---
title: "Read a centre-line paper and test frequency memory"
description: "Read Fenn and Fayer’s 2011 Figures 1–3, reproduce the Gaussian limit in Eq. (12), and test which inferences survive withheld observations."
canonical_url: "https://sandboxphysics.com/stories/multidimensional-spectroscopy/reading-and-inference/"
language: "en"
page_type: "story"
alternate_languages:
  zh-CN: "https://sandboxphysics.com/zh/stories/multidimensional-spectroscopy/reading-and-inference/"
last_modified: "2026-10-28"
---

[Multidimensional spectroscopy](https://sandboxphysics.com/stories/multidimensional-spectroscopy/)

L4 · Chapter 8

# Read a centre-line paper and test frequency memory

Sandbox Physics · 2026-10-28

Open Fenn and Fayer, Journal of Chemical Physics 135, 074502 (2011). Begin with the slicing direction in Figure 1, then the time curve in Figure 2, and finally the mixture in Figure 3. We reproduce the single-component Gaussian relation in Eq. (12) with original synthetic parameters. Reconstructing the paper’s water measurements is outside this calculation.

## Before you begin

Read chapters six and seven. Keep the author-hosted publisher-version PDF open and understand covariance, centre-line cuts and withheld prediction.

[Review the preceding chapter](https://sandboxphysics.com/stories/multidimensional-spectroscopy/spectral-diffusion/)

- Distinguish calculated spectra, experimental context and method validation.
- Independently reproduce the single-component Gaussian slope relation.
- Explain why mixtures, lifetimes and finite windows complicate inversion.

## Figure 1 establishes the slicing direction

Figure 1 appears on printed page 074502-2. Excitation frequency $\omega_\tau$ is horizontal; detection frequency $\omega_m$ is vertical. Numerical axes are in inverse centimetres. The panels use waiting times $0.2\,\mathrm{ps}$ and $2\,\mathrm{ps}$. The caption identifies calculated spectra based on a known water correlation function.

Solid lines mark cuts parallel to the detection axis. Dotted lines connect their maxima. The later peak is more symmetric and the centre line flatter. The direction matters: other papers can use opposite cuts and report an inverse slope.

Reading: Fenn and Fayer

## Figure 2 starts below unity

Figure 2 on the next page collects centre-line slopes against waiting time. The early values are already well below one, associated in the paper with a substantial homogeneous contribution. Fast fluctuations can enter the motionally narrowed component. Forcing the first point to unity and reading every amplitude from a single decay would miss this distinction.

Read Eqs. (1) and (4) next. A correlation with fast and exponential contributions enters a lineshape function through a double time integral. The centre-line series provides waiting-time information; linear absorption helps determine the fast contribution. These observations constrain different parts of the model.

Reading: Fenn and Fayer

## Reproduce the single-component limit in Eq. (12)

Equation (12) states a single-component Gaussian relation between detection-axis centre-line slope and $C(T)/C(0)$. Use standard deviation $18\,\mathrm{cm}^{-1}$, correlation $0.6$ and initial offset $10\,\mathrm{cm}^{-1}$. The predicted detection-axis maximum lies at $6\,\mathrm{cm}^{-1}$.

An independent numerical check searches slice maxima directly in a bivariate Gaussian density and fits a line through them. It does not call the response code’s slope formula. A separate frequency integral checks how Gaussian pulse selection changes covariance. Saved synthetic complex records, seeds and versions support reconstruction of the full series.

Now add independent detection-axis Gaussian broadening with standard deviation $6\,\mathrm{cm}^{-1}$. Adding variance only to that axis leaves the conditional-mean slope unchanged. Adding the same independent broadening to both axes instead gives $0.6\times324/(324+36)=0.54$. Calibration must specify which axes a process affects.

$$
\mathrm{CLS}=\frac{\operatorname{Cov}(x,y)}{\operatorname{Var}(x)}=\frac{C(T)}{C(0)}
$$

The final equality assumes the ideal single-component frequency pair with equal marginal variances. Windowing, finite pulses and overlapping pathways can modify the observed relation.

Reading: Fenn and Fayer

## Figure 3 separates two environments

Figure 3 appears on printed page 074502-5. The black curve is the OD-stretch absorption of HOD in a reverse micelle. Blue and red curves represent core and interfacial contributions. Equation (14) combines them with $a_1=0.56$. This coefficient belongs to that decomposition; it is not a universal mixture fraction.

Different excitation frequencies contain different component fractions. Their relative signal also changes if their lifetimes differ. The slice maxima depend nonlinearly on the component spectra and on their frequency- and time-dependent fractions.

Reading: Fenn and Fayer

## Lifetimes change the surviving mixture

Equation (15) makes the weight depend on frequency and wait. Let $S_1(x),S_2(x)$ be the initial spectral contributions at one excitation frequency, and $L_1,L_2$ their population lifetimes. We use $L$ here to avoid confusing lifetime with the first coherence interval.

For a new teaching example, take initial contributions 0.56 and 0.44 and lifetimes $1\,\mathrm{ps}$ and $2\,\mathrm{ps}$. After $1\,\mathrm{ps}$, the first component contributes about 0.436 of the surviving signal. The mixture changes even if each environment itself is static.

$$
f_1(x,T)=\frac{S_1(x)e^{-T/L_1}}{S_1(x)e^{-T/L_1}+S_2(x)e^{-T/L_2}}
$$

The paper’s Eq. (15), with relabelled lifetimes. The numerical lifetimes above are illustrative and are not claimed as the paper’s measurements.

Reading: Fenn and Fayer

## State exactly what has been reproduced

The reproduced result is the analytic Gaussian limit in Eq. (12) and its independent numerical checks. The interactive experiment separately generates finite-bandwidth, finite-window noisy complex signals and compares correlation models by withheld prediction. It does not digitize water spectra or recover the paper’s two-component micellar correlations.

A continuously fluctuating liquid also changes frequency during coherence intervals. Our frequency-pair model freezes that evolution. When its correlation time approaches or falls below the coherence window, it remains a defined phenomenological model, but quantitative solvent inference requires the full stochastic process and response convolution.

Reading: Tokmakoff

## A reading sequence for the next paper

Record version, figure numbers, axes and units, and identify each panel as measurement, calculation or ideal reference. Trace processing from the original signal, including phase, window and slice direction. Then list which observation determines each parameter and whether independent data check predictions.

If two microscopic models give indistinguishable spectra within the available bandwidth and noise, design a discriminating observation. Changing polarization, extending the waiting-time range or adding structural information can be more informative than adding parameters to the same map.

## Try the prediction

### [Compare Eq. (12) with noisy inference](https://sandboxphysics.com/labs/spectral-diffusion-inference/)

1. Use low noise and broad bandwidth to inspect the waiting-time centre lines.
2. Compare held-out errors and identify late records that distinguish an apparent static part from a slow decay.
3. Increase noise and explain why occasional estimates outside the physical range should remain visible as diagnostics instead of being clipped into apparent agreement.

## Exercises and explanations

1. Does a flatter mixture centre line mean that both components lost the same amount of frequency memory?

   Read the explanation

   No. Spectral overlap, frequency-dependent fractions and different population lifetimes also change the centre line. Component absorption, lifetimes and an appropriate mixture response are needed.

2. Why does reproducing Eq. (12) not reproduce the entire paper?

   Read the explanation

   It checks one explicitly defined limit. The paper also studies two-component conditions, specific spectra and a back-calculation algorithm, requiring additional parameters, data and implementation. The course labels each adopted and unadopted part.

[Previous: How a molecule forgets its earlier frequency](https://sandboxphysics.com/stories/multidimensional-spectroscopy/spectral-diffusion/)

## Sources and calculation scope

1. [Tokmakoff, Two-dimensional spectroscopy chapter (June 2011)](https://doi.org/10.6082/uchicago.3314)

   In the same deposited notes, the section whose printed page numbers restart at 1: pp. 4–5, Eqs. (1)–(3), connect response, Fourier transformation and heterodyne detection; pp. 10–11 discuss coupled-vibration peak pairs. No source figure is redistributed.

2. [Fenn and Fayer, Extracting 2D IR frequency-frequency correlation functions from two component systems (2011)](https://web.stanford.edu/group/fayer/articles/407.pdf)

   Journal of Chemical Physics 135, 074502, publisher version, DOI 10.1063/1.3625278. Read Figs. 1–3 and Eqs. (1)–(4), (12), (14)–(15). The bounded numerical benchmark is the single-component Gaussian slope relation in Eq. (12), with original synthetic parameters; it is not a fit to the paper’s water measurements.