Multiply Scattered Light as a Probe of Flow Startup in Tight Porous Media

In very fine-grained rocks such as tight sandstones and shales, water often appears not to move at all until the pushing force exceeds a minimum value, the threshold pressure gradient. Above that value it flows, but not in simple proportion to the force. Reservoir models rely on this threshold because it limits how much rock around a well can drain. Yet nobody has watched it happen inside the pores. Every threshold value in the literature is inferred from the trickle of fluid leaving a core sample, and below the threshold that trickle is too small to tell "zero" apart from "small." We propose to measure flow startup from inside the medium using a phenomenon that imaging researchers usually treat as a nuisance: multiple scattering of light.

When laser light enters a dense, cloudy medium, it bounces off thousands of particles before it exits. The resulting speckle pattern, the grainy interference image of the exiting light, flickers whenever anything along those long paths moves. Every scattering event adds a little phase shift. A path of $N$ scatterings therefore responds to motions roughly $\sqrt{N}$ times smaller than a single bounce would detect. A water-saturated pack of sub-micron silica grains is optically thick and is also a good laboratory stand-in for tight rock. We will add a dilute nanoparticle tracer to the pore water, step the applied pressure gradient up through the expected threshold, and read off how fast the speckle loses its similarity to itself over time (its decorrelation). From that we recover the local shear rate of pore fluid, and by tomographic inversion its spatial map. The extra phase accumulated over long multiply scattered paths is what turns an unmeasurable creeping flow into a measurable signal.

Core Claim

Multiple scattering amplifies sensitivity to small motions. This makes diffusing-wave spectroscopy, which measures motion from the time-correlation of multiply scattered light, able to resolve pore-fluid shear far below the detection limit of effluent-based threshold measurements. Its time signature also separates three regimes in a single measurement: true stagnation, sub-threshold creep, and post-threshold flow. The quantity measured is the intensity autocorrelation of transmitted speckle. Its field counterpart $g_1(\tau)$ at lag time $\tau$ is, to good approximation,

$$g_1(\tau) \approx \int_0^\infty P(s)\, \exp\!\left(-\frac{s}{l^*}\,\frac{k_0^2}{3}\,\langle \Delta r^2(\tau)\rangle\right) ds,$$

where $P(s)$ is the distribution of photon path lengths $s$ through the sample, $l^*$ is the transport mean free path (the distance over which light forgets its direction), and $k_0$ is the optical wavenumber in the pore fluid. The tracer's mean-square relative displacement splits as $\langle \Delta r^2(\tau)\rangle = 6 D_B \tau + c\,\Gamma^2 l^{*2} \tau^2$. Here $D_B$ is the tracer's Brownian diffusion coefficient, $\Gamma$ is the local shear rate of the pore fluid, and $c$ is an order-one geometric constant. The linear-in-$\tau$ term comes from thermal jiggling. The quadratic term comes from directed flow, because sheared tracers separate ballistically. Brownian motion therefore sets a fixed background, and the flow term is identified by its different dependence on time and on sample thickness.

This makes the threshold question decidable. A true threshold, as implied by the boundary-layer picture of Nonlinear flow in low permeability reservoirs, predicts $\Gamma = 0$ within noise below a critical gradient. A smooth nonlinear law without a threshold predicts small but nonzero $\Gamma$ that grows continuously from zero gradient. That paper models tight rock as capillary tubes lined with an immobile layer whose thickness decays exponentially with driving force, and it fits this from bulk flow rates on six cores of 4.2 to 8.3 mD. We test the same model's interior prediction directly, rather than its integrated outflow.

Research Program

The first component is a family of optically thick analog media whose permeabilities match the tight cores in the file, roughly 3 to 35 mD following the range of A Study on Characteristics of Oil–Water Relative Permeability Curves. We will sinter silica powders of controlled grain size into slabs 2 to 10 mm thick. With water in the pores, the silica–water refractive index contrast gives $l^*$ in the tens of microns, so the slabs span tens to hundreds of transport lengths. We will characterize pore-throat size with NMR relaxation using the power-law radius conversion validated in Characterization of Pore Throat Size Distribution in Tight Sandstones, which matched mercury intrusion to 8.6% average error. We will also measure permeability conventionally, so that each slab has a standard threshold estimate to compare against. Varying salinity, including the roughly 10,500 mg/L brine Song et al. used, changes the strength of surface adhesion. This gives a physical control: the boundary-layer model predicts that the threshold should shift with salinity, while optical artifacts should not.

The second component is the measurement itself. A long-coherence laser and a fast photon-counting or camera-based correlator will record transmitted speckle while a syringe pump or pressure controller steps the gradient in fine increments from zero through and beyond the expected threshold. A tracer of 20 to 50 nm particles at very low volume fraction gives the flow something to move. We will repeat measurements across tracer sizes to check that the tracer itself does not clog throats or alter the adhesion layer. The read-out is the local shear rate as a function of applied gradient, with a confidence interval set by the Brownian background.

The third component is spatial mapping. Feng et al. argue that the threshold restricts drainage to the region near a fracture where the local gradient exceeds the threshold. With a camera instead of a single detector, and multiple source–detector positions, we can reconstruct where in the slab fluid moves. Mathematically this is the radiative transfer equation with the motion term above acting like an extra, lag-dependent absorption. We will invert it with gradient-based tomography in the style of Multiple-Scattering Microphysics Tomography. That work recovers cloud droplet size parameters by differentiating a 3D multiple-scattering solver, and it cut cloud mass error from -55% to -8% by fitting physical parameters instead of generic voxels. We borrow that lesson and parameterize the unknown as a mobility field tied to the boundary-layer model rather than as free per-voxel shear.

The file's warnings about ill-posedness and local minima in inverse transport constrain this design. Slabs will be kept in a moderate thickness range, and the scattering structure $l^*$ will be calibrated separately in a no-flow state, so that only the dynamic term is inverted. The broader lesson that modeling scattered light, rather than discarding it, recovers lost information is supported by Photometric Stereo in a Scattering Medium. That paper reduced underwater depth error from 14% to 4% by explicitly modeling forward scatter.

Baselines

The naive reference point is the bulk effluent method used by Song et al. and Liu et al.: precise pump control, outflow weighed on a microbalance over long intervals, and a threshold read off by extrapolating the flow-rate-versus-gradient curve to zero flow. We will run it on the same slabs, at the same temperature, and over the same total measurement time as the optical method. We will also log evaporation and thermal drift so that its true detection floor is known. No reported result was found on Lacuna for the detection floor of this method in the cited studies. We will therefore report our own measured floor, in volumetric flux, rather than quote one. Beating this baseline means two things. First, a statistically resolved shear signal must appear at applied gradients where effluent flux is indistinguishable from zero. Second, the optical method must reach a threshold estimate with a narrower confidence interval in equal time.

The strongest existing in-situ method is time-resolved synchrotron micro-CT, the data source behind Learning Pore-scale Multiphase Flow from 4D Velocimetry and Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network. Both papers image tracer particles or fluid interfaces at micrometer resolution, at about 2.75 µm voxels in the latter, and learn simulators from the resulting movies. Their reported strengths are in prediction: particle position $R^2 = 0.9999$ over 30 timesteps and interface Dice above 0.98 for the GNS. The file reports no velocity detection limit for these systems, so no reported result was found on Lacuna for their slow-flow sensitivity. We cannot obtain synchrotron time for every slab. We will therefore treat their resolution as a stated limit and test it with laboratory micro-CT tracer tracking on one coarse slab. The comparison is fair only where both methods resolve flow. Our method beats this baseline if it detects motion in slabs whose throats are below the CT voxel size, where CT sees no resolvable pores at all.

The key ablation removes multiple scattering. We will repeat measurements on slabs thinner than a few transport lengths, and in a short-path backscatter geometry, keeping the laser, tracer, and correlator identical. If the amplification argument is right, flow sensitivity at fixed gradient should grow with thickness roughly as $(L/l^*)^2$, where $L$ is slab thickness, because that factor counts the scatterings on a typical transmitted path. If thin and thick samples give equal sensitivity, the multiple-scattering ingredient is doing nothing, and the proposal reduces to ordinary speckle velocimetry. A second ablation fixes the tracer and varies salinity, which separates adhesion physics from optics.

Why not just…?

Why not just measure effluent more precisely and for longer? Longer runs reduce random noise but not drift. Evaporation, temperature-driven volume changes, and pump creep produce apparent flows of the same order as sub-threshold creep. An integrated outflow also cannot say where in the sample fluid moves, which is the quantity the drainage-volume argument actually needs. The effluent baseline above is run in parallel with matched time budgets precisely to show whether it can match the optical floor. If it can, the spatial map remains the optical method's remaining contribution.

Why not use pulsed-field-gradient NMR, which already measures slow displacement inside opaque rock? This is the most serious alternative, because it works in real rock while our method needs optically clear analogs. Its limitation is that water self-diffusion displaces molecules tens of microns over typical encoding times, which swamps directed flow at nanometers per second. The optical method uses large, slowly diffusing tracers and separates flow by its $\tau^2$ signature. We will measure NMR displacement propagators on the same slabs at the same gradients. If NMR resolves the threshold equally well, we will say so, and the optical method's case narrows to spatial mapping and speed.

Why not just use X-ray micro-CT velocimetry? Lab and synchrotron CT resolve pores near or above a few microns. Tight-rock throats, and the adhesion layers that are supposed to cause the threshold, are far smaller. CT tracer tracking also needs displacements of at least a voxel between frames, which sub-threshold creep may never produce on practical timescales. The coarse-slab comparison in the baselines is designed to show where the two methods overlap.

Proposed Experiments

The decisive experiment is the gradient sweep on matched slabs. For each slab we fit the decorrelation curves at each gradient to the Brownian-plus-shear model and plot recovered shear rate against applied gradient, with salinity as a second axis. Three outcomes are possible. Shear indistinguishable from zero up to a sharp gradient, shifting with salinity, supports a true boundary-layer threshold. Smoothly rising shear from zero gradient means the reported thresholds reflect measurement floors rather than physics. A threshold that does not move with salinity points to a mechanism other than adhesion. Each outcome changes how reservoir models should treat the threshold. The thickness ablation tests whether multiple scattering supplies the claimed sensitivity gain.

A second experiment validates the spatial maps. A slab is made with a deliberate permeability contrast, half coarse and half fine, and placed under a gradient chosen to exceed the threshold of only the coarse half. The reconstruction should show flow confined to that half. Error is measured against the known layout, and we report how quickly reconstruction quality degrades as optical thickness rises toward the saturation regime the file flags as a physical limit.

A third experiment extends the quasi-threshold behavior of Liu et al. to two phases. They found the apparent threshold peaks and then falls as water saturation rises. We will introduce a weakly scattering immiscible oil phase and track whether interface motion, which also decorrelates speckle, begins at saturation-dependent gradients matching that rise-and-fall shape.

A stretch goal tests thin, near-infrared-transparent slices of natural quartz-rich tight sandstone. Clay absorption may prevent the diffuse regime there, and we will report that as a limitation if it does.