Two classical bits, one transmitted qubit and a shared pair
How can sending one qubit communicate two classical bits? The question is incomplete until the initial resources are named. In superdense coding, Alice and Bob already share an entangled pair before Alice learns which message to send. Alice applies one of four operations to her half, sends that qubit to Bob, and Bob measures the two qubits together. The communicated object is one qubit; the pre-shared entangled pair is an additional resource.
Definition 1.4.1
(Superdense coding protocol)
Alice and Bob share the Bell pair , Alice holding the first qubit. To send a message , Alice applies a local operation to her qubit and transmits it to Bob; Bob then measures both qubits. The resources are one transmitted qubit and one previously shared entangled pair.
The protocol sends one of four classical messages with certainty in an ideal noiseless model. We derive its circuit and then prove two limitations: Bob learns nothing about the message before the qubit arrives, and a single two-dimensional system without a pre-shared resource cannot perfectly encode four messages in one use. Neither statement requires an informal claim that a qubit is literally a classical bit.
The four encodings
Let the message be . Alice and Bob initially hold the first and second qubits, respectively, of
Alice applies to her qubit, with the rightmost operator acting first, then transmits it. With the displayed phase convention, the four joint states just before Bob decodes are
They are mutually orthonormal Bell states. A global minus sign would not change a physical state or a measurement probability; the relative signs inside each state matter. Bob can therefore distinguish the four messages by a joint measurement in this basis.
Common mistake
(Without the gates, the sign is invisible)
If Bob measured the two qubits directly in the computational basis, he could not tell from : both and give outcome or with probability each. The bit lives in the relative sign, which a computational-basis measurement does not see. Decoding therefore needs a basis change first.
Decoding with ordinary gates
Bob applies controlled-NOT with the received qubit as control and his original qubit as target, then applies to the received qubit. Finally he measures both qubits in the computational basis.
Theorem 1.4.2
(Correctness of superdense decoding)
For every , Bob's gates map to . His computational-basis measurement therefore returns the first bit and the second bit with probability one.
Proof
(Tracking the state through the circuit)
For every ,
The first arrow uses and then . The second arrow replaces the target by . For the third, put ; since for bits, the state is , because . Since , the last gate maps it to .
The overall factor is a global phase and has no effect on the outcome. The order of the output bits follows from the chosen encoding and wire order; swapping either convention can change the labels in a table without changing the physical protocol.
Proof X-ray
Why the two bits appear on different wires
The decoding works because the two bits are stored in two different features of the Bell state, and the gates separate them. The bit-flip label determines which computational-basis pair appears: the controlled-NOT copies the difference of the two wires into the second qubit, so the second qubit becomes exactly , independent of . After that gate, the relative phase remains on the first qubit, and Hadamard turns that phase information into the computational-basis bit . The one non-obvious point is that the sign factor differs from only by a global factor , so the phase bit is unaffected by . Both operations are unitary; the classical message appears only when Bob measures.
Worked example 1.4.3
(Sending the message (1,1))
If Alice sends , she prepares . Bob's controlled-NOT changes it to . Hadamard on the first qubit yields , so Bob reads with probability one. The minus sign on the whole final state is unobservable.
Worked example 1.4.4
(Sending the message (0,1): a pure phase)
For Alice applies only and prepares . The controlled-NOT sends and , giving . Hadamard yields , so Bob reads . This is the case in which Alice flips no bit: the message is carried entirely by a sign that the decoding turns into a measurable bit.
Bob's state before transmission
Alice's choice alone cannot signal to Bob.
Proposition 1.4.5
(Bob's marginal does not depend on Alice's operation)
For any unitary applied to Alice's half of , Bob's reduced density operator is . In particular it is independent of the message .
Proof
(Tracing out Alice's qubit)
For any unitary ,
The vectors and are orthonormal. The partial trace over Alice's qubit sends a cross term to ; the diagonal terms each have trace one, contributing .
The probability of every measurement outcome available on Bob's qubit before transmission is therefore independent of Alice's message. The dependence becomes accessible only after Bob receives the other qubit and can measure the pair.
Worked example 1.4.6
(A unitary outside the four encodings)
Let Alice apply the Hadamard gate, which is not one of . Then , with . Tracing out Alice's qubit leaves for Bob again. No local choice on Alice's side changes what Bob sees on his qubit alone.
Counterexample mode
Entanglement does not send a message by itself
The false claim is that Bob can read Alice's two bits as soon as she chooses a local gate, since the joint Bell state changes. His local state is still for all four choices, so even a different measurement strategy on his qubit alone has the same outcome distribution. The repaired statement is that Alice's operation selects a distinguishable joint state, which Bob can access only after communication of her qubit.
Why the resource count is not a contradiction
We first prove a dimension bound for perfect discrimination.
Theorem 1.4.7
(Dimension bound for perfect discrimination)
Suppose density operators on a -dimensional space are identified with certainty by a measurement with effects , , so . Then the supports of the are mutually orthogonal, and .
Proof
(Orthogonal supports)
Each . The condition forces to act as the identity on the support of ; the zero probabilities for force it to vanish on the support of . Eigenspaces of the Hermitian operator for eigenvalues 1 and 0 are orthogonal. Hence the nonzero supports of distinct states are mutually orthogonal, and .
Without a pre-shared resource, suppose Alice can encode a classical message as an arbitrary one-qubit state and sends that qubit to Bob. Bob may attach a fixed independent ancilla in state and perform a joint POVM with effects ; it induces the POVM on the received qubit, where is the ancilla. Each is positive, , and . The dimension bound therefore gives at most two messages perfectly distinguishable in one use. For example, the four states cannot be perfectly identified from one unknown qubit: the first two already span the space, leaving no orthogonal support for the others.
With exactly one shared Bell pair and no other pre-shared quantum system, after Alice sends one qubit Bob holds a four-dimensional two-qubit system. A fixed independent ancilla can again be absorbed into the measurement, so the effective dimension is four. The four orthogonal encodings attain the resulting ceiling of four perfectly distinguishable messages in this setting. The bound is about zero-error, one-use discrimination under the stated resources. It is not a proof of a general noisy-channel capacity theorem, nor does it say that every use of one qubit without entanglement yields a deterministic bit under noise.
Common mistake
(The shared pair is part of the cost)
It is wrong to say that one qubit carries two bits with no further resource. Without the pre-shared pair, one qubit distinguishes at most two messages perfectly. The pair has to be distributed beforehand, and after decoding the four possible states are computational-basis product states, so the original Bell pair is consumed by this protocol.
Teleportation has a different input and resource task: it transfers an unknown qubit state using a shared pair and two classical bits. Neither protocol supplies free communication or violates no-cloning.
Practice with solutions
Quick check
(Alice encodes . Write the joint state before Bob decodes and the two measured output bits in the order used above.)
Use the table of four encodings.
Solution · Answer
The encoded state is . Bob reads with certainty.
Quick check
(What is Bob's reduced state before transmission when Alice chooses either or ? Can a local measurement distinguish those choices?)
Use the marginal computation for a general unitary .
Solution · Answer
In both cases his density operator is ; every local measurement therefore has the same outcome distribution.
Quick check
(Why can no measurement perfectly distinguish , and when a single unknown qubit is supplied?)
Count orthogonal supports in a two-dimensional space.
Solution · Answer
Perfect discrimination would require three nonzero mutually orthogonal supports in a two-dimensional space. Equivalently, and already span that space, so no third nonzero state can be orthogonal to both.
Quick check
(In the stated one-Bell-pair protocol, how many classical messages are perfectly distinguished, how many qubits are transmitted during encoding, and what shared resource was needed beforehand?)
Separate the transmitted qubit from the pre-shared resource.
Solution · Answer
Four messages are distinguished; Alice sends one qubit after applying her gate, and the parties must already share one Bell pair. The protocol's resource count includes both transmission and the shared pair.