Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Which-Path Information & the Quantum Eraser

Complementarity, entanglement and restoring fringes.

Mark which arm a photon took and the fringe vanishes. Measure the mark in a different basis and the fringe comes back, but only after you sort your data. The quantum eraser shows that complementarity is about information and entanglement, not about clumsy disturbance.

What you’ll learn

  • State and prove Englert's duality relation D2+V2≤1D^2 + V^2 \le 1, with equality for pure markers.
  • Derive the conditional fringe and anti-fringe produced by measuring a marker in a conjugate basis.
  • Prove that no operation on the marker changes the unconditional fringe, and use it to dissolve delayed-choice paradoxes.
  • Model decoherence as which-path information spread over many environmental systems.

Wave and particle knowledge, quantified

Let BS₁ put the photon in the two arms with probabilities w0,w1w_0, w_1, and let a marker end in state ρ0\rho_0 or ρ1\rho_1 depending on the arm. Three numbers describe what you can know:

  • Predictability P=∣w0−w1∣P = \abs{w_0 - w_1}: path knowledge available beforehand.
  • Visibility V=2∣ρ01path∣V = 2\abs{\rho^{\text{path}}_{01}}: the fringe contrast, set by the path qubit's coherence.
  • Distinguishability D=∥w0ρ0−w1ρ1∥1D = \norm{w_0\rho_0 - w_1\rho_1}_1: the bias of the best guess of the path from the marker, Pguess=12(1+D)P_{\text{guess}} = \tfrac12(1 + D).
DerivationProof
  1. ∣Ψ⟩=w0 ∣0⟩∣e0⟩+w1 ∣1⟩∣e1⟩,V=2w0w1 ∣γ∣,  γ=⟨e0∣e1⟩\ket{\Psi} = \sqrt{w_0}\,\ket{0}\ket{e_0} + \sqrt{w_1}\,\ket{1}\ket{e_1}, \qquad V = 2\sqrt{w_0w_1}\,\abs{\gamma},\ \ \gamma = \braket{e_0|e_1}

    Pure case. The off-diagonal element of the path's reduced state is w0w1⟨e1∣e0⟩\sqrt{w_0w_1}\braket{e_1|e_0}.

  2. A=w0∣e0⟩ ⁣⟨e0∣−w1∣e1⟩ ⁣⟨e1∣:Tr⁡A=w0−w1,det⁡A=−w0w1(1−∣γ∣2)A = w_0\dyad{e_0} - w_1\dyad{e_1}: \quad \Tr A = w_0 - w_1, \quad \det A = -w_0w_1\left(1 - \abs{\gamma}^2\right)

    Restrict to span⁡{∣e0⟩,∣e1⟩}\operatorname{span}\{\ket{e_0}, \ket{e_1}\}. The eigenvalues have opposite signs.

  3. D=∥A∥1=λ+−λ−=(Tr⁡A)2−4det⁡A=1−4w0w1∣γ∣2D = \norm{A}_1 = \lambda_+ - \lambda_- = \sqrt{(\Tr A)^2 - 4\det A} = \sqrt{1 - 4w_0w_1\abs{\gamma}^2}

    Use (w0−w1)2+4w0w1=1(w_0 - w_1)^2 + 4w_0w_1 = 1.

  4. D2+V2=1−4w0w1∣γ∣2+4w0w1∣γ∣2=1D^2 + V^2 = 1 - 4w_0w_1\abs{\gamma}^2 + 4w_0w_1\abs{\gamma}^2 = 1

    Equality for pure markers.

  5. Mixed markers: purify them by including the environment, ∣E0⟩,∣E1⟩\ket{E_0}, \ket{E_1}. VV is unchanged (it depends only on the path's reduced state), while the trace norm can only shrink under the partial trace over the environment: D≤DpureD \le D_{\text{pure}}. Hence D2+V2≤1D^2 + V^2 \le 1. Finally P=∣Tr⁡A∣≤∥A∥1=DP = \abs{\Tr A} \le \norm{A}_1 = D. ■\blacksquare

Check your understanding

BS₁ is unbalanced, with path weights 0.80.8 and 0.20.2, and there is no marker at all. What are VV and DD?

  1. V=1V = 1, D=0D = 0
  2. V=0.8V = 0.8, D=0.6D = 0.6
  3. V=0.6V = 0.6, D=0.8D = 0.8
  4. V=0.64V = 0.64, D=0.36D = 0.36
Show the answer

B. V=0.8V = 0.8, D=0.6D = 0.6

With identical markers (γ=1\gamma = 1), V=20.8⋅0.2=0.8V = 2\sqrt{0.8\cdot0.2} = 0.8 and D=1−4⋅0.16=0.6=∣0.8−0.2∣=PD = \sqrt{1 - 4\cdot 0.16} = 0.6 = \abs{0.8 - 0.2} = P. All path knowledge is prior knowledge, and D2+V2=1D^2 + V^2 = 1.

Erasing, and sorting

Take a perfect marker, ∣e0⟩=∣0⟩\ket{e_0} = \ket0 and ∣e1⟩=∣1⟩\ket{e_1} = \ket1 (a CNOT from path to marker), so V=0V = 0. Scully and Drühl (1982) noticed that measuring the marker in the conjugate basis ∣±⟩\ket{\pm} erases the which-path information. Erasure does not restore the fringe in the raw data. It splits the data into two subsets with complementary fringes.

DerivationConditional fringes
  1. 12∣0⟩(∣0⟩+eiφ∣1⟩)+12∣1⟩(∣0⟩−eiφ∣1⟩)\tfrac12\ket{0}\big(\ket{0} + e^{i\varphi}\ket{1}\big) + \tfrac12\ket{1}\big(\ket{0} - e^{i\varphi}\ket{1}\big)

    Path ⊗\otimes marker, after BS₁, the CNOT, the phase and BS₂.

  2. ⟨±∣(∣0⟩+eiφ∣1⟩)=12(1±eiφ)\bra{\pm}\big(\ket0 + e^{i\varphi}\ket1\big) = \tfrac{1}{\sqrt2}\big(1 \pm e^{i\varphi}\big)

    Project the marker onto ∣±⟩\ket{\pm}.

  3. P(D0,+)=12cos⁡2φ2,P(D0,−)=12sin⁡2φ2P(\text{D0}, +) = \tfrac12\cos^2\tfrac{\varphi}{2}, \qquad P(\text{D0}, -) = \tfrac12\sin^2\tfrac{\varphi}{2}

    Joint probabilities.

  4. P(D0∣+)=cos⁡2φ2,P(D0∣−)=sin⁡2φ2,P(D0)=12P(\text{D0}\mid +) = \cos^2\tfrac{\varphi}{2}, \quad P(\text{D0}\mid -) = \sin^2\tfrac{\varphi}{2}, \quad P(\text{D0}) = \tfrac12

    A fringe and an anti-fringe, each of full visibility, summing to the flat distribution.

Delayed choice without retrocausality

In the delayed-choice eraser of Kim, Yu, Kulik, Shih and Scully (2000), entangled photon pairs are created; the signal photon is detected first, and its idler partner, which carries the which-path information, is routed afterward at random either to detectors that reveal the path or through a beam splitter that erases it. Fringes appear only in the signal counts sorted by coincidence with the erasing detectors. Later experiments made the choice space-like separated from the signal detection (Ma et al., 2013).

Nothing here needs influence on the past. Measurements on different subsystems commute, so the joint distribution of signal and idler outcomes is the same whichever is measured first. The late choice only decides which conditional distributions you can form from data that was already fixed, and the theorem above guarantees the signal's own statistics never depend on it.

Check your understanding

In a delayed-choice eraser the idlers are measured a year after the signal photons hit the screen. By looking at the screen alone, can you tell in advance which basis will be chosen for the idlers?

  1. Yes: the screen shows fringes if the idlers will be erased
  2. Yes, but only statistically, with many photons
  3. No: the screen always shows the flat sum of fringe and anti-fringe, whatever is chosen later
  4. No, because the screen pattern is destroyed by the later measurement
Show the answer

C. No: the screen always shows the flat sum of fringe and anti-fringe, whatever is chosen later

The signal's reduced state is independent of any operation on the idler (no-signalling), so the screen shows the same featureless pattern in every case. Fringes exist only in subsets selected using the idler results.

Decoherence: which-path information you cannot collect

A real environment learns the path through many weak interactions: scattered gas molecules, thermal photons, phonons. If the kk-th environmental particle ends in ∣e0(k)⟩\ket{e^{(k)}_0} or ∣e1(k)⟩\ket{e^{(k)}_1}, the records multiply:

V=∏k=1K∣⟨e0(k)∣e1(k)⟩∣.V = \prod_{k=1}^{K} \abs{\braket{e^{(k)}_0|e^{(k)}_1}}.

Even if every single scattering event is almost uninformative, the visibility decays exponentially in their number. Collisional decoherence of fullerene matter waves shows exactly this exponential loss of contrast with background gas pressure (Hornberger et al., 2003).

In principle the eraser still works, but it would require measuring every environmental particle in a conjugate basis. That is why decoherence is effectively irreversible: the which-path information is not destroyed, it is just spread where nobody can collect it (Zurek's environment-induced superselection).

Key takeaways

  • Which-path distinguishability and fringe visibility trade off exactly: D2+V2≤1D^2 + V^2 \le 1, with equality for pure joint states (Englert), and P≤DP \le D.
  • Measuring a perfect marker in a conjugate basis yields a fringe and an anti-fringe in the sorted subsets, which sum to a flat pattern.
  • No operation on the marker can change the unconditional statistics at the detectors, so delayed-choice erasers involve no retrocausality.
  • Decoherence is which-path information spread over many environmental systems: overlaps multiply, so visibility decays exponentially.

Check yourself

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

Question 1

After the marker is measured in the ∣±⟩\ket{\pm} basis, how do the unsorted D0 counts depend on φ\varphi?

  1. They show a full-visibility fringe cos⁡2(φ/2)\cos^2(\varphi/2)
  2. They show a half-visibility fringe
  3. They are flat at 12\tfrac12 for every φ\varphi
  4. They show the anti-fringe sin⁡2(φ/2)\sin^2(\varphi/2)
Show the answer

C. They are flat at 12\tfrac12 for every φ\varphi

The fringe (++ subset) and the anti-fringe (−- subset) each occur half the time and sum to 12\tfrac12. The eraser changes only conditional statistics.

Question 2

Pure markers, equal path weights, ∣⟨e0∣e1⟩∣=0.6\abs{\braket{e_0|e_1}} = 0.6. What is the distinguishability DD?

  1. 0.60.6
  2. 0.40.4
  3. 0.360.36
  4. 0.80.8
Show the answer

D. 0.80.8

V=0.6V = 0.6, and for pure markers D=1−V2=0.8D = \sqrt{1 - V^2} = 0.8.

Question 3

Each of 500 environmental particles records the path only very weakly, with overlap 0.990.99. What fringe visibility remains?

  1. about 99%99\%
  2. about 50%50\%
  3. about 0.7%0.7\%
  4. exactly 00
Show the answer

C. about 0.7%0.7\%

V=0.99500=e500ln⁡0.99≈e−5.0≈0.007V = 0.99^{500} = e^{500\ln 0.99} \approx e^{-5.0} \approx 0.007. Weak records compound exponentially.

Question 4

After the photon has been detected, which operation on the marker alone can change P(D0)P(\text{D0})?

  1. Measuring the marker in the ∣±⟩\ket{\pm} basis
  2. Measuring the marker in the ∣0⟩,∣1⟩\ket{0},\ket{1} basis
  3. Applying a Hadamard to the marker and then discarding it
  4. No operation on the marker alone can change it
Show the answer

D. No operation on the marker alone can change it

The path's reduced state is invariant under any channel on the marker, so its detection statistics are fixed once the marking interaction is done. Choices on the marker affect only correlations.

Further reading

  • M. O. Scully & K. Drühl, "Quantum eraser: A proposed photon correlation experiment concerning observation and 'delayed choice' in quantum mechanics", Phys. Rev. A 25, 2208 (1982).
  • D. M. Greenberger & A. Yasin, "Simultaneous wave and particle knowledge in a neutron interferometer", Phys. Lett. A 128, 391 (1988).
  • B.-G. Englert, "Fringe visibility and which-way information: an inequality", Phys. Rev. Lett. 77, 2154 (1996).
  • Y.-H. Kim, R. Yu, S. P. Kulik, Y. Shih & M. O. Scully, "Delayed 'choice' quantum eraser", Phys. Rev. Lett. 84, 1 (2000).
  • W. H. Zurek, "Decoherence, einselection, and the quantum origins of the classical", Rev. Mod. Phys. 75, 715 (2003).