distinguishes all four with certainty. The encoding map is the protocol's engine. Alice applies one of four Pauli operators to her qubit only, and each choice sends the shared $|\Phi^+\rangle$ to a different Bell state. This is the step where two classical bits get written into the joint state.

(\,U_{ab}\otimes I\,)\,|\Phi^+\rangle = |\beta_{ab}\rangle,\qquad U_{00}=I,\;\;U_{01}=X,\;\;U_{10}=Z,\;\;U_{11}=ZX \qquad (3)

where: $ab$ — the two classical bits Alice wishes to send; $U_{ab}$ — the Pauli operation applied to qubit $A$ ; $|\beta_{ab}\rangle$ — the resulting Bell state, from $\{|\Phi^+\rangle,|\Psi^+\rangle,|\Phi^-\rangle,|\Psi^-\rangle\}$ . Reading: the bit $a$ controls the phase flip $Z$ and the bit $b$ controls the bit flip $X$ , so $U_{ab}=Z^a X^b$ up to a global phase. The transmitted qubit carries no local information at any point. Tracing Bob's half out of any Bell state yields the same maximally mixed marginal, identical for all four messages, which is why interception of the channel qubit is worthless.

\rho_A = \operatorname{Tr}_B\big(|\beta_{ab}\rangle\langle\beta_{ab}|\big) = \tfrac{1}{2}\,I \qquad \forall\, ab \qquad (4)

where: $\operatorname{Tr}_B$ — partial trace over Bob's qubit; $\tfrac12 I$ — the maximally mixed single-qubit state. Reading: the four encoded states have identical marginals, so no measurement on the channel qubit alone can distinguish the messages — the information is nonlocal. The whole feat is summarized by a resource inequality, the grammar of quantum Shannon theory. It says the resources on the left can accomplish anything the resources on the right can, and the equality is achieved by the protocol above.

1\;\text{qubit} + 1\;\text{ebit} \;\geq\; 2\;\text{cbits} \qquad (5)

where: ebit — one shared maximally entangled Bell pair; cbit — one noiseless classical bit; $\geq$ — "can simulate / is at least as powerful as". Reading: run the same relation backwards with the channels swapped and you get teleportation, $1\text{ ebit} + 2\text{ cbits} \geq 1\text{ qubit}$ . Finally, the Holevo bound is the ceiling the protocol must respect, and the reason the second bit must be bought with entanglement rather than smuggled in the qubit.

\chi \;=\; S\!\Big(\textstyle\sum_x p_x \rho_x\Big) - \sum_x p_x\, S(\rho_x) \;\leq\; \log_2 d \qquad (6)

where: $\chi$ — Holevo quantity, an upper bound on accessible information; $S$ — von Neumann entropy; $d$ — Hilbert-space dimension. Reading: for an unassisted single qubit $d=2$ so $\chi\leq 1$ ; the pre-shared ebit lifts the effective signal space to the four-dimensional two-qubit space where $\chi$ can reach $\log_2 4 = 2$ .

Derivation AThe four Pauli operators generate the four Bell states

The encoding step is the entire trick of the protocol, so it earns a direct verification rather than an appeal to authority. We act with each of the four Pauli operators on Alice's half of the shared $|\Phi^+\rangle$ , read off the resulting two-qubit state, and confirm that the four outputs are exactly the four Bell states — and that they are mutually orthogonal. Orthogonality is not a nicety here: it is precisely what lets Bob name the state with certainty, so establishing it is the same as establishing that the channel really delivers two clean bits and not some blurred fraction of them.

Step 1. Message $00$ , operator $I$ : $(I\otimes I)|\Phi^+\rangle = |\Phi^+\rangle = \tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ . The state is untouched. Why: the "do nothing" branch has to be one of the four outcomes, and it anchors the dictionary at $|\Phi^+\rangle$ .

Step 2. Message $01$ , operator $X$ : $X$ swaps $|0\rangle\leftrightarrow|1\rangle$ on Alice's qubit, so $|00\rangle\to|10\rangle$ and $|11\rangle\to|01\rangle$ , giving $(X\otimes I)|\Phi^+\rangle = \tfrac{1}{\sqrt2}(|10\rangle+|01\rangle) = |\Psi^+\rangle$ . Why: a bit flip moves the pair off the "same" subspace onto the "different" subspace, converting a $\Phi$ into a $\Psi$ .

Step 3. Message $10$ , operator $Z$ : $Z$ leaves $|0\rangle$ alone and sends $|1\rangle\to-|1\rangle$ , so $|11\rangle\to-|11\rangle$ , giving $(Z\otimes I)|\Phi^+\rangle = \tfrac{1}{\sqrt2}(|00\rangle-|11\rangle) = |\Phi^-\rangle$ . Why: a phase flip flips the relative sign, the only other degree of freedom in the Bell basis besides the bit pattern.

Step 4. Message $11$ , operator $ZX$ : apply $X$ then $Z$ . After $X$ the state is $\tfrac{1}{\sqrt2}(|10\rangle+|01\rangle)$ ; then $Z$ on the first qubit flips the sign of $|10\rangle$ , giving $\tfrac{1}{\sqrt2}(|01\rangle-|10\rangle) = |\Psi^-\rangle$ . Note $ZX = iY$ up to phase, so this branch is the $Y$ -Pauli. Why: combining the flip and the phase reaches the last, antisymmetric Bell state, completing the basis.

Step 5. Orthonormality: each output is manifestly normalized, and any two differ either in their bit pattern (the $\Phi$ pair versus the $\Psi$ pair are supported on orthogonal computational states) or in a relative sign (which flips one inner-product term), so all six cross inner products vanish: $\langle\beta_{a'b'}|\beta_{ab}\rangle = \delta_{aa'}\delta_{bb'}$ . Why: distinctness is not enough — mutual orthogonality is what makes a single measurement resolve all four without error.

Go deeper · Derivation B · 1 minDecoding by a CNOT and a Hadamard

Derivation BDecoding by a CNOT and a Hadamard

Bob cannot read the message until he can name which Bell state he holds, and a projective Bell measurement is not something laboratory hardware performs directly — the entangled basis must first be rotated into the computational basis, where an ordinary readout works. The standard decoder applies a CNOT controlled by the transmitted qubit $A$ onto Bob's stored qubit $B$ , followed by a Hadamard on $A$ , after which measuring both qubits in the computational basis returns the two bits. We push all four Bell states through this fixed circuit and confirm the four outputs are the four distinct strings $ab$ .

Step 1. The circuit is $(H_A\otimes I_B)\,\mathrm{CNOT}_{A\to B}$ , then computational readout. CNOT flips $B$ iff $A=1$ ; the Hadamard on $A$ turns $\tfrac{1}{\sqrt2}(|0\rangle\pm|1\rangle)$ into $|0\rangle$ or $|1\rangle$ . Why: CNOT untangles the correlation into a product, and the Hadamard converts the surviving $\pm$ superposition into a definite bit.

Step 2. On $|\Phi^+\rangle$ : CNOT gives $\tfrac{1}{\sqrt2}(|00\rangle+|10\rangle)=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)|0\rangle$ , then $H_A$ gives $|00\rangle$ — outcome $ab=00$ . Why: the symmetric, same-bit Bell state decodes to the all-zero string, matching $U_{00}=I$ .

Step 3. On $|\Psi^+\rangle$ : CNOT gives $\tfrac{1}{\sqrt2}(|01\rangle+|11\rangle)=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)|1\rangle$ , then $H_A$ gives $|01\rangle$ — outcome $ab=01$ , matching $U_{01}=X$ . Why: the different-bit symmetric state carries $b=1$ in Bob's qubit, which the CNOT deposits cleanly.

Step 4. On $|\Phi^-\rangle$ : CNOT gives $\tfrac{1}{\sqrt2}(|00\rangle-|10\rangle)=\tfrac{1}{\sqrt2}(|0\rangle-|1\rangle)|0\rangle$ , then $H_A$ gives $|10\rangle$ — outcome $ab=10$ , matching $U_{10}=Z$ . Why: the relative minus sign becomes $a=1$ under the Hadamard, which reads phase as a bit.

Step 5. On $|\Psi^-\rangle$ : CNOT gives $\tfrac{1}{\sqrt2}(|01\rangle-|11\rangle)=\tfrac{1}{\sqrt2}(|0\rangle-|1\rangle)|1\rangle$ , then $H_A$ gives $|11\rangle$ — outcome $ab=11$ , matching $U_{11}=ZX$ . The four outputs $\{00,01,10,11\}$ are distinct, so the decoder is deterministic. Why: with all four Bell states mapped to all four strings, Bob's ordinary two-qubit measurement is a perfect Bell-state analyzer.

Go deeper · Derivation C · 1 minWhy one unassisted qubit cannot beat one bit

Derivation CWhy one unassisted qubit cannot beat one bit

It is tempting to imagine the qubit Alice sends simply "holds" two bits, but that would smash a hard limit on quantum communication, so the second bit must be located somewhere other than the channel qubit. Here we prove the limit itself. For any ensemble of single-qubit signal states an eavesdropper or receiver might prepare and measure, the Holevo quantity — an upper bound on the accessible classical information — cannot exceed one bit, because a single qubit's von Neumann entropy is at most $\log_2 2 = 1$ . The bound forces the resolution: the extra bit is supplied by the pre-shared ebit, which enlarges the signal space to four dimensions.

Step 1. Alice, unassisted, encodes her message in single-qubit states $\rho_x$ sent with probabilities $p_x$ ; the receiver's accessible information obeys $I_{\text{acc}}\leq\chi$ . Why: the Holevo theorem is the tight information ceiling for any preparation-and-measurement scheme, so bounding $\chi$ bounds everything.

Step 2. Write the Holevo quantity $\chi = S(\bar\rho) - \sum_x p_x S(\rho_x)$ with $\bar\rho = \sum_x p_x \rho_x$ . Since $S(\rho_x)\geq 0$ , the subtracted term only helps, so $\chi \leq S(\bar\rho)$ . Why: dropping a non-negative term gives a clean upper bound that depends only on the average state.

Step 3. The average state $\bar\rho$ is a single-qubit density matrix, and the entropy of a $d$ -dimensional state is at most $\log_2 d$ , so $S(\bar\rho)\leq \log_2 2 = 1$ . Hence $\chi\leq 1$ . Why: a lone qubit lives in a two-dimensional space, and no ensemble of such states carries more than one accessible bit.

Step 4. In superdense coding the distinguishable signal states are the two-qubit Bell states, living in a four-dimensional space made accessible by the shared ebit, so the ceiling rises to $\log_2 4 = 2$ . The doubling is bought, not free. Why: entanglement does not overfill the qubit — it changes which Hilbert space the message really lives in. *The key numerical anchors are $\log_2 2 = 1$ bit for an unassisted single qubit and $\log_2 4 = 2$ bits for the four-dimensional two-qubit space made available by the ebit.*

Limiting cases

The advantage is only as good as the shared entanglement. Strip it away and the protocol degrades to an ordinary one-bit channel; degrade the pair partially and the capacity slides smoothly between one and two bits.

Case	Result	Meaning
No shared entanglement	$\rho_A = \tfrac{1}{2}I \Rightarrow C = 1\text{ bit}$	Without an ebit the single qubit obeys the Holevo bound; the protocol collapses to plain one-bit transmission.
Perfect ebit	$\|\Phi^+\rangle \Rightarrow C = 2\text{ bits}$	The four encoded states are exactly orthogonal, so Bob decodes both bits deterministically.
Partially entangled pair	$\|\psi\rangle=\cos\theta\|00\rangle+\sin\theta\|11\rangle \Rightarrow C = 1 + S(\rho_A)$	For $\theta\neq\pi/4$ the encoded states are non-orthogonal; the recoverable information lies strictly between 1 and 2 bits.
Noisy / decohered pair	$\rho = (1-p)\|\Phi^+\rangle\langle\Phi^+\| + p\,\tfrac{I}{4}$	Depolarization blurs the Bell states, lowering the rate below 2 and eventually below 1 as $p$ grows.
Linear-optics Bell measurement	$C = \log_2 3 \approx 1.58\text{ bits}$	Only two of the four Bell states are unambiguously distinguishable with linear optics, capping the gain short of two.

Common trapThe single qubit carries two bits on its own

It does not. Without Bob's pre-shared half, the qubit Alice sends is the maximally mixed $\tfrac12 I$ , identical for every message, and carries exactly zero accessible information — an eavesdropper who grabs it in transit learns nothing. The "doubling" is a property of the joint two-qubit state, and it is paid for by the ebit that was distributed in advance. This never violates the Holevo bound: that bound limits an unassisted single qubit to one bit, whereas superdense coding places the message in the four-dimensional two-qubit space, where up to two bits are allowed. Nor does the protocol send anything faster than light — the qubit still travels down a physical channel at or below $c$ , and until it arrives Bob has only his useless mixed half.

Worked examples

1 · Sending the message 11, traced end to end (fully worked)

Alice wants to send $ab=11$ . Follow the shared state through her encoding, the transmission, and Bob's decoding, tracking amplitudes and the transmitted qubit's marginal.

Step 1. Start from the shared ebit $|\Phi^+\rangle = \tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ , Alice holding $A$ and Bob holding $B$ . The message $11$ selects the operator $U_{11}=ZX$ .

Step 2. Encode. Apply $X$ to Alice's qubit: $\tfrac{1}{\sqrt2}(|10\rangle+|01\rangle)$ . Then apply $Z$ to Alice's qubit, flipping the sign of $|10\rangle$ : $\tfrac{1}{\sqrt2}(|01\rangle-|10\rangle) = |\Psi^-\rangle$ . The shared pair is now the singlet, $|\Psi^-\rangle$ .

Step 3. Transmission. Alice sends qubit $A$ to Bob. Its marginal is $\rho_A = \operatorname{Tr}_B|\Psi^-\rangle\langle\Psi^-| = \tfrac12\big(|0\rangle\langle0| + |1\rangle\langle1|\big) = \tfrac12 I$ — maximally mixed, so anyone intercepting only this qubit learns nothing about $ab$ .

Step 4. Decode. Bob now holds both qubits and runs the decoder. CNOT gives $\tfrac{1}{\sqrt2}(|01\rangle-|11\rangle)=\tfrac{1}{\sqrt2}(|0\rangle-|1\rangle)|1\rangle$ ; the Hadamard on $A$ gives $|11\rangle$ . Measuring in the computational basis yields $ab=11$ with probability 1.

Step 5. Consistency check: the encode–decode pair is deterministic, $U_{11}=ZX$ maps $|\Phi^+\rangle\to|\Psi^-\rangle$ and the decoder maps $|\Psi^-\rangle\to|11\rangle$ , closing the loop. Two classical bits delivered by transmitting one qubit, with no channel leakage.

2 · Capacity from a partially entangled pair (fully worked numbers)

Alice and Bob share $|\psi\rangle = \cos\theta\,|00\rangle + \sin\theta\,|11\rangle$ with $\theta=\pi/6$ instead of the maximally entangled $\theta=\pi/4$ . Using the standard four-Pauli encoding, how many classical bits can Alice convey?

Step 1. The capacity of the four-Pauli superdense scheme on a shared pure state is $C = 1 + S(\rho_A)$ , where $\rho_A$ is Alice's marginal and $S$ is the von Neumann entropy in bits. This interpolates between $1$ (no entanglement) and $2$ (maximal entanglement).

Step 2. Compute the marginal. Tracing out Bob, $\rho_A = \cos^2\theta\,|0\rangle\langle0| + \sin^2\theta\,|1\rangle\langle1|$ . For $\theta=\pi/6$ , $\cos^2(\pi/6)=3/4$ and $\sin^2(\pi/6)=1/4$ , so $\rho_A = \operatorname{diag}(3/4,\,1/4)$ .

Step 3. Its entropy is the binary entropy $S(\rho_A) = H(1/4) = -\tfrac34\log_2\tfrac34 - \tfrac14\log_2\tfrac14 = \tfrac34(0.4150) + \tfrac14(2) = 0.3113 + 0.5 = \mathbf{0.811\ \text{bits}}$ .

Step 4. Capacity: $C = 1 + 0.811 = \mathbf{1.811\ \text{bits}}$ . The imperfect pair leaks about $0.19$ bit of the ideal two, because the four encoded states are no longer perfectly orthogonal and Bob's measurement occasionally confuses them. At $\theta=\pi/6$ the channel carries 1.81 bits, versus 2 bits for a perfect ebit and 1 bit with no entanglement.

Physics of the portrait

Superdense coding did not fall out of an experiment — it fell out of taking entanglement seriously as a resource you could budget, spend, and convert. That accounting-first way of thinking is the signature of Charles Bennett, who spent a career turning slippery quantum phenomena into hard currency.

Origins · 1 minCharles Bennett and how this was discovered

Charles Bennett

born 1943, American physicist and information theorist at IBM Research · the founding of quantum information theory

Charles H. Bennett is, more than any single figure, the architect of quantum information as a discipline of countable resources. In the 1970s he established the thermodynamics of computation, showing that logically reversible computation can in principle dissipate no energy and clarifying Landauer's principle that only erasure has an unavoidable cost. In 1984, with Gilles Brassard, he invented the BB84 protocol, the first quantum key distribution scheme. Then came the pair of results that define this article's neighbourhood: superdense coding with Stephen Wiesner in 1992, and quantum teleportation in 1993 with Brassard, Crépeau, Jozsa, Peres, and Wootters — discovered almost together and quickly recognized as time-reverse duals. Bennett went on to help build the quantitative theory of entanglement itself: entanglement distillation, the entanglement of formation, and the entanglement-assisted classical capacity (with Shor, Smolin, and Thapliyal). His unifying move throughout was to treat qubits, ebits, and cbits as interconvertible units obeying resource inequalities — the language in which superdense coding reads as a single clean line.

History

1973

— Holevo proves an unassisted qubit carries at most one accessible classical bit

1992

— Bennett and Wiesner propose superdense coding

1993

— quantum teleportation is discovered, the time-reverse dual of dense coding

1996

— Mattle, Weinfurter, Kwiat, and Zeilinger demonstrate dense coding with entangled photons

2004

— Schaetz and co-workers realize dense coding with a complete Bell analysis on trapped ions

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, §2.3 (superdense coding) and §1.3.6 · Preskill, Lecture Notes for Physics 219, Ch. 4 · Watrous, The Theory of Quantum Information, chapter on resource conversion · Wilde, Quantum Information Theory, Ch. 6 (three unit protocols) and Ch. 11 (Holevo bound) · Mermin, Quantum Computer Science, Ch. 6 · Bennett and Wiesner, Phys. Rev. Lett. 69, 2881 (1992). The encoding identities $(U_{ab}\otimes I)|\Phi^+\rangle = |\beta_{ab}\rangle$ follow from direct Pauli matrix computation; the CNOT–Hadamard decoder maps the four Bell states to the four distinct computational outputs $\{00,01,10,11\}$ . The partial-entanglement capacity $C=1+S(\rho_A)$ is the standard pure-state dense-coding result, giving $H(1/4)=0.811$ bits and hence $C=1.811$ bits at $\theta=\pi/6$ .

Prerequisites. Bell-State Measurement · Pauli operators and the Bell basis · Entanglement and the maximally entangled state. Leads to. Quantum Teleportation · Entanglement Swapping · Entanglement-assisted classical capacity.

Ask the tutor: Ask me to walk any message $ab$ through the full encode-transmit-decode cycle, or to work out the capacity of a noisy or partially entangled pair step by step.

How can one qubit carry two bits without breaking the Holevo bound?

The Holevo bound limits the accessible information of an unassisted qubit to one bit, and superdense coding never violates it. The message is not stored in the single qubit Alice transmits — its state is the maximally mixed $\tfrac12 I$ for all four messages, which carries zero accessible information on its own. What actually carries the two bits is the joint two-qubit state, which lives in a four-dimensional Hilbert space where up to $\log_2 4 = 2$ bits are allowed. That four-dimensional signal space is made available by the pre-shared ebit. So the bound applies to a different object (one qubit, $d=2$ ) than the one doing the work (two qubits, $d=4$ ), and both statements are true at once.

What exactly is an ebit, and where does the second bit come from?

An ebit is one unit of entanglement: a shared maximally entangled two-qubit state such as $|\Phi^+\rangle$ , one half held by each party. It is the currency of entanglement-assisted communication. In superdense coding the first bit is the ordinary capacity of sending one qubit, and the second bit is "paid for" by consuming the ebit. Concretely, the ebit is what makes Alice's four local Pauli operations map to four orthogonal global states — without it, four Paulis on a product state would give at most two distinguishable outcomes. The ebit must be distributed in advance, before Alice knows her message, which is why dense coding is useful precisely when entanglement is cheap to pre-share but channel use is precious.

Why is the transmitted qubit useless to an eavesdropper?

Because its reduced density matrix is $\tfrac12 I$ — the maximally mixed state — for every one of the four messages. An eavesdropper who taps the channel and captures Alice's qubit holds a state that is statistically identical regardless of whether she sent 00, 01, 10, or 11, so no measurement can extract any information about the message. This is a direct consequence of $|\Phi^+\rangle$ being maximally entangled: tracing out one half of any Bell state always leaves the other half maximally mixed. The message only becomes readable when the two halves are physically reunited and measured together, which is exactly what makes the correlation, not the qubit, the carrier of information.

How is superdense coding the mirror image of teleportation?

They are time-reverse duals with the quantum and classical channels swapped. Teleportation consumes one ebit and two classical bits to transmit one qubit: $1\text{ ebit} + 2\text{ cbits} \geq 1\text{ qubit}$ . Superdense coding consumes one ebit and one qubit to transmit two classical bits: $1\text{ qubit} + 1\text{ ebit} \geq 2\text{ cbits}$ . In both, the shared ebit is the catalyst, and both use the same hardware — a Bell measurement and local Pauli corrections. Run one protocol's circuit backwards and reinterpret which lines are quantum and which are classical, and you land on the other. This symmetry is why Bennett's group discovered them within a year of each other, and why they are taught as the two founding "unit protocols".

What happens if the shared pair is only partially entangled or noisy?

The clean two-bit result degrades gracefully. For a partially entangled pure state $\cos\theta|00\rangle+\sin\theta|11\rangle$ , the four Pauli-encoded states are no longer perfectly orthogonal, and the capacity falls to $1 + S(\rho_A)$ bits — exactly $1$ when $\theta=0$ (no entanglement) and $2$ when $\theta=\pi/4$ (maximal). For a noisy pair, modelled as a Werner-like mixture of $|\Phi^+\rangle$ with the maximally mixed state, the Bell states blur together and the rate drops below two and eventually below one as the noise grows. The lesson is the same throughout: the advantage tracks the quality of the entanglement, not the qubit.

Why can't linear optics do a complete Bell measurement, and what is the fix?

With only beam splitters, phase shifters, and photodetectors — linear optics — one can unambiguously identify only two of the four Bell states of a photon pair (the two antisymmetric-under-swap ones), while the other two produce identical detector patterns and cannot be told apart. That leaves three distinguishable outcomes, capping the dense-coding channel at $\log_2 3 \approx 1.58$ bits instead of the ideal two. The workarounds add a resource: hyperentanglement encodes an auxiliary degree of freedom (such as momentum or orbital angular momentum) so the Bell analyzer becomes complete, or one uses an atom–photon interface or trapped ions, where deterministic entangling gates give a full Bell measurement. Trapped-ion experiments reached the complete two-bit capacity as early as 2004.

Does superdense coding let Alice signal faster than light?

No. The two bits do not become available to Bob until Alice's qubit physically arrives, and that qubit travels down an ordinary channel at or below the speed of light. Before it arrives, Bob holds only his half of the ebit, whose marginal is the useless $\tfrac12 I$ — he can extract nothing. Entanglement supplies correlation, but correlation alone transmits no message: Alice's local Pauli operation has no observable effect on Bob's qubit until the two halves are brought together. This is the same no-signalling principle that protects relativity in every entanglement-based protocol; superdense coding merely uses the pre-shared correlation to double the classical payload of a legitimately transmitted qubit.