arXiv:2609.31043
Single-shot coherent process tomography and mid-infrared polarimetry with undetected photons
Marthe Zeja, Wen-Zhe Yan, Zhibo Hou, Sven Ramelow
Abstract
Measurements with undetected photons infer properties of a probe beam that is never detected, transferring mid-infrared information onto silicon-friendly wavelengths. Tomography in this paradigm has so far relied on switched settings or scanned phases, referenced across a drift-sensitive campaign; no protocol reads all Jones parameters from one frame. We show that a folded nonlinear interferometer with a diagonally pumped crossed-crystal source performs coherent process tomography of the undetected beam in a single spectrometer frame: it reconstructs all seven parameters of the beam's round-trip Jones matrix, the coherent part of the channel. Birefringent group-delay walk-off, normally a calibration nuisance, acts as a frequency multiplexer: each Jones-matrix element is assigned its own carrier, a spectral fringe period paired with a detector port. Because the multiplexing exploits the down-conversion bandwidth rather than merely tolerating it, a single Fourier transform per detector trace returns all four moduli and all three relative phases across that bandwidth, i.e., the full round-trip Jones spectrum up to a global phase, with no scan or setting change (after a one-time reference frame). Simultaneity makes every relative phase immune to common-mode drift, and a dark carrier, a self-interference term that unitarity forces to vanish, provides a built-in null test. For realistic parameters of a periodically poled KTP source the scheme yields wavelength-resolved mid-infrared polarimetry across 3.5-4.2 microns (~80 spectral points per Jones element) on a silicon camera, extending to depth-resolved Jones matrices, i.e., polarization-sensitive optical coherence tomography with undetected photons. A Fisher-information analysis, the first we are aware of in undetected-photon tomography, sets the precision budget and compares delay multiplexing to sequential protocols at equal photon number.
AI Summary
The paper asks whether polarization-process tomography with undetected photons can recover a complex Jones matrix without switching settings or scanning phases. It proposes a low-gain, folded nonlinear interferometer with diagonally pumped crossed crystals. Birefringent delays assign four probe–reference interference terms to distinguishable spectral carriers across two simultaneously recorded signal-polarization ports. Calibrated Fourier demodulation recovers four moduli and three relative phases of the undetected idler’s round-trip Jones matrix $M$, up to global phase, from one exposure. An additional carrier vanishes by unitarity when environmental loss modes are included, providing an alignment/readout null test. The results are theoretical and numerical. A ppKTP design maps a silicon-detected spectrum near $800\,\mathrm{nm}$ onto $3.5$–$4.2\,\mu\mathrm{m}$ idler polarimetry, with approximately 80 independent spectral points per matrix element. Monte Carlo reconstruction at $10^6$ detected photons matches classical Cramér–Rao bounds within 5%. At equal photon budget, multiplexed and two-setting sequential tomography have approximately equal Fisher information without drift; multiplexing avoids inter-frame phase-reference errors. Quantum-Fisher and Holevo-bound calculations quantify readout inefficiency and the information cost of leaving the idler undetected. The reconstruction requires reference calibration, dispersion-aware demodulation, simultaneous port acquisition, and wavelength-registration tracking. Carrier separation limits spectral resolution and sample differential group delay, approximately $2\,\mathrm{ps}$ single-pass for the example placement. Only the coherent channel amplitude is recovered, not general decoherence; reciprocal round trips have five independent parameters and do not determine the single-pass matrix. Experimental implementation remains prospective.
The authors used Claude (Anthropic; Claude Opus 5 and Claude Fable 5.1, accessed 2026) as an AI coding and writing assistant during the preparation of this work.
Error found: The abstract's one-FFT reconstruction description omits the family-specific calibrated dispersion correction or fitting required by the realistic implementation. The advertised approximately 80 independent spectral points per Jones element are not generally supported by the specified matched-window readout, because its 5 ps windows exceed the reported minimum local carrier separation of 3.9 ps.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 27 pages, 3 figures, 5 tables