Disorder-Induced Capillary Bursts Control Intermittency in Slow Imbibition

A multiscale analysis of the spatially averaged velocity of an imbibition front VlðtÞ measured at scalel reveals that the slow front dynamics is intermittent: the distributions of ΔVlðτÞ 1⁄4 Vlðtþ τÞ − VlðtÞ evolve continuously through time scales τ, from heavy-tailed to Gaussian—reached at a time lag τc set by the extent of the medium heterogeneities. Intermittency results from capillary bursts triggered from the smallest scale of the disorder up to the scale lc at which viscous dissipation becomes dominant. The effective number of degrees of freedom of the front l=lc controls its intensity.

Among the striking spatiotemporal features of porous media flows, persistent intermittent properties of Lagrangian velocities have been observed very recently in numerical simulations [12].Intermittency is a key concept in hydrodynamic turbulence, associated with the occurrence of bursts of intense motion within more quiescent fluid flow [13].It leads to strong deviations from Gaussian statistics, that become larger when considering fluctuations at smaller scales [14,15].In the context of interfacial dynamics, small-scale intermittency was observed in gravity-capillary wave turbulence [16], a system which strongly differs from high Reynolds hydrodynamic turbulence [17].Thus, understanding the physical mechanism of intermittency remains a challenge [16][17][18][19][20]. Recently, it was suggested that it could be related to the properties of the fluctuations of the energy flux shared by different systems displaying an energy cascade [16,17].
In this Letter, we perform a multiscale analysis (in space and time) of the spatially averaged velocity of a fluid front slowly invading a disordered medium.While in Ref. [6] we studied the statistical properties of this velocity, the novel study presented here reveals that this velocity displays complex temporal correlations and exhibits all the distinguishing features of an intermittent dynamics.More importantly, expanding the experimental parameter space by using fluids of different viscosities and a wider range of imposed flow rates, we disclose which variables control intermittency in this system.Our work provides the first experimental evidence and detailed characterization of intermittency in fluid invasion of confined heterogeneous media, and brings new insight into the origin of intermittency in extended nonequilibrium systems.
Our experiments emulated the invasion of an open fracture of variable aperture by a wetting, viscous fluid.The experimental setup, shown in the top panel of Fig. 1, has been described in detail elsewhere [4,7].We recall, briefly, its main features.It consists of two horizontal, parallel glass plates (190 × 500 mm 2 ), separated by a narrow gap spacing.Nonoverlapping square patches (0.4 × 0.4 mm 2 ) randomly distributed in space and filling 35% of the bottom plate provide dichotomic fluctuations of the gap thickness between 0.40 and 0.46 mm, modifying the permeability κ of the cell [21].In these conditions, the lateral extent of the islands formed by adjacent patches follows an exponential distribution, with a characteristic length l d ¼ 0.6 mm [8].Silicone oil was injected at a constant flow rate, and displaced the air initially present.The nearly flat initial front became progressively rougher, until a statistically stationary state was reached in which the front width saturated to a constant value.We recorded the advancement of the fluid front, in the stationary state, with high spatial and temporal resolution (r ¼ 0.106 mm=pixel and up to 200 frames=s).The size of the recorded frames was L ¼ 136 mm in the transverse direction (x) and 25 to 45 mm in the propagation direction (y).
We used silicone oils of five different dynamic viscosities, from μ ¼ 10 to 350 cP, and similar oil-air surface tension around γ ¼ 21 mN=m.The imposed flow rates produced mean front velocities in the range 0.04 < v < 0.6 mm=s, giving capillary numbers Ca ¼ μv=γ between 6 × 10 −5 and 2 × 10 −3 .For each given experimental condition (v; μ), we performed 15 to 20 experiments with different disorder realizations.The interface profiles hðx; tÞ were obtained from the recorded snapshots by applying an edge-tracking algorithm.Measuring the amount of time spent by the front on each ðx; yÞ position, wtðx; y ¼ hðx; tÞÞ, we computed the map of local front velocities vðx; yÞ ¼ r=wtðx; yÞ [22], where r is the spatial resolution specified earlier.Finally, the global velocity of the front on scale l was obtained by averaging a spatiotemporal map vðx; tÞ of these local velocities over a window of lateral size l, V l ðtÞ ¼ ð1=lÞ R l vðx; tÞdx.Figure 1 shows a typical velocity signal V l ðtÞ, and its time derivative.
Because of the medium heterogeneities, capillary pressure and permeability variations distort the imbibition front, while viscous pressure and surface tension tend to restore its flatness (see the videos in the Supplemental Material [23]).For slow displacements (capillary regime) the oil-air interface develops long range correlations, with a lateral correlation length given by the crossover length scale l c ¼ ffiffiffiffiffiffiffiffiffiffi κ=Ca p at which viscosity becomes more effective than surface tension in damping the capillary-pressure fluctuations [3,7,10].For fast displacements, both capillarypressure and permeability fluctuations become relevant [7].In contrast to the dynamics observed in a compact porous medium with characteristic pore size bursts [24], the interfacial dynamics in our disordered confined medium consists of power-law distributed avalanches, both in sizes and durations.Their cutoffs imposed by the finite correlation length l c diverge for v → 0 [8], signaling the presence of a critical depinning transition [10].Within the range of parameters explored in the present experiments, corresponding to slow displacements, the lateral correlation length took values in the range 3 < l c < 15 mm, much larger than the spatial resolution of the images and the characteristic length of the medium heterogeneities, l d ¼ 0.6 mm.The possibility of tuning independently l c and l allowed experimental studies of "finite-size scaling," a tool often used in theoretical analysis of critical phenomena [25].
Spatially localized avalanches, which may occur simultaneously along the interface, can lead to strong temporal fluctuations of the global advancement of the imbibition front.In order to study the fluctuations of the spatially averaged (global) velocity V l ðtÞ, and specifically its intermittent character, we performed a multiscale analysis of the signal with the multifractal formalism introduced by Parisi and Frisch for fully developed turbulence [13].We examined the statistical properties of the velocity increments ΔV l ðτÞ ¼ V l ðt þ τÞ − V l ðtÞ for various time delays τ and length scales l.First, we present the results corresponding to a given set of conditions: v ¼ 0.053 mm=s and μ ¼ 50 cP, with a measuring window l ¼ L=8 ¼ 17 mm (≃1.6l c ). Figure 2 (top panel) shows the distributions of the normalized velocity increments Y ¼ ðΔV l − hΔV l iÞ=σ for increasing time lags τ. hΔV l i stands for the ensemble average of ΔV l and σ for its standard deviation.Interestingly, we observe that the shape of these distributions evolved through the temporal scales τ.For short time lags, we measured distributions with exponential tails and a peak at Y ¼ 0 (zero acceleration).This shape progressively evolved towards nearly Gaussian distributions at longer time lags.This result unveils that the dynamics of the invading front is highly intermittent, with complex temporal correlations on short time intervals, which we quantified through the analysis of the fluctuations of V l ðtÞ.The fluctuations are locally described by the singularity exponents hðtÞ given by ΔV l ðτÞ ∼ τ hðtÞ as τ → 0 þ .Nonlocal singularities can be characterized by the p-order structure functions S p ðτÞ ≡ hjΔV l ðτÞj p i, from which the spectra of global exponents ξ p can be defined as S p ðτÞ ∼ τ ξ p .Homogeneous fluctuations [hðtÞ ¼ H for all t] lead to ξ p ∼ p, while nonhomogeneous temporal fluctuations whose singularity exponents vary with time-a hallmark of intermittency-result in a nonlinear dependence of ξ p with p [26,27].In order to unveil deviations from monofractality and Gaussian statistics, we computed the normalized p-root structure functions C N p ðτÞ ≡ C p =R G p , where C p ≡ S 1=p p and the normalizing factor is the ratio of structure functions for a Gaussian distribution, which depends only on p [28].The bottom panel of Fig. 2 shows a double-logarithmic plot of C N p ðτÞ for p ¼ 1; …; 5.At short time lags, C N p ∼ τ ξ p =p over more than one decade.The inset shows that ξ p depends nonlinearly on p, reflecting the intermittent dynamics of V l ðtÞ.At large τ, C N p ðτÞ collapsed onto a single curve for all p as expected for Gaussian statisticsobserved in the top panel of Fig. 2.
Intermittency was observed for all the experimental conditions explored.But, more interestingly, we noticed that the tails of the distributions of velocity increments for a given time lag τ were systematically larger in some limitsthus, revealing a stronger intermittent character.Larger tails were observed when either V l ðtÞ was measured at smaller length scales l [29] or the imposed flow rate (and hence, v) or the oil viscosity μ were smaller, as shown in Fig. 3. To quantify this systematic evolution, we computed the kurtosis K (a measurement of the relative importance of the tails) and skewness Sk (a measurement of the asymmetry) of the ΔV l distributions [30].The bottom panel of Fig. 4 shows the systematic decrease of K for various l towards a plateau at large time lags, for a given set of experiments performed at v ¼ 0.13 mm=s and μ ¼ 50 cP.For measuring windows larger than the lateral extent of correlated motion l > l c , K reached the expected Gaussian value K G ¼ 3, but it was significantly larger for l < l c due to finite-size effects.The decrease of the kurtosis at short τ could be approximated by a power law K ∼ τ −α , reflecting the nonsimilarity of the statistical distributions at different time lags, and thus, once again, intermittent temporal fluctuations of the global front velocity up to a characteristic time τ c , as observed previously in Fig. 2. The exponent α measuring the intensity of the intermittent character of the velocity signal [20] increased systematically as l approached l c and saturated to a nearly constant value for l ≤ l c , as shown in the inset of Fig. 4. It is worth noting that, in addition to being heavy tailed, the distributions for short τ had a small positive asymmetry, which decreased systematically with l, as shown in the top panel of Fig. 4.
To compare the various experiments performed in very different conditions, and avoid finite-size effects due to the correlation length scale l c (and specifically, its evolution with v and μ), we carried out the same analysis for spatially averaged velocity V l measured at scales l such that l=l c FIG. 3   was constant.This ratio provides an estimate of the effective number of statistically independent degrees of freedom of the invading front, which controls the non-Gaussian asymmetric shape of the distributions of velocity fluctuations [6].Interestingly, when measured at scales l ¼ l c and a given time lag τ, we observed that the distributions of velocity increments ΔV l ðτÞ collapsed into the same shape for experiments performed at different viscosities but the same velocity (left top panel in Fig. 5).However, in these conditions the shape of the distributions of the velocity increments still evolved systematically with the imposed flow rate (right top panel in Fig. 5).Such behavior is confirmed by the temporal evolution of the kurtosis K for V l measured at scales l ¼ l c , as shown in the bottom panels of Fig. 5.The evolution of K and, more specifically, the characteristic time delay τ c at which K reaches a plateau (close to the value K G ¼ 3) did not depend on oil viscosity, but τ c decreased systematically with imposed flow rate.Moreover, the intensity of the intermittent character of the velocity signal measured by the exponent α appeared to be the same for the various experimental conditions.Therefore, rescaling the time lag τ by the characteristic time scale τ c collapsed the kurtosis for all the experiments, as shown in the inset of the right bottom panel of Fig. 5.We found that τ c ∼ 1=v with a proportionality coefficient around 0.6 mm-independently of l, μ, v.This value coincides with l d , the mean extent of the disorder patches, and thus, τ c ¼ l d =v can be identified as the average time spent by the fluid front to advance over the typical distance l d .Interestingly, while local bursts of activity can extend on scales much larger than this characteristic length of the medium heterogeneities l d and last over time scales much longer than τ c , the latter determines the temporal range of intermittency of the global (spatially averaged) velocity signal.
Finally, this statistical analysis of ΔV l ðτÞ allowed us to reach the main result of our study, namely to identify the controlling parameters of the observed intermittent dynamics.Figure 6 shows the normalized statistical distributions of spatially averaged velocity increments ΔV l ðτÞ obtained under various experimental conditions (μ; v), but measured at length scales l and time lags τ such that both ratios l=l c and τ=τ c are fixed.The data collapse shows that the distributions of velocity increments depend specifically on those two parameters only.While the characteristic time τ c sets the temporal range over which intermittency is observed, the ratio l=l c controls its intensity.
To conclude, slow imbibition of a disordered confined medium (a laboratory model of an open fracture) showed all the characteristic features of intermittent dynamics.As shown in Fig. 1, periods of low velocities (slightly below the imposed mean front velocity v) alternated with periods of very large velocity excursions above v.In the former, the acceleration was small and strongly correlated to the velocity, while in the latter the acceleration fluctuated strongly.Such intermittent dynamics results from local bursts of fast cooperative motion, triggered from the smallest length scales of the medium heterogeneities up to the correlation length l c -the scale at which viscous pressure damping becomes more effective than surface tension, which depends upon capillary number and medium permeability.Both l c and τ c vanish at high flow rates.Hence, intermittency is present only in slow imbibition displacements in the capillary regime.By analogy with intermittency and the flux of energy across different scales in fully developed turbulence [13,20], we suggest that the positive asymmetry of the statistical distributions of velocity increments observed in our experiments could be the signature of an inverse cascade resulting from fluid incompressibility, which is at the origin of the correlation length l c [10].Extension of the analysis presented here to other problems of front propagation in random media such as fluid imbibition in compact porous media [31], contact line dynamics [32], and planar crack front displacements [33], could bring further insight into this question.Our model experiment of fluid front invasion at low Ca, and the systems mentioned, are very different from hydrodynamic or wave turbulence.Hence, we expect that the results presented in this Letter will motivate new theoretical and experimental approaches of intermittency.

FIG. 1 (
FIG. 1 (color online).Top: Sketch of the model disordered medium.It is formed by two parallel, closely spaced glass plates (G).Copper patches (Cu) of lateral size 0.4 × 0.4 mm 2 and 0.06 mm height, randomly deposited without overlapping, cover 35% of the bottom fiber glass plate (FG).Middle: Global velocity, V l ðtÞ, computed from an experiment with v ¼ 0.053 mm=s and μ ¼ 50 cP (l c ≃ 11 mm), observed on a scale l ¼ L=8 ¼ 17 mm.Bottom: Corresponding timederivative, A l ðtÞ.

FIG. 2 (
FIG. 2 (color online).Top: semilog plot of PðYÞ vs Y ¼ ðΔV l − hΔV l iÞ=σ for increasing time lags τ, shifted vertically for visual clarity, for experiments at v ¼ 0.053 mm=s and μ ¼ 50 cP, and analyzed at l ¼ L=8 ¼ 17 mm.The dashed curve represents a Gaussian pdf.Bottom: Corresponding p-root normalized structure functions C N p ðτÞ for p ¼ f1ð∘Þ; 2; 3; 4; 5ð▵Þg.The inset shows the evolution of the exponent ξ p with p; error bars are smaller than symbol size.

FIG. 4 (
FIG. 4 (color online).Semilog plot of the skewness (top) and log-log plot of the kurtosis (bottom) of PðΔV l Þ for various l as function of τ.The horizontal dashed line corresponds to K G ¼ 3.The inset shows the exponent α of the power-law decay of the kurtosis as a function of l=l c .In these experimental conditions l c ¼ 6.7 mm.

FIG. 5 (
FIG. 5 (color online).Semilog plots of the normalized distributions of velocity increments (shifted arbitrarily for different v for visual clarity) for experiments analyzed at l ¼ l c (top) and log-log plots of their corresponding kurtosis for all time lags τ explored (bottom).Left panels display results for different μ but fixed v. Right panels display the systematic evolution with v (from 0.036 to 0.55 mm=s) for a single μ.Vertical lines give the value of τ used to compute the distributions shown in the top panels.Inset: collapse of the kurtosis for all experiments analyzed at l ¼ l c when τ is rescaled by τ c ¼ l d =v.