Skip to main content
Back to timeline
arXivSource publication:

Coherent scattering beyond average relaxation: a predictive-sufficiency test shows which dynamics correlation functions leave out

Synopsis

X-ray photon correlation spectroscopy (XPCS) usually compresses a speckle movie into intensity correlation functions, and that compression can discard information needed for later dynamics. An "information microscope" compares prediction from measurable XPCS observables alone with prediction after adding a learned representation of the present speckle sequence; a gain means the conventional description has left information unused, and a library of candidate observables built from speckle intensities is then searched to recover it. In simulated Bragg scattering, two added observables raise the prediction score from 0.347 to 0.412, while the learned representation retains a 0.047 increment.

Source-provided article image: Predictive sufficiency as an information microscope for time-resolved coherent scattering
arXiv

Interpretation

A time-resolved speckle record can be tested for whether a chosen set of XPCS observables retains the information needed to describe subsequent measured scattering dynamics. Conventional practice reduces the speckle movie to intensity autocorrelation, relaxation time, fluctuation amplitude and similar descriptors; these give physical insight, but whether the reduction discards information relevant to later dynamics is usually not tested. In stationary and smoothly aging Brownian-dynamics controls, once the explicit description includes correlations at several wave vectors and delays, their temporal fluctuations, and correlations over successive intervals of the present window, adding the learned speckle representation yields no statistically resolved improvement.

In simulated coherent scattering around a Bragg reflection, the spatial distribution of speckle activity and the anisotropy of the pattern carry information beyond average intensity decorrelation. Using only the conventional intensity autocorrelation at a four-frame delay and a spatial Pearson correlation between speckle patterns to predict the same quantities 16 frames later in a non-overlapping window gives a score of 0.347. Adding the learned representation raises the score to 0.459; adding two observables built from speckle intensities raises it to 0.412 and cuts the remaining learned increment to 0.047, and five observables give 0.417 with 0.042.

In a controlled small-angle scattering model, temporal fluctuations of speckle activity provide sensitivity to the spatial extent of cooperative motion, while the mean correlation mainly constrains the collective displacement amplitude. With the mean intensity autocorrelation and a fluctuation-like quantity as the compact description, prediction of the same later quantities scores 0.523. Adding the learned representation raises it to 0.616; two automatically selected observables give 0.602 with a remaining increment of 0.014, five give 0.609 with 0.008, and cooperative domain size is only weakly constrained by the compact description, with both enriched descriptions clearly improving its reconstruction.

In a two-dimensional glass former, retaining time-resolved speckle correlations together with instantaneous radial intensity recovers information missing from a fixed mean-relaxation reference. With a fixed-prefix mean correlation as reference, adding the learned representation raises the prediction score by 0.196. In the largest test cohort, five scattering observables raise the score from -0.173 to 0.041; adding instantaneous radial intensity raises it from 0.041 to 0.056, and the remaining learned contribution is no longer resolved.

Perspective

The demonstrated scope is three simulated benchmarks: Bragg scattering, a controlled small-angle scattering model and a two-dimensional glass former, with both the observable search and the predictive readout performed on these simulated speckle data; the authors note that predictive sufficiency depends on the chosen future scattering quantity and on the time scales the measurement can reach.

The authors state that the predictive gain is not a direct measurement of mutual information and is not a proof of information-theoretic sufficiency; the absence of a resolved gain holds only within the tested observation window, target, prediction model and data. For Bragg and cooperative small-angle scattering, observable selection stops at no more than five additions on validation data, and this stopping rule is not evidence of universal sufficiency; the glass causal search and the physically motivated state ladder are separate benchmarks whose scores are not directly interchangeable.