Superdense Coding
One qubit, handed from Alice to Bob, can carry at most one classical bit of freshly
chosen information — that is a hard law (Holevo's bound). So how could Alice possibly cram two
classical bits into a single qubit she sends? The trick, called superdense coding, is
that Alice and Bob prepared for this in advance: they share an entangled
Bell pair.
With that shared resource already in hand, one physical qubit travelling across the channel delivers a
full two-bit message. It is the exact mirror image of
quantum teleportation,
where two classical bits move one qubit.
The magic ingredient is that Alice can steer the joint state of the pair by touching only her
own half. A gentle nudge to her qubit rotates the shared Bell state into a different Bell state — and
because the four Bell states are perfectly distinguishable, Bob can read back exactly which nudge she
chose.
The setup: a pre-shared Bell pair
Before any message exists, someone prepares the Bell state |\Phi^+\rangle and
gives qubit A to Alice and qubit B to Bob,
who may then walk to opposite ends of the world:
|\Phi^+\rangle = \tfrac{1}{\sqrt2}\big(|00\rangle + |11\rangle\big).
Later Alice decides on a two-bit message — one of 00, 01, 10, 11 — and applies
exactly one of the four Pauli
gates to her qubit. Each choice rotates the shared pair into a different Bell state.
She then physically sends her single qubit to Bob. Now Bob holds both halves, measures them
together in the Bell basis, and recovers both bits. One qubit crossed the channel; two bits arrived.
The four-way dictionary
Applying a Pauli to Alice's half of |\Phi^+\rangle lands the pair on one of the
four Bell states. Reveal the rows one at a time — this table is the whole codebook:
Read the states off with the identities (I\otimes I)|\Phi^+\rangle = |\Phi^+\rangle,
(Z\otimes I)|\Phi^+\rangle = |\Phi^-\rangle,
(X\otimes I)|\Phi^+\rangle = |\Psi^+\rangle, and
(ZX\otimes I)|\Phi^+\rangle = -|\Psi^-\rangle. Four inputs, four
orthogonal outputs — that orthogonality is exactly why Bob can never confuse them.
The circuit end to end
The whole protocol is three moves on two wires. Alice owns the top wire, Bob the bottom. After Alice's
Pauli, her qubit crosses to Bob (the dashed hand-off); Bob then runs the Bell measurement — a
CNOT then a Hadamard, the reverse of how a Bell pair is built — and measures both wires
to read the two bits. Step through it:
A Bell measurement is just "un-building" the entanglement: the CNOT copies the correlation back onto the
control wire and the Hadamard collapses the superposition, so the two qubits come out as an ordinary
two-bit number — precisely the message Alice encoded.
Worked example: Alice encodes “10”
Say Alice wants to send the bits 10. The table tells her to apply
X to her qubit. Acting on the first slot of
|\Phi^+\rangle, the bit-flip swaps 0\leftrightarrow 1
in each term:
(X\otimes I)\,\tfrac{1}{\sqrt2}\big(|00\rangle + |11\rangle\big) = \tfrac{1}{\sqrt2}\big(|10\rangle + |01\rangle\big) = |\Psi^+\rangle.
The shared pair is now |\Psi^+\rangle. Alice ships her single qubit to Bob,
who holds both halves and runs his decoder. Recall |\Psi^+\rangle came from the
computational input |10\rangle under "Hadamard then CNOT," so running that
circuit backwards (CNOT, then Hadamard on the top wire) returns
|10\rangle:
|\Psi^+\rangle \;\xrightarrow{\;\text{CNOT}\;}\; \tfrac{1}{\sqrt2}\big(|11\rangle + |01\rangle\big) \;\xrightarrow{\;H_1\;}\; |10\rangle.
Bob measures and reads 1 then 0 with certainty —
Alice's two bits, delivered by one qubit. Every other message decodes the same way: the Bell state
is the message.
Worked example: why “01” needs Z
For the message 01 the table says apply Z. The
phase-flip leaves |0\rangle alone and negates |1\rangle,
so acting on Alice's slot:
(Z\otimes I)\,\tfrac{1}{\sqrt2}\big(|00\rangle + |11\rangle\big) = \tfrac{1}{\sqrt2}\big(|00\rangle - |11\rangle\big) = |\Phi^-\rangle.
Notice Alice changed nothing about which outcomes are possible —
|\Phi^+\rangle and |\Phi^-\rangle both give
00 or 11 on a plain joint measurement. All she
altered is the relative phase, the invisible-looking minus sign. Yet a full Bell
measurement (which interferes the qubits) sees that phase perfectly and pulls out
01. This is why superdense coding needs the Bell measurement, not two ordinary
ones: half the message lives in a phase that only interference can reveal.
The mirror of teleportation
Superdense coding and teleportation are two readings of the same resource — a shared Bell pair — run in
opposite directions:
-
Teleportation: spend 1 shared Bell pair and send
2 classical bits to move 1 unknown
qubit of state.
-
Superdense coding: spend 1 shared Bell pair and send
1 qubit to move 2 classical
bits.
Read one line, then swap "qubit" with "two classical bits" and reverse the arrow, and you get the other.
Entanglement is the currency; the two protocols just exchange it for different things.
- a pre-shared Bell pair |\Phi^+\rangle lets Alice send
two classical bits by transmitting just one qubit;
- she encodes with one Pauli on her half:
I\!\to\!\Phi^+ (00), Z\!\to\!\Phi^- (01),
X\!\to\!\Psi^+ (10), ZX\!\to\!\Psi^- (11);
- Bob decodes with a Bell measurement (CNOT, then Hadamard, then measure both),
perfectly distinguishing the four orthogonal Bell states;
- it is the dual of teleportation — and Holevo still holds, because the pair was
shared in advance at the cost of prior communication.
Picture the channel as a single lane just wide enough for one qubit. Naively a qubit carries one bit, so
two bits should need two trips. Superdense coding sends only one qubit down the lane and still delivers
two bits — the second bit rides along on the entanglement Alice and Bob set up earlier. It feels like
getting a package through a one-item slot by having already mailed half of it last week. The lane never
widened; the preparation did the work.
It is tempting to conclude "a qubit carries two bits, so Holevo is wrong." It is not. The qubit only
carries two bits because of the pre-shared entanglement, and establishing that Bell pair
already required sending a qubit earlier (or otherwise distributing entanglement). Count honestly and two
qubits crossed the channel in total, for two bits — one bit per qubit, exactly as Holevo demands. Strip
away the shared pair and a lone qubit, with no entangled partner waiting at the far end, carries just
one bit. The doubling is a gift of the shared resource, not a loophole in the bound.