A Monomer-Referenced Witness for Electronic Coherence in Light-Harvesting Complexes

Ultrafast spectroscopists keep seeing oscillations ("quantum beats") in the signals from photosynthetic antenna proteins. They still cannot say with confidence whether these beats come from electronic coherence, meaning the excitation is shared in a superposition across several pigments, or only from atoms vibrating inside each pigment. A clean theoretical test exists. If you integrate a two-dimensional electronic spectrum over all excitation and detection frequencies, vibrational beats cancel and only electronic coherence survives. That cancellation is exact only for infinitely short laser pulses and for transition strengths that do not depend on molecular geometry. Neither condition holds in a real laboratory. The leftover vibrational "leakage" can imitate coherence, or hide it.

This note proposes to handle that leakage the way reactor-neutrino physicists handle uncertainty in the source flux. Daya Bay's near detector, which sees the same source through the same hardware, predicts what the far detector should see. In our version, the "near detector" is the isolated, uncoupled pigment. It has zero electronic coherence by definition, yet it carries the same vibrational modes and passes through the same laser pulses and optics. We measure its integrated signal in the same session as the coupled complex and treat its residual oscillations as a direct calibration of the leakage. We then transport that leakage to the aggregate with a small physical model whose few unknowns are profiled as nuisance parameters, in the precision-physics sense of unknown parameters fitted alongside the quantity of interest. Any oscillation in the aggregate that exceeds the transported leakage is evidence for electronic coherence. Because every claim comes with a calibrated false-positive rate, it can be falsified.

Core Claim

The integrated pump-probe witness of A witness for coherent electronic vs vibronic-only oscillations in ultrafast spectroscopy defines

$$PP'(t_2)=\iint d\omega_1\,d\omega_3\,S_{2D}(\omega_1,t_2,\omega_3).$$

Here $S_{2D}$ is the two-dimensional spectrum at excitation frequency $\omega_1$, detection frequency $\omega_3$ and waiting time $t_2$. That paper proved $PP'$ is vibration-free only in the impulsive, Condon limit, and left open the finite-pulse and geometry-dependent-dipole (non-Condon) residuals.

Our claim is that these residuals are mostly common-mode between a chromophore and its aggregate. They can therefore be measured on the monomer and predicted for the aggregate, rather than assumed away. The test statistic is a profile likelihood ratio for an electronic-coherence amplitude $A$ in the model

$$PP'_{\rm agg}(t_2)=\mathcal{T}_\theta\!\left[PP'_{\rm mon}\right](t_2)+A\sum_k c_k\cos(\Omega_k t_2+\phi_k)e^{-t_2/\tau_k}+\epsilon(t_2).$$

In this model:

  • $\mathcal{T}_\theta$ is a transfer map from the measured monomer leakage to the aggregate leakage.
  • $\theta$ collects a handful of nuisance parameters: pulse-envelope mismatch, non-Condon slope, and the reduction of vibrational coupling strength that comes from exciton delocalization.
  • $\Omega_k$ are excitonic gap frequencies with relative weights $c_k$, phases $\phi_k$ and dephasing times $\tau_k$.
  • $\epsilon$ is measurement noise.

We test $A=0$ against $A>0$ after maximizing the likelihood over $\theta$. What is new is the measured null reference and the explicit nuisance treatment. The witness alone supplies neither. Ad hoc fits of decaying sinusoids to absolute spectra absorb leakage silently.

Why this framing, and what prior work fixes

Several papers supply the motivation.

Together these results argue against settling the question by fitting a full ab initio forward model to absolute spectra. The forward model's own systematic error is larger than the effect being sought.

Two further papers constrain the design. Coherent exciton dynamics in supramolecular light-harvesting nanotubes revealed by ultrafast quantum process tomography reconstructs the full dynamical map of C8S3 nanotubes from eight transient-grating experiments. It reports inter-wall coherence of 200 ± 120 fs, but it enforces physicality with semidefinite programming, which folds pulse calibration and phase drift into the fitted map. The method also scales exponentially with pigment count. Local protein solvation drives direct down-conversion in phycobiliprotein PC645 via incoherent vibronic transport reproduces the measured PC645 down-conversion rate (1.6 vs 1.7 ps⁻¹) without long-lived electronic delocalization. PC645 is therefore a natural system where a well-calibrated witness should usually return $A\approx0$.

Two papers supply methods.

We do not inherit that paper's second-order cumulant assumption for the aggregate. We use it only to bound the monomer non-Condon slope.

Research Program

Simulation stage. We build a simulated testbed in which the ground truth is known. For each system we generate 2D spectra with non-Markovian dynamics, including:

The systems are:

  • the monomer, coherent dimer and incoherent dimer of the witness paper, extended to finite pulses;
  • Nile blue in ethanol, a strongly vibronic real monomer with a validated neural potential;
  • a PC645 dimer model, the expected negative;
  • FMO under both the OLB and SAB site energies;
  • an inner/outer-wall C8S3 model, the expected positive.

Transfer map. We derive $\mathcal{T}_\theta$ analytically to first order in pulse duration and non-Condon slope. Delocalization shrinks the effective vibrational coupling by roughly the participation ratio, and the leakage amplitude is rescaled accordingly. We then check where this first-order map breaks down.

Statistics. We calibrate the test the way collider searches do. Thousands of null pseudo-experiments, generated with $A=0$ and $\theta$ drawn from its prior, give the empirical distribution of the test statistic and the real false-positive rate. Coverage of the confidence interval on $A$ and on the coherence lifetime is checked the same way.

Experimental protocol. The protocol interleaves monomer and aggregate measurements shot-block by shot-block with identical pulses. This is the analogue of identical detectors at different baselines. It cancels pulse shape and slow phase drift to first order, so only the differential residual has to be modeled.

Baselines

The naive reference point is windowed Fourier or decaying-sinusoid fitting of 2D cross-peak volumes versus $t_2$. It is included because it remains common practice. The witness paper already shows it fires on uncoupled monomers, so we expect a high false-positive rate on every vibronic-only system. Beating it means our false-positive rate stays at its nominal level while theirs does not.

The strongest competitor is the bare integrated witness $PP'$ of A witness for coherent electronic vs vibronic-only oscillations in ultrafast spectroscopy, applied directly to finite-pulse, non-Condon data and thresholded on oscillation amplitude. It is the only existing method aimed at exactly this question. It will run on the same simulated and measured spectra, with its threshold tuned on the same null pseudo-experiments, so that both methods are compared at matched nominal false-positive rate. I found no reported result on Lacuna for the bare witness's false-positive rate under finite pulses or non-Condon coupling; establishing that number is itself a first deliverable. Beating it means higher detection power at the same calibrated false-positive rate on the coherent dimer and C8S3 models, and staying near nominal false-positive rate on PC645 and Nile blue, where we predict the bare witness will exceed it.

For the experimental positive control, the reference number is the tomography result of 200 ± 120 fs inter-wall coherence in C8S3 (Coherent exciton dynamics in supramolecular light-harvesting nanotubes revealed by ultrafast quantum process tomography). Our lifetime estimate should agree with it within errors, and our interval should be narrower. Our method needs two samples rather than eight pulse configurations, which is where the narrower interval should come from.

Two ablations isolate the key ingredients. The first removes the measured monomer reference and lets $\mathcal{T}_\theta$ act on a simulated monomer instead. This tells us whether the measured null, rather than the nuisance model, does the work. The second keeps the monomer but fixes $\theta$ at nominal values without profiling, which tests whether nuisance profiling is needed.

Why not just…?

Why not just use shorter pulses until the impulsive limit holds? Shorter pulses reduce only the finite-bandwidth part of the leakage. The non-Condon part comes from the molecule, not the laser, so it persists at any pulse length. The simulation sweep from 10 to 40 fs will show directly how much residual remains at the shortest pulses currently achievable. If leakage at 10 fs falls below the noise for every test system, the monomer reference becomes unnecessary for those systems. We would report that honestly as a negative result for the instrumental half of the method.

Why not just fit a full ab initio forward model to the aggregate spectrum? This is the real threat when the Hamiltonian is well known. For FMO it is not: OLB and SAB change the trap population by more than 100% (Absence of Quantum Oscillations and Dependence on Site Energies in Electronic Excitation Transfer in the Fenna-Matthews-Olson Trimer). We will run the forward-fit baseline under both parameter sets and check whether its coherence verdict flips. Our statistic depends on site energies only through the frequencies $\Omega_k$, which we scan over, so we predict it will not flip. If the forward fit stays stable across OLB and SAB, our advantage there shrinks to robustness against pulse miscalibration alone.

Why not just do quantum process tomography everywhere? Tomography gives the full dynamical map, but the number of required state preparations grows exponentially with pigment count. It also absorbs drift through semidefinite projection rather than isolating it (Coherent exciton dynamics in supramolecular light-harvesting nanotubes revealed by ultrafast quantum process tomography). We use it as the reference on C8S3, where it is feasible, and not as a competitor on FMO, where it is not.

Why not just bound leakage by simulating the monomer instead of measuring it? The first ablation tests exactly this. If simulated and measured monomers give the same false-positive rate, the experimental reference adds nothing beyond cancelling drift. The measured reference earns its place only if real monomer residuals contain structure, such as chirp, scatter or solvent modes, that the simulation misses.

Decisive Checks

The method is supported if four things hold:

  1. Across null pseudo-experiments on Nile blue and PC645 with randomized $\theta$, the false-positive rate of our test stays within its nominal 5%, while the bare witness exceeds it whenever pulses are longer than about 15 fs or the non-Condon slope is nonzero.
  2. On the simulated coherent dimer and C8S3 model, detection power at matched false-positive rate exceeds that of the bare witness.
  3. On FMO, the verdict on $A$ is the same under OLB and SAB.
  4. On measured C8S3 versus its monomer dye, the recovered inter-wall lifetime agrees with 200 ± 120 fs.

The method fails if the first-order transfer map leaves biased residuals on the incoherent dimer, meaning aggregate leakage is not transportable from the monomer. In that case the near-detector analogy breaks, and only the bare-witness false-positive characterization remains as the contribution.