A click without a mode label
A broadband herald reports an idler photon without identifying its mode. For each possible idler mode, one column of the pair matrix gives the signal amplitudes. Amplitudes within that column remain coherent. Ignoring distinct idler outcomes means mixing their probabilities. Matrix multiplication performs exactly this sum.
Complex conjugation reverses the imaginary part and is written with a star. Transposition followed by conjugation gives the adjoint . The signal matrix entry multiplies signal amplitudes from the same idler column, then sums over columns. Diagonal entries are probabilities; off-diagonal entries describe coherence.
Reading: A. Bhattacharjee et al.
Rotate one contribution to coherence
For , the first column contributes one quarter to signal coherence. The second contributes . At zero phase they point together; after half a turn they cancel. The two diagonal probabilities remain one half each.
At zero phase the columns are identical, so learning the idler mode would not alter the signal wave packet. At half a turn the columns describe orthogonal signal superpositions. The idler can identify which one accompanied it. Ignoring that information leaves an equal mixture.
All phases aligned
One arrow reversed
Find the paired modes with Schmidt decomposition
A suitable choice of modes puts the pair state into a sum of matched signal-idler pairs. Each signal mode then accompanies just one idler mode. This is Schmidt decomposition. Its nonnegative weights are eigenvalues of the signal density matrix: the matrix scales each corresponding mode without changing its direction.
The singular-value decomposition writes . Signal Schmidt modes are columns of ; idler modes are columns of . Their weights are squared singular values. The conjugation matters for rebuilding a complex biphoton amplitude.
Reading: A. Eckstein
A different basis leaves purity unchanged
Changing mode basis resembles changing coordinates on a map. A unitary change preserves lengths and inner products. Matrix entries change, but eigenvalues and purity do not. A mixture cannot become pure through relabeling modes.
This gives useful independent checks. Direct coefficient multiplication, eigenvalues of the reduced matrix and singular values of the pair matrix should agree. Unit trace and nonnegative eigenvalues are required before the weights can represent probabilities.