Skip to content

Lesson 10: Quantum Teleportation, Superdense Coding and Quantum Error Correction

10.1 Learning Outcomes

By the end of this hour a student will be able to:

  1. Derive the quantum teleportation protocol step by step and state its resource requirements.
  2. Explain why teleportation does not violate no-cloning or no-signalling.
  3. Derive superdense coding and state the duality between it and teleportation.
  4. Explain why classical repetition codes fail and how stabiliser measurement solves the problem.
  5. Construct the 3-qubit bit-flip and phase-flip codes and describe the 9-qubit Shor code.
  6. State the quantum threshold theorem and interpret physical-vs-logical qubit overhead.

10.2 Quantum Teleportation

The task. Alice holds an unknown qubit \(|\psi\rangle = \alpha|0\rangle + \beta|1\rangle\). She wants Bob to end up holding exactly \(|\psi\rangle\). They share one entangled pair and a classical channel. Alice may not send the qubit itself.

Resources: 1 shared ebit (\(|\Phi^+\rangle\)) + 2 classical bits + local operations.

The circuit

                        Alice
              +-----------------------------+
 q0 |psi> ----------*--------[H]------[M]---------.
                    |                             |  m0
 q1 (EPR) ---------[X]----------------[M]------.  |
              +----------------------------+  |  |  m1
                                               |  |
 q2 (EPR) -------------------------------[X]--[Z]------> |psi>
                        Bob                ^    ^
                                           |    |
                                     controlled by m1, m0

Reading the diagram: * is a CNOT control, [X] its target, [M] a measurement, and the final [X]/[Z] on Bob's wire are applied only if the corresponding classical bit is 1.

Qubits 1 and 2 begin in the Bell state \(|\Phi^+\rangle\).

Full derivation

Initial state (qubit 0 is Alice's unknown; qubits 1–2 are the shared pair):

\[ |\Psi_0\rangle = |\psi\rangle_0 \otimes |\Phi^+\rangle_{12} = \big(\alpha|0\rangle + \beta|1\rangle\big)\otimes\frac{1}{\sqrt2}\big(|00\rangle+|11\rangle\big) \]
\[ = \frac{1}{\sqrt2}\Big[\alpha|000\rangle + \alpha|011\rangle + \beta|100\rangle + \beta|111\rangle\Big] \]

Step 1 — CNOT (control \(q_0\), target \(q_1\)). Flip \(q_1\) where \(q_0=1\):

\[ |\Psi_1\rangle = \frac{1}{\sqrt2}\Big[\alpha|000\rangle + \alpha|011\rangle + \beta|110\rangle + \beta|101\rangle\Big] \]

Step 2 — Hadamard on \(q_0\). Substitute \(|0\rangle \to \frac{|0\rangle+|1\rangle}{\sqrt2}\) and \(|1\rangle\to\frac{|0\rangle-|1\rangle}{\sqrt2}\):

\[ |\Psi_2\rangle = \frac{1}{2}\Big[ \alpha(|0\rangle+|1\rangle)|00\rangle + \alpha(|0\rangle+|1\rangle)|11\rangle + \beta(|0\rangle-|1\rangle)|10\rangle + \beta(|0\rangle-|1\rangle)|01\rangle\Big] \]

Step 3 — regroup by Alice's two qubits \(q_0q_1\):

\[ |\Psi_2\rangle = \frac{1}{2}\Big[ |00\rangle\big(\alpha|0\rangle + \beta|1\rangle\big) + |01\rangle\big(\alpha|1\rangle + \beta|0\rangle\big) \]
\[ + \;|10\rangle\big(\alpha|0\rangle - \beta|1\rangle\big) + |11\rangle\big(\alpha|1\rangle - \beta|0\rangle\big)\Big] \]

Read this carefully. Bob's qubit already carries \(\alpha\) and \(\beta\) in all four branches — up to one of exactly four known corrections.

Step 4 — Alice measures \(q_0q_1\). Each outcome has probability \(\frac14\) (note \(|\alpha|^2+|\beta|^2=1\)), and this is what makes the protocol non-signalling: Alice's statistics carry no information about \(|\psi\rangle\).

Step 5 — Bob's correction, determined by the two classical bits \(m_0m_1\):

Alice's bits \(m_0m_1\) Bob's state Correction Why
\(00\) \(\alpha\lvert 0\rangle + \beta\lvert 1\rangle\) \(I\) already correct
\(01\) \(\alpha\lvert 1\rangle + \beta\lvert 0\rangle\) \(X\) bit flipped
\(10\) \(\alpha\lvert 0\rangle - \beta\lvert 1\rangle\) \(Z\) phase flipped
\(11\) \(\alpha\lvert 1\rangle - \beta\lvert 0\rangle\) \(ZX\) both

General rule: apply \(X^{m_1}Z^{m_0}\). Bob then holds exactly \(|\psi\rangle\).

Why no laws are broken

  • No-cloning is respected. Alice's measurement destroys her copy. At the end exactly one copy of \(|\psi\rangle\) exists, in Bob's hands. Teleportation moves a state; it does not duplicate it.
  • No-signalling is respected. Without the two classical bits, Bob's qubit is in the maximally mixed state \(\rho_B = I/2\) — no information at all. The protocol is limited by the classical channel, hence by the speed of light.
  • \(|\psi\rangle\) is never learned. Neither party discovers \(\alpha\) or \(\beta\); measuring to find them would require infinitely many copies.

Applications

Quantum repeaters (entanglement swapping), distributed quantum computing, and — importantly for fault tolerance — gate teleportation, the standard way to apply a \(T\) gate on encoded qubits using a pre-distilled magic state.

10.3 Superdense Coding

The task (the mirror image). Alice sends two classical bits to Bob by transmitting one qubit, using one pre-shared ebit.

Protocol.

  1. Alice and Bob share \(|\Phi^+\rangle = \frac{1}{\sqrt2}(|00\rangle+|11\rangle)\).
  2. To encode bits \(b_1b_0\), Alice applies \(Z^{b_1}X^{b_0}\) to her qubit only:
Bits \(b_1b_0\) Alice applies Resulting Bell state
\(00\) \(I\) \(\lvert \Phi^+\rangle\)
\(01\) \(X\) \(\lvert \Psi^+\rangle\)
\(10\) \(Z\) \(\lvert \Phi^-\rangle\)
\(11\) \(ZX\) \(\lvert \Psi^-\rangle\)
  1. Alice sends her single qubit to Bob.
  2. Bob, now holding both qubits, performs a Bell measurement (CNOT then \(H\), then measure both) and recovers \(b_1b_0\) with certainty, because the four Bell states are orthogonal.

Verification for \(b_1b_0 = 01\):

\[ (X\otimes I)\frac{1}{\sqrt2}(|00\rangle+|11\rangle) = \frac{1}{\sqrt2}(|10\rangle+|01\rangle) = |\Psi^+\rangle \ \checkmark \]

The duality.

\[ \text{Teleportation: } 1 \text{ ebit} + 2 \text{ cbits} \Rightarrow 1 \text{ qubit sent} \]
\[ \text{Superdense coding: } 1 \text{ ebit} + 1 \text{ qubit} \Rightarrow 2 \text{ cbits sent} \]

Holevo's bound (one qubit carries at most one classical bit) is not violated: the second bit was carried by the pre-shared entanglement, distributed earlier.

10.4 Quantum Error Correction

Three obstacles that classical QEC does not face

  1. No-cloning — you cannot make three copies of \(|\psi\rangle\).
  2. Measurement destroys — reading the qubit to check it collapses the superposition.
  3. Continuous errors — a qubit can suffer \(R_x(0.01)\), not just a discrete flip. There are uncountably many possible errors.

The three ideas that solve them

  1. Encode, don't copy. Spread one logical qubit across several physical qubits in an entangled state.
  2. Measure the syndrome, not the data. Measure operators that reveal which error occurred without revealing \(\alpha\) or \(\beta\).
  3. Error discretisation. A continuous error is a superposition of \(\{I, X, Y, Z\}\); the syndrome measurement projects it onto one of these. Correcting \(X\), \(Z\) and \(Y = iXZ\) therefore corrects everything. This is the theorem that makes QEC possible at all.

The 3-qubit bit-flip code

Encoding.

\[ |0\rangle_L = |000\rangle, \qquad |1\rangle_L = |111\rangle \]
\[ \alpha|0\rangle+\beta|1\rangle \ \longmapsto\ \alpha|000\rangle + \beta|111\rangle \]

Circuit: two CNOTs from \(q_0\) to \(q_1\) and \(q_2\). Note this is not cloning — the encoded state is entangled, not \(|\psi\rangle^{\otimes3}\).

Syndrome measurement. Measure the two stabilisers

\[ S_1 = Z_1Z_2, \qquad S_2 = Z_2Z_3 . \]

These are parity checks. They commute with each other and with the logical operators, so they reveal the error location without disturbing \(\alpha,\beta\).

\(S_1\) \(S_2\) Diagnosis Correction
\(+1\) \(+1\) no error none
\(-1\) \(+1\) flip on qubit 1 \(X_1\)
\(-1\) \(-1\) flip on qubit 2 \(X_2\)
\(+1\) \(-1\) flip on qubit 3 \(X_3\)

Corrects any single bit flip. Fails on two or more.

Failure probability. If each qubit flips independently with probability \(p\):

\[ P_{\text{fail}} = 3p^2(1-p) + p^3 = 3p^2 - 2p^3 \approx 3p^2 . \]

For \(p = 0.01\): \(P_{\text{fail}} \approx 2.98\times10^{-4}\) — a 34-fold improvement. The code helps whenever \(3p^2 < p\), i.e. \(p < 1/3\). This threshold behaviour is the seed of the full threshold theorem.

The 3-qubit phase-flip code

A \(Z\) error is a bit flip in the Hadamard basis, because \(HZH = X\). So conjugate the whole bit-flip code by Hadamards:

\[ |0\rangle_L = |{+}{+}{+}\rangle, \qquad |1\rangle_L = |{-}{-}{-}\rangle \]

with stabilisers \(X_1X_2\) and \(X_2X_3\).

The 9-qubit Shor code

Concatenate the two: encode with the phase-flip code, then encode each of its three qubits with the bit-flip code.

\[ |0\rangle_L = \frac{1}{2\sqrt2}\big(|000\rangle+|111\rangle\big)^{\otimes 3} \]
\[ |1\rangle_L = \frac{1}{2\sqrt2}\big(|000\rangle-|111\rangle\big)^{\otimes 3} \]

This corrects an arbitrary single-qubit error — \(X\), \(Z\), \(Y\), or any continuous rotation — on any one of the nine qubits. It was the first QEC code (Peter Shor, 1995) and proved that fault-tolerant quantum computing is possible in principle.

Other important codes.

Code \([[n,k,d]]\) Note
Shor \([[9,1,3]]\) first code
Steane \([[7,1,3]]\) CSS code from the classical Hamming \([7,4,3]\) code
Perfect / Laflamme \([[5,1,3]]\) smallest possible distance-3 code
Surface code \([[d^2,1,d]]\) 2-D nearest-neighbour only; today's leading candidate

Notation \([[n,k,d]]\): \(n\) physical qubits, \(k\) logical qubits, distance \(d\), correcting \(t = \lfloor (d-1)/2\rfloor\) errors.

The threshold theorem

Theorem (Aharonov–Ben-Or; Knill–Laflamme–Zurek; Kitaev). If the physical error rate per gate is below a constant threshold \(p_{\text{th}}\), then arbitrarily long quantum computations can be performed reliably, with only polylogarithmic overhead in the number of qubits and gates.

For the surface code, \(p_{\text{th}} \approx 1\%\) — within reach of today's hardware. The price is overhead: roughly \(10^3\)–\(10^4\) physical qubits per logical qubit for cryptographically relevant computations. This is exactly why estimates for "breaking RSA-2048 with Shor" quote millions of physical qubits.

10.5 Solved Problems

Problem 1 — Teleporting a specific state

Alice teleports \(|\psi\rangle = \frac{1}{\sqrt5}\big(|0\rangle + 2|1\rangle\big)\) and measures \(m_0m_1 = 10\). Write Bob's pre-correction state, the correction, and verify the final state.

Solution. Here \(\alpha = \frac{1}{\sqrt5}\), \(\beta = \frac{2}{\sqrt5}\).

From the branch table, outcome \(10\) leaves Bob with

\[ \alpha|0\rangle - \beta|1\rangle = \frac{1}{\sqrt5}|0\rangle - \frac{2}{\sqrt5}|1\rangle . \]

Correction: \(m_0 = 1\), \(m_1 = 0\), so apply \(X^{0}Z^{1} = Z\):

\[ Z\left(\frac{1}{\sqrt5}|0\rangle - \frac{2}{\sqrt5}|1\rangle\right) = \frac{1}{\sqrt5}|0\rangle + \frac{2}{\sqrt5}|1\rangle = |\psi\rangle \ \checkmark \]

Note that the probability of this outcome is \(\frac14\) regardless of \(\alpha,\beta\) — Alice learns nothing about the state she teleported.

Problem 2 — Superdense coding, full trace

Alice wants to send \(11\). Trace the protocol and verify Bob decodes correctly.

Solution.

Shared state: \(|\Phi^+\rangle = \frac{1}{\sqrt2}(|00\rangle+|11\rangle)\).

Encode: \(b_1b_0 = 11\), so Alice applies \(ZX\) to her qubit.

\[ (X\otimes I)|\Phi^+\rangle = \tfrac{1}{\sqrt2}\big(|10\rangle + |01\rangle\big) \]
\[ (Z\otimes I)\tfrac{1}{\sqrt2}\big(|10\rangle+|01\rangle\big) = \tfrac{1}{\sqrt2}\big(-|10\rangle + |01\rangle\big) = |\Psi^-\rangle \]

Decode (Bell measurement): apply CNOT\(_{0\to1}\) then \(H\) on \(q_0\).

\[ \mathrm{CNOT}\,\tfrac{1}{\sqrt2}\big(|01\rangle - |10\rangle\big) = \tfrac{1}{\sqrt2}\big(|01\rangle - |11\rangle\big) = \tfrac{1}{\sqrt2}\big(|0\rangle - |1\rangle\big)\otimes|1\rangle \]
\[ (H\otimes I): \quad H\tfrac{1}{\sqrt2}(|0\rangle-|1\rangle) = H|-\rangle = |1\rangle \]
\[ \Rightarrow\ |11\rangle . \]

Bob measures and reads \(b_1b_0 = 11\) ✓ with probability 1. Two classical bits delivered by one transmitted qubit.

Problem 3 — Bit-flip code syndrome decoding

The state \(\alpha|000\rangle + \beta|111\rangle\) suffers an \(X\) error on qubit 2. Compute the syndrome and show that correction restores the state exactly, without revealing \(\alpha,\beta\).

Solution.

Corrupted state: \(X_2\big(\alpha|000\rangle + \beta|111\rangle\big) = \alpha|010\rangle + \beta|101\rangle\).

Measure \(S_1 = Z_1Z_2\) (parity of qubits 1 and 2):

  • \(|010\rangle\): bits \(0,1\) → parity odd → eigenvalue \(-1\)
  • \(|101\rangle\): bits \(1,0\) → parity odd → eigenvalue \(-1\)

Both branches give \(-1\), so the measurement returns \(S_1 = -1\) without collapsing the superposition — this is the key point.

Measure \(S_2 = Z_2Z_3\):

  • \(|010\rangle\): bits \(1,0\) → odd → \(-1\)
  • \(|101\rangle\): bits \(0,1\) → odd → \(-1\)

\(S_2 = -1\).

Syndrome \((-1,-1)\) ⟶ table says error on qubit 2 ⟶ apply \(X_2\):

\[ X_2\big(\alpha|010\rangle + \beta|101\rangle\big) = \alpha|000\rangle + \beta|111\rangle \ \checkmark \]

Why \(\alpha,\beta\) stay secret. The stabilisers act identically on \(|0\rangle_L\) and \(|1\rangle_L\), so their eigenvalues depend only on the error, never on the encoded amplitudes. This is the central design principle of stabiliser codes.

Problem 4 — Error discretisation

A qubit in the bit-flip code suffers the continuous rotation \(R_x(\epsilon) = \cos\frac{\epsilon}{2}I - i\sin\frac{\epsilon}{2}X\) on qubit 1. Show that syndrome measurement corrects it exactly.

Solution. Write \(c = \cos\frac{\epsilon}{2}\), \(s = \sin\frac{\epsilon}{2}\). The corrupted state is

\[ |\psi'\rangle = \big(cI - is X_1\big)|\psi_L\rangle = c\,|\psi_L\rangle - is\,X_1|\psi_L\rangle . \]

This is a superposition of "no error" and "\(X_1\) error".

Measuring the syndrome \((S_1, S_2)\) projects onto one of the error subspaces:

  • With probability \(|c|^2 = \cos^2\frac{\epsilon}{2}\): syndrome \((+1,+1)\), state collapses to \(|\psi_L\rangle\). No error occurred.
  • With probability \(|s|^2 = \sin^2\frac{\epsilon}{2}\): syndrome \((-1,+1)\), state collapses to \(X_1|\psi_L\rangle\). Apply \(X_1\) to restore \(|\psi_L\rangle\).

In both cases the final state is exactly \(|\psi_L\rangle\) — the small continuous error has been digitised by the measurement and then removed. No approximation, no residual error.

Significance. This is why a code that corrects only the discrete set \(\{X,Y,Z\}\) corrects the continuum of all possible single-qubit errors. Without it, QEC would be impossible.

Problem 5 — Logical error rate and the value of concatenation

Physical qubits have error rate \(p = 10^{-3}\). Compute the logical error rate for (a) the 3-qubit code, (b) one level of concatenation (\(9\) qubits), (c) two levels (\(27\) qubits). How many levels are needed for \(10^{-15}\)?

Solution. For a distance-3 code, \(p_L \approx 3p^2\) (fails only on 2 or more errors).

(a) \(p_L = 3(10^{-3})^2 = 3\times10^{-6}\).

(b) Concatenation applies the same formula to the previous level's rate:

\[ p_L^{(2)} = 3\left(3\times10^{-6}\right)^2 = 3 \times 9\times10^{-12} = 2.7\times10^{-11} . \]

(c) \(p_L^{(3)} = 3\left(2.7\times10^{-11}\right)^2 = 3\times 7.29\times10^{-22} = 2.19\times10^{-21}\).

So three levels already beat \(10^{-15}\) comfortably; two levels (\(2.7\times10^{-11}\)) do not.

Cost. Level \(k\) uses \(3^k\) physical qubits per logical qubit:

Level Physical qubits Logical error rate
1 3 \(3\times10^{-6}\)
2 9 \(2.7\times10^{-11}\)
3 27 \(2.2\times10^{-21}\)

Double-exponential suppression for exponential cost — the essence of the threshold theorem. Note that with \(p = 0.4 > 1/3\) the recursion diverges: concatenation makes things worse. This is the threshold in action.

10.6 Practice Problems

  1. Alice teleports \(|+\rangle\) and measures \(11\). What is Bob's pre-correction state and what correction restores \(|+\rangle\)? [Ans: \(\frac{1}{\sqrt2}(|1\rangle-|0\rangle)\); apply \(ZX\)]
  2. Verify \((Z\otimes I)|\Phi^+\rangle = |\Phi^-\rangle\).
  3. Show that Bob's reduced density matrix before receiving Alice's bits is \(I/2\), proving no-signalling.
  4. Encode \(\frac{1}{\sqrt2}(|0\rangle+|1\rangle)\) in the phase-flip code and show a \(Z_2\) error is detected.
  5. For the 3-qubit code with \(p = 0.05\), compute the logical error rate and the improvement factor. [Ans: \(7.1\times10^{-3}\); ≈ 7×]
  6. Why is the surface code preferred over the Shor code on superconducting hardware?
  7. Explain why teleportation requires two classical bits and not one.

10.7 Summary

  • Teleportation: 1 ebit + 2 cbits move an unknown qubit. Alice's Bell measurement leaves Bob's qubit correct up to one of four Pauli corrections \(X^{m_1}Z^{m_0}\). No cloning (Alice's copy is destroyed) and no signalling (Bob sees \(I/2\) until the bits arrive).
  • Superdense coding: the dual protocol — 1 ebit + 1 qubit send 2 classical bits, decoded by a Bell measurement.
  • QEC cannot copy. It encodes into entangled states and measures stabilisers, which reveal the error without revealing the data.
  • Error discretisation means correcting \(\{X,Y,Z\}\) corrects every continuous error.
  • 3-qubit codes fix bit flips or phase flips; the 9-qubit Shor code fixes both. \([[n,k,d]]\) notation; the surface code leads today.
  • Threshold theorem: below \(p_{\text{th}}\approx1\%\), arbitrarily long computation is possible with polylog overhead — but with \(10^3\)–\(10^4\) physical qubits per logical qubit in practice.

Unit I is now complete. You should be able to: manipulate states and gates algebraically, reason on the Bloch sphere, compute measurement statistics, detect entanglement, and explain why interference (not parallelism) is the source of quantum power.

Next lesson: L11 — Quantum oracles and phase kick-back (Unit II begins).

Practise now: unit1-qubit-lab.html → Teleportation tab runs the protocol shot by shot and shows Bob's state before and after correction; the QEC tab lets you inject errors and read syndromes.