Simple Algorithm
Quantum Computing · Excited State · 22 min

Bell States & Creating Entanglement

The four Bell states, and the circuits that make them.

Four special states form a basis of two qubits, and every one of them is maximally entangled. They are the workhorses of quantum information: teleportation, superdense coding, entanglement swapping and Bell tests all start from one. Here is how to make them, recognize them and undo them.

What you’ll learn

  • Write the four Bell states, show they form an orthonormal basis, and derive CNOT (H⊗I)∣xy⟩=∣βxy⟩\mathrm{CNOT}\,(H\otimes I)\ket{xy} = \ket{\beta_{xy}}.
  • Identify Bell states by their XXXX and ZZZZ eigenvalues and compute their correlation functions.
  • Use the transpose trick (A⊗I)∣Φ+⟩=(I⊗AT)∣Φ+⟩(A\otimes I)\ket{\Phi^+} = (I\otimes A^T)\ket{\Phi^+} and the singlet's U⊗UU\otimes U invariance.
  • Explain why local operations cannot create entanglement, and which two-qubit gates can.
  • Build a GHZ state and contrast its correlations and fragility with those of a Bell pair.

The Bell basis

∣Φ±⟩=12(∣00⟩±∣11⟩),∣Ψ±⟩=12(∣01⟩±∣10⟩)\ket{\Phi^\pm} = \tfrac{1}{\sqrt2}\big(\ket{00}\pm\ket{11}\big),\qquad \ket{\Psi^\pm} = \tfrac{1}{\sqrt2}\big(\ket{01}\pm\ket{10}\big)
The four Bell states. Any two of them are orthogonal: they either share no basis kets, or share both with a relative sign that cancels the overlap. So they form an orthonormal basis of C2⊗C2\C^2\otimes\C^2.

Compactly, ∣βxy⟩=12(∣0,y⟩+(−1)x∣1,yˉ⟩)\ket{\beta_{xy}} = \tfrac{1}{\sqrt2}\big(\ket{0,y}+(-1)^x\ket{1,\bar y}\big), so β00=Φ+\beta_{00} = \Phi^+, β01=Ψ+\beta_{01} = \Psi^+, β10=Φ−\beta_{10} = \Phi^- and β11=Ψ−\beta_{11} = \Psi^-. Each has a coefficient matrix with ∣det⁡C∣=12\abs{\det C} = \tfrac12, hence concurrence 1: all four are maximally entangled. A single circuit makes all of them:

DerivationPreparing the Bell states
  1. ∣x⟩∣y⟩ → H⊗I  12(∣0⟩+(−1)x∣1⟩)∣y⟩\ket{x}\ket{y}\ \xrightarrow{\ H\otimes I\ }\ \tfrac{1}{\sqrt2}\big(\ket0+(-1)^x\ket1\big)\ket{y}

    H∣x⟩=12(∣0⟩+(−1)x∣1⟩)H\ket x = \tfrac{1}{\sqrt2}\big(\ket0+(-1)^x\ket1\big): the first input bit becomes a relative sign.

  2. → CNOT  12(∣0⟩∣y⟩+(−1)x∣1⟩∣y⊕1⟩)=∣βxy⟩\xrightarrow{\ \mathrm{CNOT}\ }\ \tfrac{1}{\sqrt2}\big(\ket{0}\ket{y}+(-1)^x\ket{1}\ket{y\oplus1}\big) = \ket{\beta_{xy}}

    The CNOT flips the target in the ∣1⟩\ket1 branch: the second bit becomes the parity, whether the two qubits agree or disagree.

  3. (H⊗I) CNOT ∣βxy⟩=∣xy⟩(H\otimes I)\,\mathrm{CNOT}\,\ket{\beta_{xy}} = \ket{xy}

    The circuit is unitary and maps the computational basis onto the Bell states, which proves again that they are orthonormal. Running it backwards (CNOT, then HH, then a computational-basis measurement) is a Bell measurement: it reads out xx and yy.

Check your understanding

Starting from ∣Φ+⟩\ket{\Phi^+}, you apply ZZ to qubit 1 instead of qubit 0. Which state do you get?

  1. ∣Φ+⟩\ket{\Phi^+}
  2. ∣Ψ−⟩\ket{\Psi^-}
  3. ∣Ψ+⟩\ket{\Psi^+}
  4. ∣Φ−⟩\ket{\Phi^-}
Show the answer

D. ∣Φ−⟩\ket{\Phi^-}

(I⊗Z)∣Φ+⟩=12(∣00⟩−∣11⟩)=∣Φ−⟩(I\otimes Z)\ket{\Phi^+} = \tfrac{1}{\sqrt2}(\ket{00}-\ket{11}) = \ket{\Phi^-}, the same as ZZ on qubit 0. In general (A⊗I)∣Φ+⟩=(I⊗AT)∣Φ+⟩(A\otimes I)\ket{\Phi^+} = (I\otimes A^T)\ket{\Phi^+}, and ZT=ZZ^T = Z.

Fingerprints: stabilizers and correlations

Each Bell state is a joint eigenstate of the commuting observables X⊗XX\otimes X and Z⊗ZZ\otimes Z, and the pair of eigenvalues identifies it uniquely: these two operators are the stabilizers of the Bell basis. Because Y⊗Y=−(X⊗X)(Z⊗Z)Y\otimes Y = -(X\otimes X)(Z\otimes Z), the YYYY eigenvalue is then fixed too.

StateZZZZXXXXYYYYResults along a common axis
∣Φ+⟩\ket{\Phi^+}+1+1+1+1−1-1equal for zz and xx, opposite for yy
∣Φ−⟩\ket{\Phi^-}+1+1−1-1+1+1equal for zz and yy, opposite for xx
∣Ψ+⟩\ket{\Psi^+}−1-1+1+1+1+1opposite for zz, equal for xx and yy
∣Ψ−⟩\ket{\Psi^-}−1-1−1-1−1-1opposite along every axis

Measuring along unit vectors a⃗\vec a (Alice) and b⃗\vec b (Bob), the correlation is E(a⃗,b⃗)=⟨β∣(a⃗⋅σ⃗)⊗(b⃗⋅σ⃗)∣β⟩E(\vec a,\vec b) = \bra{\beta}(\vec a\cdot\vec\sigma)\otimes(\vec b\cdot\vec\sigma)\ket{\beta}. Two identities turn such computations into one-liners.

DerivationThe transpose trick and the singlet's symmetry
  1. (A⊗I)∣Φ+⟩=12∑i,jAji∣j⟩∣i⟩=(I⊗AT)∣Φ+⟩(A\otimes I)\ket{\Phi^+} = \tfrac{1}{\sqrt2}\sum_{i,j}A_{ji}\ket{j}\ket{i} = (I\otimes A^{T})\ket{\Phi^+}

    Expand ∣Φ+⟩=12∑i∣ii⟩\ket{\Phi^+} = \tfrac{1}{\sqrt2}\sum_i\ket{ii} and relabel the sum: an operator on one half can be moved to the other half as its transpose.

  2. (U⊗U‾)∣Φ+⟩=(I⊗U‾UT)∣Φ+⟩=∣Φ+⟩(U\otimes\overline{U})\ket{\Phi^+} = (I\otimes\overline{U}U^{T})\ket{\Phi^+} = \ket{\Phi^+}

    For unitary UU, U‾UT=UU†‾=I\overline{U}U^{T} = \overline{UU^\dagger} = I, so ∣Φ+⟩\ket{\Phi^+} is invariant under "opposite" rotations U⊗U‾U\otimes\overline U:

  3. (U⊗U)∣Ψ−⟩=(I⊗U XZ UT)∣Φ+⟩=det⁡(U) ∣Ψ−⟩(U\otimes U)\ket{\Psi^-} = (I\otimes U\,XZ\,U^{T})\ket{\Phi^+} = \det(U)\,\ket{\Psi^-}

    Write the singlet as ∣Ψ−⟩=(I⊗XZ)∣Φ+⟩\ket{\Psi^-} = (I\otimes XZ)\ket{\Phi^+}. For every 2×22\times2 matrix MM, M(XZ)MT=det⁡(M) XZM(XZ)M^{T} = \det(M)\,XZ, because XZ=(0−110)XZ = \begin{pmatrix}0&-1\\1&0\end{pmatrix} is antisymmetric. So the singlet is invariant, up to a phase, under identical rotations of both qubits:

  4. EΨ−(a⃗,b⃗)=− a⃗⋅b⃗,EΦ+(a⃗,b⃗)=axbx−ayby+azbzE_{\Psi^-}(\vec a,\vec b) = -\,\vec a\cdot\vec b,\qquad E_{\Phi^+}(\vec a,\vec b) = a_xb_x-a_yb_y+a_zb_z

    Rotational invariance forces the singlet's (bilinear) correlation to be a multiple of a⃗⋅b⃗\vec a\cdot\vec b, and perfect anticorrelation at a⃗=b⃗\vec a = \vec b fixes the multiple. For Φ+\Phi^+, read it off the table (⟨XX⟩=⟨ZZ⟩=1\expval{XX} = \expval{ZZ} = 1, ⟨YY⟩=−1\expval{YY} = -1, and mixed terms such as ⟨XZ⟩\expval{XZ} vanish). In the xx–zz plane these become ∓cos⁡(θA−θB)\mp\cos(\theta_A-\theta_B).

Predict before you look

Alice and Bob share ∣Ψ−⟩\ket{\Psi^-} and measure at Bloch angles 0∘0^\circ and 120∘120^\circ. What percentage of pairs give the same result?

Show the answer

75 %

E=−cos⁡120∘=+12E = -\cos120^\circ = +\tfrac12, and P(same)=1+E2=75%P(\text{same}) = \tfrac{1+E}{2} = 75\%. At 120∘120^\circ the singlet's anticorrelation has turned into a majority of agreements.

What can create entanglement?

Entanglement therefore needs an interaction: a two-qubit gate that is not of the form A⊗BA\otimes B. CNOT and CZ are maximally entangling: on suitable product inputs they output Bell states. SWAP is a two-qubit gate that never entangles, since SWAP(∣a⟩⊗∣b⟩)=∣b⟩⊗∣a⟩\mathrm{SWAP}(\ket a\otimes\ket b) = \ket b\otimes\ket a, while its square root SWAP\sqrt{\mathrm{SWAP}} is maximally entangling. Adding classical communication does not help: LOCC can consume or redistribute entanglement but never create it. That is the starting point of Entanglement as a Resource (LOCC).

PlatformHow entanglement is typically made
Photonsspontaneous parametric down-conversion in a nonlinear crystal emits polarization-entangled pairs
Trapped ionslaser-driven gates, such as the Mølmer–Sørensen gate, that couple the ions through shared motional modes
Superconducting qubitsmicrowave-activated or tunable-coupler gates of CZ, iSWAP or cross-resonance type
Neutral atomsRydberg blockade: exciting one atom shifts its neighbor out of resonance, giving a CZ gate
A qualitative overview; the hardware topic covers performance.
Check your understanding

Which two-qubit gate can never turn a product state into an entangled state?

  1. CNOT
  2. SWAP\sqrt{\mathrm{SWAP}}
  3. SWAP
  4. CZ
Show the answer

C. SWAP

SWAP maps ∣a⟩⊗∣b⟩\ket a\otimes\ket b to ∣b⟩⊗∣a⟩\ket b\otimes\ket a, another product state. CNOT and CZ produce maximally entangled states from ∣+⟩∣0⟩\ket{+}\ket{0} and ∣+⟩∣+⟩\ket+\ket+, and SWAP\sqrt{\mathrm{SWAP}} maps ∣01⟩\ket{01} to a maximally entangled state.

Three qubits: the GHZ state

One more CNOT spreads the entanglement to a third qubit:

∣GHZ⟩=12(∣000⟩+∣111⟩).\ket{\mathrm{GHZ}} = \tfrac{1}{\sqrt2}\big(\ket{000}+\ket{111}\big).

All three qubits agree in the ZZ basis, and measuring any one of them in ZZ collapses the other two. But GHZ entanglement is fragile: tracing out one qubit leaves 12(∣00⟩ ⁣⟨00∣+∣11⟩ ⁣⟨11∣)\tfrac12\big(\dyad{00}+\dyad{11}\big), a classically correlated mixture with no entanglement at all.

Think first Is there a three-qubit state that, unlike GHZ, stays entangled when one qubit is lost?

Yes: the W state 13(∣001⟩+∣010⟩+∣100⟩)\tfrac{1}{\sqrt3}(\ket{001}+\ket{010}+\ket{100}). Tracing out any one qubit leaves an entangled two-qubit state, with concurrence 23\tfrac23. GHZ and W are so different that no local operations, not even probabilistic ones, convert one into the other.

Key takeaways

  • The Bell states ∣βxy⟩=CNOT (H⊗I)∣xy⟩\ket{\beta_{xy}} = \mathrm{CNOT}\,(H\otimes I)\ket{xy} form an orthonormal basis of maximally entangled states; reversing the circuit is a Bell measurement.
  • Each Bell state is fixed by its XXXX and ZZZZ eigenvalues, and Paulis on a single qubit cycle through all four.
  • (A⊗I)∣Φ+⟩=(I⊗AT)∣Φ+⟩(A\otimes I)\ket{\Phi^+} = (I\otimes A^T)\ket{\Phi^+} and (U⊗U)∣Ψ−⟩=det⁡U ∣Ψ−⟩(U\otimes U)\ket{\Psi^-} = \det U\,\ket{\Psi^-} give the correlations, e.g. EΨ−=−a⃗⋅b⃗E_{\Psi^-} = -\vec a\cdot\vec b.
  • Local operations preserve Schmidt rank: creating entanglement requires an interaction such as CNOT or CZ (but never SWAP).
  • GHZ extends the Bell pair to three qubits, with all-or-nothing correlations but complete fragility under the loss of one qubit.

Check yourself

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

Question 1

Which Bell state does CNOT (H⊗I)\mathrm{CNOT}\,(H\otimes I) produce from ∣11⟩\ket{11}?

  1. ∣Φ−⟩\ket{\Phi^-}
  2. ∣Ψ+⟩\ket{\Psi^+}
  3. ∣Ψ−⟩\ket{\Psi^-}
  4. ∣Φ+⟩\ket{\Phi^+}
Show the answer

C. ∣Ψ−⟩\ket{\Psi^-}

H∣1⟩=∣−⟩H\ket1 = \ket-, so 12(∣0⟩−∣1⟩)∣1⟩↦12(∣01⟩−∣10⟩)=∣Ψ−⟩\tfrac{1}{\sqrt2}(\ket0-\ket1)\ket1\mapsto\tfrac{1}{\sqrt2}(\ket{01}-\ket{10}) = \ket{\Psi^-}: x=1x = 1 gives the minus sign and y=1y = 1 the anticorrelation.

Question 2

Alice and Bob share ∣Ψ−⟩\ket{\Psi^-} and both measure along the same, arbitrary axis n⃗\vec n. Their results are:

  1. always opposite
  2. always equal
  3. opposite along zz only, unrelated otherwise
  4. uncorrelated
Show the answer

A. always opposite

(U⊗U)∣Ψ−⟩=det⁡U ∣Ψ−⟩(U\otimes U)\ket{\Psi^-} = \det U\,\ket{\Psi^-}: rotating both measurement axes together changes nothing, so the perfect zz-anticorrelation holds along every axis, E=−n⃗⋅n⃗=−1E = -\vec n\cdot\vec n = -1.

Question 3

A two-qubit state satisfies (X⊗X)∣ψ⟩=∣ψ⟩(X\otimes X)\ket\psi = \ket\psi and (Z⊗Z)∣ψ⟩=−∣ψ⟩(Z\otimes Z)\ket\psi = -\ket\psi. Which state is it?

  1. ∣Φ+⟩\ket{\Phi^+}
  2. ∣Φ−⟩\ket{\Phi^-}
  3. ∣Ψ−⟩\ket{\Psi^-}
  4. ∣Ψ+⟩\ket{\Psi^+}
Show the answer

D. ∣Ψ+⟩\ket{\Psi^+}

ZZ=−1ZZ = -1 means the bits disagree (the Ψ\Psi family), and XX=+1XX = +1 picks the plus sign: ∣Ψ+⟩=12(∣01⟩+∣10⟩)\ket{\Psi^+} = \tfrac{1}{\sqrt2}(\ket{01}+\ket{10}).

Question 4

One qubit of a GHZ state is lost (traced out). The remaining two qubits are:

  1. in a Bell state
  2. in the separable mixture 12(∣00⟩ ⁣⟨00∣+∣11⟩ ⁣⟨11∣)\tfrac12(\dyad{00}+\dyad{11})
  3. maximally mixed, I/4I/4
  4. in the pure product state ∣+⟩∣+⟩\ket{+}\ket{+}
Show the answer

B. in the separable mixture 12(∣00⟩ ⁣⟨00∣+∣11⟩ ⁣⟨11∣)\tfrac12(\dyad{00}+\dyad{11})

Tr⁡3∣GHZ⟩ ⁣⟨GHZ∣=12(∣00⟩ ⁣⟨00∣+∣11⟩ ⁣⟨11∣)\Tr_3\dyad{\mathrm{GHZ}} = \tfrac12(\dyad{00}+\dyad{11}): perfectly correlated in ZZ, but a mixture of product states, hence not entangled. The coherence between ∣000⟩\ket{000} and ∣111⟩\ket{111} needs all three qubits.

Go deeper

Further reading

  • M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, §1.3.6 (Bell states) and §2.3 (superdense coding) (Cambridge, 2010).
  • J. S. Bell, "On the Einstein Podolsky Rosen paradox", Physics 1, 195 (1964).
  • D. M. Greenberger, M. A. Horne, A. Shimony & A. Zeilinger, "Bell's theorem without inequalities", Am. J. Phys. 58, 1131 (1990).
  • N. D. Mermin, "Quantum mysteries revisited", Am. J. Phys. 58, 731 (1990).