Simple Algorithm
Quantum Computing · Excited State · 20 min

No-Cloning & No-Signalling

You cannot copy an unknown quantum state, or use entanglement to signal.

Quantum mechanics is linear, and linearity alone rules out a universal copier. The same structure guarantees that entanglement, for all its instant correlations, can never carry a message. In this lesson you prove both theorems and see why each one needs the other.

What you’ll learn

  • Prove the no-cloning theorem from linearity and, more sharply, from the preservation of inner products.
  • Characterize exactly which sets of states can be cloned, and build a cloner for the XX basis.
  • Prove the no-signalling theorem for an arbitrary local operation using the partial trace.
  • Explain why a perfect cloner would allow superluminal signalling, and place no-cloning among its related no-go theorems.

What a cloner would have to do

Model a would-be copier in full generality. It receives the input system AA in an unknown state ∣ψ⟩\ket\psi, a blank system BB in a fixed state ∣e⟩\ket{e}, and its own machinery MM in a fixed state ∣m⟩\ket{m}. Closed quantum evolution is unitary, and any physical process, including measurements whose outcomes are later discarded, can be modeled as a unitary on a larger system (Stinespring dilation). So nothing is lost by asking for a single unitary UU with

U(∣ψ⟩A∣e⟩B∣m⟩M)=∣ψ⟩A∣ψ⟩B∣mψ⟩M.U\big(\ket\psi_A\ket{e}_B\ket{m}_M\big)=\ket\psi_A\ket\psi_B\ket{m_\psi}_M .

The machine may end in a state ∣mψ⟩\ket{m_\psi} that depends on the input; this also absorbs any global phase.

Some sets can be cloned. The CNOT clones {∣0⟩,∣1⟩}\{\ket0,\ket1\} with blank ∣0⟩\ket0 and no machine at all. Any orthonormal basis {∣k⟩}\{\ket{k}\} can be cloned the same way: rotate it to the computational basis, fan out with CNOTs, and rotate both outputs back. The real question is whether one unitary can clone states that are not orthogonal.

Test this with two copiers: the usual one for the ZZ basis, and one you build for the XX basis.

Check your understanding

A unitary UU clones {∣0⟩,∣1⟩}\{\ket0,\ket1\}, and another unitary VV clones {∣+⟩,∣−⟩}\{\ket+,\ket-\}. Which statement is true?

  1. Applying UU and then VV clones all four states
  2. A cleverly designed unitary with a large enough machine register clones all four
  3. No unitary clones all four, because ∣0⟩\ket0 and ∣+⟩\ket+ are neither orthogonal nor equal
  4. The combination (U+V)/2(U+V)/\sqrt2 clones all four
Show the answer

C. No unitary clones all four, because ∣0⟩\ket0 and ∣+⟩\ket+ are neither orthogonal nor equal

You will prove next that a clonable set must be pairwise orthogonal or identical. Since ∣⟨0∣+⟩∣=1/2\abs{\braket{0|+}}=1/\sqrt2, no unitary, with any machine, clones both ∣0⟩\ket0 and ∣+⟩\ket+.

Proof 1: linearity

DerivationNo-cloning from linearity
  1. U∣0⟩∣e⟩=∣0⟩∣0⟩,U∣1⟩∣e⟩=∣1⟩∣1⟩U\ket0\ket{e}=\ket0\ket0,\qquad U\ket1\ket{e}=\ket1\ket1

    Suppose UU clones both basis states. (Leave out the machine for now; Proof 2 puts it back.)

  2. U(α∣0⟩+β∣1⟩)∣e⟩=α∣00⟩+β∣11⟩U\big(\alpha\ket0+\beta\ket1\big)\ket{e}=\alpha\ket{00}+\beta\ket{11}

    Linearity: the superposition passes through term by term.

  3. ∣ψ⟩∣ψ⟩=α2∣00⟩+αβ∣01⟩+αβ∣10⟩+β2∣11⟩\ket\psi\ket\psi=\alpha^2\ket{00}+\alpha\beta\ket{01}+\alpha\beta\ket{10}+\beta^2\ket{11}

    What a genuine copy of ∣ψ⟩=α∣0⟩+β∣1⟩\ket\psi=\alpha\ket0+\beta\ket1 would be.

  4. αβ=0 ⟹ ∣ψ⟩∈{∣0⟩,∣1⟩} (up to phase)\alpha\beta=0\ \Longrightarrow\ \ket\psi\in\{\ket0,\ket1\}\ \text{(up to phase)}

    Match the ∣01⟩\ket{01} and ∣10⟩\ket{10} coefficients (even allowing an overall phase eiγe^{i\gamma} on the right): they force αβ=0\alpha\beta=0. Every genuine superposition is mis-copied.

  5. Worse than mis-copied: for αβ≠0\alpha\beta\neq0 the output α∣00⟩+β∣11⟩\alpha\ket{00}+\beta\ket{11} is entangled. A basis copier converts superposition into entanglement with the blank, exactly what you saw in the circuit.

Predict before you look

The CNOT copier receives ∣+⟩\ket+. What is the fidelity ⟨+∣ρ∣+⟩\bra{+}\rho\ket{+} of each output qubit's reduced state ρ\rho with ∣+⟩\ket+?

Show the answer

0.5

The output is Φ+\Phi^+, and each half of a Bell pair is maximally mixed: ρ=Tr⁡B∣Φ+⟩ ⁣⟨Φ+∣=I/2\rho=\Tr_B\dyad{\Phi^+}=\id/2, so ⟨+∣ρ∣+⟩=12\bra+\rho\ket+=\tfrac12. Averaged over all pure inputs, the CNOT copier's single-copy fidelity is the average of ∣α∣4+∣β∣4\abs\alpha^4+\abs\beta^4, which is 23\tfrac23. The best possible universal copier reaches 56\tfrac56, as the Rydberg State lesson on optimal cloning shows.

Proof 2: inner products

Proof 1 assumed the copier works on a whole basis. A stronger statement needs only two states and allows any machine. The tool is the inner product, which every unitary preserves.

DerivationNo-cloning from inner products
  1. U∣ψ⟩∣e⟩∣m⟩=∣ψ⟩∣ψ⟩∣mψ⟩,U∣ϕ⟩∣e⟩∣m⟩=∣ϕ⟩∣ϕ⟩∣mϕ⟩U\ket\psi\ket e\ket m=\ket\psi\ket\psi\ket{m_\psi},\qquad U\ket\phi\ket e\ket m=\ket\phi\ket\phi\ket{m_\phi}

    Suppose UU clones two states ∣ψ⟩\ket\psi and ∣ϕ⟩\ket\phi.

  2. ⟨ϕ∣ψ⟩ ⟨e∣e⟩ ⟨m∣m⟩=⟨ϕ∣ψ⟩2 ⟨mϕ∣mψ⟩\braket{\phi|\psi}\,\braket{e|e}\,\braket{m|m}=\braket{\phi|\psi}^2\,\braket{m_\phi|m_\psi}

    Take the inner product of the two equations. The left-hand side uses U†U=IU^\dagger U=\id.

  3. s=s2 ∣⟨mϕ∣mψ⟩∣≤s2,s≡∣⟨ϕ∣ψ⟩∣∈[0,1]s=s^2\,\abs{\braket{m_\phi|m_\psi}}\le s^2,\qquad s\equiv\abs{\braket{\phi|\psi}}\in[0,1]

    Take absolute values. By Cauchy–Schwarz the machine overlap is at most 1.

  4. s≤s2 on [0,1] ⟹ s∈{0,1}s\le s^2\ \text{on }[0,1]\ \Longrightarrow\ s\in\{0,1\}

    So the two states are orthogonal or identical up to phase. Conversely, a pairwise-orthogonal set extends to an orthonormal basis, which can be cloned. ■\blacksquare

Check your understanding

Which of these assumptions does the inner-product proof not need?

  1. That the copier preserves inner products (is unitary on input, blank and machine)
  2. That it copies both ∣ψ⟩\ket\psi and ∣ϕ⟩\ket\phi
  3. That the blank and the machine start in the same states for both inputs
  4. That the machine ends in the same state for both inputs
Show the answer

D. That the machine ends in the same state for both inputs

The final machine states ∣mψ⟩\ket{m_\psi}, ∣mϕ⟩\ket{m_\phi} may differ: they enter only through ∣⟨mϕ∣mψ⟩∣≤1\abs{\braket{m_\phi|m_\psi}}\le1, which is all the proof uses.

No-signalling

Now the other half of the story. Alice and Bob share a state ρAB\rho_{AB}, possibly entangled. Alice does anything to her system: a gate, a measurement, a noisy process. Can Bob, looking only at his own system, tell what she did?

Everything Bob can observe locally is fixed by his reduced density matrix ρB=Tr⁡AρAB\rho_B=\Tr_A\rho_{AB}: any measurement he makes, with POVM elements EkE_k, has outcome probabilities Tr⁡(EkρB)\Tr(E_k\rho_B). So the question is whether Alice can change ρB\rho_B.

DerivationThe no-signalling theorem
  1. ρAB ↦ ∑k(Kk⊗I) ρAB (Kk⊗I)†,∑kKk†Kk=I\rho_{AB}\ \mapsto\ \sum_k (K_k\otimes\id)\,\rho_{AB}\,(K_k\otimes\id)^\dagger,\qquad \sum_k K_k^\dagger K_k=\id

    Any operation by Alice, including a measurement whose outcome Bob does not know, is described by Kraus operators KkK_k acting on AA alone.

  2. ρB′=∑kTr⁡A ⁣[(Kk⊗I) ρAB (Kk†⊗I)]=Tr⁡A ⁣[(∑kKk†Kk⊗I) ρAB]\rho_B'=\sum_k\Tr_A\!\big[(K_k\otimes\id)\,\rho_{AB}\,(K_k^\dagger\otimes\id)\big]=\Tr_A\!\Big[\big(\textstyle\sum_k K_k^\dagger K_k\otimes\id\big)\,\rho_{AB}\Big]

    The partial trace over AA is cyclic for operators that act on AA only.

  3. ρB′=Tr⁡A ρAB=ρB\rho_B'=\Tr_A\,\rho_{AB}=\rho_B

    Completeness. Bob's state is untouched, whatever Alice chose to do. ■\blacksquare

A concrete case: for Φ+\Phi^+, if Alice measures in ZZ, Bob's qubit becomes ∣0⟩\ket0 or ∣1⟩\ket1 with probability ½ each; if she measures in XX, it becomes ∣+⟩\ket+ or ∣−⟩\ket-. Two different ensembles, one density matrix:

12∣0⟩ ⁣⟨0∣+12∣1⟩ ⁣⟨1∣  =  12∣+⟩ ⁣⟨+∣+12∣−⟩ ⁣⟨−∣  =  I2.\tfrac12\dyad0+\tfrac12\dyad1\;=\;\tfrac12\dyad{+}+\tfrac12\dyad{-}\;=\;\tfrac{\id}{2}.

Cloning would break causality

This is exactly where a cloner would break things. If Bob could turn his qubit into NN perfect copies, Alice's two choices would leave him with

ρZ(N)=12(∣0⟩ ⁣⟨0∣⊗N+∣1⟩ ⁣⟨1∣⊗N)vs.ρX(N)=12(∣+⟩ ⁣⟨+∣⊗N+∣−⟩ ⁣⟨−∣⊗N).\rho_Z^{(N)}=\tfrac12\big(\dyad{0}^{\otimes N}+\dyad{1}^{\otimes N}\big)\quad\text{vs.}\quad\rho_X^{(N)}=\tfrac12\big(\dyad{+}^{\otimes N}+\dyad{-}^{\otimes N}\big).

For N≥2N\ge2 these differ. Measuring all copies in ZZ gives all-equal outcomes for ρZ(N)\rho_Z^{(N)}, but for ρX(N)\rho_X^{(N)} all-equal outcomes occur only with probability 21−N2^{1-N}. Bob would learn Alice's choice instantly.

TaskPossible?Notes
Clone an unknown pure stateNoNo-cloning (1982)
Clone states promised to come from an orthogonal setYesMeasure and re-prepare, or a CNOT-style unitary
Approximately clone every qubit stateYesBest symmetric 1→21\to2 fidelity is 5/65/6, reached by the Bužek–Hillery cloner (1996)
Clone exactly but probabilistically, heralding successOnly for linearly independent setsProbabilistic cloning (Duan–Guo, 1998)
Broadcast non-commuting mixed statesNoNo-broadcasting (Barnum, Caves, Fuchs, Jozsa & Schumacher, 1996)
Delete one of two copies of an unknown stateNoNo-deleting (Pati & Braunstein, 2000)
Signal using entanglement aloneNoNo-signalling

Key takeaways

  • A unitary clones a set of pure states iff they are pairwise orthogonal or equal up to phase. Linearity and inner-product preservation each prove it.
  • A basis copier entangles instead of copying: α∣0⟩+β∣1⟩↦α∣00⟩+β∣11⟩\alpha\ket0+\beta\ket1\mapsto\alpha\ket{00}+\beta\ket{11}.
  • No-signalling: Tr⁡A[(E⊗id)ρAB]=ρB\Tr_A[(\mathcal E\otimes\mathrm{id})\rho_{AB}]=\rho_B for every local channel E\mathcal E, so Alice's choices are invisible to Bob alone.
  • A perfect cloner would distinguish ensembles with the same density matrix, which is superluminal signalling. The two no-go theorems protect each other.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

Which set of states can a single unitary (with blank and machine) clone perfectly?

  1. {∣0⟩,∣+⟩}\{\ket0,\ket+\}
  2. All states on the equator of the Bloch sphere
  3. {∣0⟩,∣1⟩,∣+⟩}\{\ket0,\ket1,\ket+\}
  4. {∣+i⟩,∣−i⟩}\{\ket{+i},\ket{-i}\}
Show the answer

D. {∣+i⟩,∣−i⟩}\{\ket{+i},\ket{-i}\}

∣+i⟩\ket{+i} and ∣−i⟩\ket{-i} are orthogonal, so a unitary (rotate them to ∣0⟩,∣1⟩\ket0,\ket1, CNOT, rotate back) clones them. Every other option contains a pair whose overlap is strictly between 0 and 1.

Question 2

Alice and Bob share Φ+\Phi^+. Without telling Bob, Alice either does nothing, applies XX, or measures in the XX basis. Bob's reduced density matrix in the three cases is:

  1. I/2\id/2 in all three cases
  2. I/2\id/2, I/2\id/2, and then ∣+⟩ ⁣⟨+∣\dyad{+} or ∣−⟩ ⁣⟨−∣\dyad{-}
  3. ∣0⟩ ⁣⟨0∣\dyad0, ∣1⟩ ⁣⟨1∣\dyad1, and I/2\id/2
  4. It depends on how far apart they are
Show the answer

A. I/2\id/2 in all three cases

Each is a local operation on AA, so ρB=Tr⁡AρAB=I/2\rho_B=\Tr_A\rho_{AB}=\id/2 is unchanged. After an XX measurement Bob's qubit is ∣+⟩\ket+ or ∣−⟩\ket-, but since he does not know which, his description is the equal mixture, again I/2\id/2.

Question 3

Why can't a large machine register rescue a cloner of ∣0⟩\ket0 and ∣+⟩\ket+?

  1. Large machines decohere too quickly
  2. The machine would need infinitely many qubits
  3. The machine overlap ∣⟨mϕ∣mψ⟩∣≤1\abs{\braket{m_\phi|m_\psi}}\le1 only shrinks the right side of s=s2∣⟨mϕ∣mψ⟩∣s=s^2\abs{\braket{m_\phi|m_\psi}}, which with s=1/2s=1/\sqrt2 is already too small
  4. The machine would become entangled with Bob
Show the answer

C. The machine overlap ∣⟨mϕ∣mψ⟩∣≤1\abs{\braket{m_\phi|m_\psi}}\le1 only shrinks the right side of s=s2∣⟨mϕ∣mψ⟩∣s=s^2\abs{\braket{m_\phi|m_\psi}}, which with s=1/2s=1/\sqrt2 is already too small

With s=1/2s=1/\sqrt2, cloning would need 1/2=12∣⟨mϕ∣mψ⟩∣≤121/\sqrt2=\tfrac12\abs{\braket{m_\phi|m_\psi}}\le\tfrac12: impossible for any machine of any size.

Question 4

Why would a perfect cloner enable faster-than-light signalling?

  1. Copies of a qubit would travel faster than the original
  2. Bob could copy his half of an entangled pair many times and distinguish two ensembles that share the same density matrix
  3. Cloning creates entanglement, and entanglement acts faster than light
  4. Alice could clone her classical bits before sending them
Show the answer

B. Bob could copy his half of an entangled pair many times and distinguish two ensembles that share the same density matrix

Alice's basis choice prepares {∣0⟩,∣1⟩}\{\ket0,\ket1\} or {∣±⟩}\{\ket\pm\} at Bob's end, both with ρB=I/2\rho_B=\id/2. Many copies would make the two ensembles distinguishable (ρZ(N)≠ρX(N)\rho_Z^{(N)}\neq\rho_X^{(N)}), so Bob would learn her choice with no classical message at all.

Question 5

Which of these does quantum mechanics allow?

  1. Copying any unknown qubit exactly, with a heralded success probability of ½
  2. Deleting one of two copies of an unknown state while leaving the other intact
  3. Using shared entanglement alone to send one bit
  4. Producing two imperfect copies of an unknown qubit, each with fidelity 5/65/6
Show the answer

D. Producing two imperfect copies of an unknown qubit, each with fidelity 5/65/6

Approximate cloning is allowed: the optimal universal cloner gives each copy fidelity 5/65/6. Exact probabilistic cloning works only for linearly independent sets, deleting is ruled out by the no-deleting theorem, and signalling by no-signalling.

Go deeper

Further reading

  • W. K. Wootters & W. H. Zurek, "A single quantum cannot be cloned", Nature 299, 802 (1982).
  • D. Dieks, "Communication by EPR devices", Physics Letters A 92, 271 (1982).
  • M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, §1.3.5 and Box 12.1 (Cambridge, 2010).
  • V. Scarani, S. Iblisdir, N. Gisin & A. Acín, "Quantum cloning", Reviews of Modern Physics 77, 1225 (2005).