Quantifying active diffusion in an agitated fluid

Pierre-Yves Gires , Mithun Thampi and Matthias Weiss *
Experimental Physics I, University of Bayreuth, Universitätsstr. 30, D-95447 Bayreuth, Germany. E-mail: matthias.weiss@uni-bayreuth.de

Received 7th July 2020 , Accepted 11th September 2020

First published on 11th September 2020

Mixing of reactants in microdroplets predominantly relies on diffusional motion due to small Reynolds numbers and the resulting absence of turbulent flows. Enhancing diffusion in microdroplets by an auxiliary noise source is therefore a topical problem. Here we report on how the diffusional motion of tracer beads is enhanced upon agitating the surrounding aqueous fluid with miniaturized magnetic stir bars that are compatible with microdroplets and microfluidic devices. Using single-particle tracking, we demonstrate via a broad palette of measures that local stirring of the fluid at different frequencies leads to an enhanced but apparently normal and homogenous diffusion process, i.e. diffusional steps follow the anticipated Gaussian distribution and no ballistic motion is observed whereas diffusion coefficients are significantly increased. The signature of stirring is, however, visible in the power-spectral density and in the velocity autocorrelation function of trajectories. Our data therefore demonstrate that diffusive mixing can be locally enhanced with miniaturized stir bars while only moderately affecting the ambient noise properties. The latter may also facilitate the controlled addition of nonequilibrium noise to complex fluids in future applications.

1 Introduction

In recent years, a multitude of physico-chemical studies on and with complex fluids has been based on small sample volumes, from the microliter down to the femtoliter range. Supposedly most prominent are microfluidics-based experiments that have been utilized for a large variety of topical problems in physics, chemistry, and biology (see ref. 1–4 for an overview). Especially in the life sciences, microdroplets have emerged as an important tool. In the realm of synthetic biology, for example, microdroplets of cell-derived and biomimetic fluids have been used to study gene transcription,5–7 immunoassays8 and directed evolution,9 or the self-organization of cytoskeletal filaments.10,11 In fact, studying dynamic self-organization and nonequilibrium thermodynamics of reconstituted biological and biomimetic systems via microdroplets has lately gained considerable momentum (see, for example, ref. 12 for a recent review).

Despite the frequent use of microdroplets, an inherent limitation of such tiny sample volumes is often a poor, diffusion-based mixing since low Reynolds numbers prohibit any turbulence. A typical flow velocity of v = 100 μm s−1 for water in a microfluidic device with a typical dimension of L = 50 μm yields, for example, Re = ρvL/η < 0.01. The remaining laminar flow is hence well described by the Stokes equation and mixing of reactants, e.g. enzymes and substrate proteins, solely relies on diffusion. Making matters worse, diffusion in complex fluids, e.g. in cell extract droplets, is often slowed down or even anomalous due to (macromolecular) crowding.13,14 As a result, encounter rates for interaction partners are even further decreased, hence limiting the progress and timing of desired reactions. Facilitating diffusion processes in microdroplets by adding auxiliary fluctuations in a controlled fashion is therefore of high interest. Besides such a direct technical benefit, adding controlled active noise to complex fluids is also a promising tool to further probe predictions from stochastic thermodynamics15,16 or to study noise-dependent changes in emergent phenomena, e.g. for the self-assembly of a mitotic spindle apparatus inside a microdroplet.11

Here we report on how the diffusional motion of tracer beads in aqueous microdroplets is enhanced upon agitating the fluid with micron-sized magnetic stir bars. In particular, using single-particle tracking data (SPT), we demonstrate via a broad palette of measures that locally stirring the fluid with an ensemble of stir bars at different frequencies leads to an enhanced but apparently normal and homogenous diffusion process of immersed beads: diffusional steps follow the expected Gaussian distribution and no ballistic or superdiffusive motion is observed, yet diffusion coefficients are increased up to twofold. Signatures of the local stirring are, however, clearly visible in the power-spectral density and in the velocity autocorrelation function of trajectories, indicating the addition of an auxiliary, non-white noise. Our data therefore demonstrate that local diffusive mixing can be enhanced with miniaturized stir bars while only moderately affecting the ambient noise properties. We envisage our stirring approach to be a versatile tool for microfluidic applications in which flow-driven mixing is impractical, e.g. for isolated microdroplets. In addition, the observation of only modest changes in the diffusion characteristics may be of significance for studying nonequilibrium noise under well-defined conditions.

2 Materials and methods

2.1 Sample preparation and stirring

Magnetic stir bars (SBs) were synthesized as described before in detail.17 In brief, SBs were assembled from a suspension of spherical Fe3O4 nanoparticles (diameter 30 nm; SOR-30, Ocean NanoTech LLC). After ligand exchange, hydrophilized nanoparticles were aligned in solution by a static magnetic field, imposed by a neodymium magnet in a custom-made holder. Particles were subsequently connected permanently by a silica coating, hence fixing the magnetic axes and eventually yielding ferrimagnetic rods. Electron microscopy revealed that all SBs had a slender, rod-like shape: typical diameters were in the range ∼300 nm and lengths varied from few hundred nanometers up to 20 μm, following roughly an exponential profile with a mean length of approximately 4 μm.

Starting from purified SBs, the sample was produced by mixing 8 μl of the stir-bar suspension with 1.5 μl water and 0.5 μl of a suspension of fluorescent beads (200 nm FluoSpheres carboxylate, yellow-green 505/515, 2% solid from Invitrogen; diluted to a ratio 1[thin space (1/6-em)]:[thin space (1/6-em)]104 and sonicated for 15 min). For each experiment, 0.4 μl of this solution was pipetted into a 4 μl squalene droplet between two 15 mm coverslips using a spacer made from four layers of magic tape (thickness of 50 μm each) and one double side tape (height of 87.5 μm). The sandwiched sample was then mounted on a custom-made aluminium holder and positioned in the center of two pairs of custom-made Helmholtz coils with which SBs were addressed (see ref. 17 for all relevant details). A custom-made microcontroller enabled us to tune the alternating magnetic field in the frequency range f ∈ [0,10] Hz with an adjustable magnetic field strength (here chosen to be 0.5 mT). A schematic representation of the resulting sample is shown in Fig. 1.

image file: d0cp03629c-f1.tif
Fig. 1 Schematic representation of an aqueous droplet with magnetic stir bars and fluorescent beads (sample thickness h = 290 μm, diameter d = 620 μm). Due to an alternating magnetic field B, stir bars undergo a synchronous rotation with angular frequency ωB while fluorescent beads show a vivid diffusional motion (see supplementary movies in the ESI, for an example).

2.2 Imaging and tracking

Imaging was performed on an inverted epifluorescence microscope (Leica DMI6000B) using a 10× objective (HC PL APO 10×/2.3 DRY) and a 1392 × 1040 pixel camera (Leica DFC360FX), yielding a pixel pitch of 645 nm. The evaluation of images with a tracking algorithm (see below) allowed for determining particle positions, i.e. fluorescence peak positions, with subpixel resolution (position uncertainty about 80 nm). Fluorescent beads were illuminated using a 480 ± 20 nm filter, the fluorescence detection range was 527 ± 15 nm. The frame time was Δt = 118 ms.

Images of moving fluorescent beads were analyzed with FIJI. Images were first subjected to a background correction (subtraction of the mean pixel intensity from a homogeneous region around the droplet). Then, for improved particle detection, a Gaussian filter (radius 0.65 pixels) was applied and the contrast was adjusted so that 0.3% of all pixels were saturated. These images were then evaluated with the FIJI plugin TrackMate using a blob diameter of 3.6 μm, a threshold of 15 and a median filter. The maximum linking distance and gap closing were set to 5 μm with zero frame gap. If aggregates were observed (by visual inspection), these were removed from the list of trajectories. For each frequency at least 246 trajectories, each with a minimum of N = 200 positions, were used for further analysis. For each condition, trajectories were cut to the same length for all analyses.

3 Results and discussion

In order to supply aqueous microdroplets with a controlled auxiliary noise, we used custom-made magnetic stir bars with a width of few 100 nm and lengths in the range of few micrometers (see ref. 17 for a detailed characterization). As shown before, these slender stir bars can be addressed by an alternating magnetic field with tunable field strength and frequency f ∈ [0,10] Hz (see Materials and methods as well as ref. 17 for details). So far, however, the effects of local stirring on diffusive motion had not been quantified. To close this gap, we immersed fluorescent beads of 100 nm radius in the microdroplet and followed their motion by single-particle tracking (SPT) in the imaging plane (see Materials and methods). A schematic representation of the sample is shown in Fig. 1. Representative movies of a large ensemble of rotating stir bars and of the vivid motion of fluorescent beads are provided in the ESI.

Particles were tracked in the absence of any stirring (f = 0 Hz) and while stir bars were agitating the fluid at a stirring frequency in the range f ∈ [0.01,10] Hz. For each frequency, more than 200 trajectories were acquired and analyzed. In particular, for each trajectory r(t) with N ≥ 200 positions and a sampling time Δt = 118 ms we first calculated the time-averaged mean square displacement (TA-MSD)

image file: d0cp03629c-t1.tif(1)
where τ = kΔt denotes the lag time. In addition to TA-MSDs, we also considered the ensemble average of all TA-MSDs, 〈r2(τ)〉t,E, at a given stirring frequency f.

In general, a scaling law

r2(τ)〉 = α(2)
is expected for both, 〈r2(τ)〉t and 〈r2(τ)〉t,E, where 0 < α < 2 reflects the type of diffusional motion. Besides normal Brownian diffusion (α = 1) also sub- and superdiffusive motion is captured by this expression (α < 1 and α > 1, respectively) whereas ballistic motion leads to α = 2. The generalized transport coefficient, K, boils down to the familiar diffusion constant for α = 1 (with a dimension-dependent constant prefactor).

Using eqn (2), we have extracted K and α from individual TA-MSDs by linear regression of log(〈r2(τ)〉t) vs. log(τ). Since eqn (1) relies on finite averages at every lag time τ, we have restricted this fit process to τ ≤ 10Δt; for larger lag times, growing statistical fluctuations due to insufficient averaging emerge that eventually can spoil the fit process. Therefore, parameters derived from TA-MSDs essentially report on short-time transport characteristics. As a result, we obtained for each stirring frequency f the experimental probability density functions (PDFs) p(α) and p(K) for the associated ensemble of trajectories. Mean values of these PDFs are named 〈α〉 and 〈K〉 in the following. Due to the restriction of the fitting procedure to early lag times, 〈α〉 and 〈K〉 encode only the short-term characteristic of trajectories. As a complement, we therefore also fitted for each stirring frequency the ensemble-averaged TA-MSD, 〈r2(τ)〉t,E. Here, the fitting process included all available lag times since the ensemble averaging smoothed out the fluctuations of individual TA-MSDs. The resulting parameters, named αe and Ke, hence encode the characteristic features over all available time scales.

In the absence of stirring (f = 0 Hz), TA-MSDs and their ensemble average showed the anticipated normal diffusion behavior 〈r2(τ)〉 ∼ τ (Fig. 2a) with individual TA-MSDs exhibiting significant fluctuations around the mean scaling due to the finite time series length.18 These fluctuations resulted in random variations of individual scaling exponents, reflected in the width of p(α) around its mean, 〈α〉 = 1.03 (Fig. 2a, inset). Analyzing the associated 〈r2(τ)〉t,E yielded αe = 1.02, i.e. both measures are in very good agreement with the expected value α = 1. It is worth noting that minor deviations from unity are frequently observed in SPT experiments (see discussion in ref. 18 and 19).

image file: d0cp03629c-f2.tif
Fig. 2 (a) Representative TA-MSDs of fluoresecent beads, 〈r2(τ)〉t (grey lines), and the ensemble average of all TA-MSDs, 〈r2(τ)〉t,E (black circles), show normal diffusion (indicated by black dashed line) when no stirring is applied (f = 0 Hz). Inset: Fitting individual TA-MSDs with eqn (2) yields a fairly narrow probability density function of scaling exponents, p(α), with a mean 〈α〉 = 1.03, consistent with normal diffusion. The width of p(α), i.e. the standard deviation σ ≈ 0.08, is mostly determined by the finite length of trajectories (cf. discussion in ref. 18). (b) Representative ensemble-averaged TA-MSDs, 〈r2(τ)〉t,E, at the indicated stirring frequencies all show an approximately normal diffusive behavior, yet with varying prefactors Ke. Especially for f = 1 Hz, also a stirring-induced bump is visible. See main text for a detailed discussion.

Being consistent with normal diffusion, the generalized transport coefficient for the acquired ensemble of two-dimensional trajectories was expected to be K = 4D0 = 9.2 μm2 s−1, where D0 ≈ 2.3 μm2 s−1 is the fluorescent beads' diffusion constant in water at room temperature according to the Einstein–Stokes equation. This expectation is nicely confirmed by the experimentally determined values 〈K〉 ≈ 9.8 μm2 s−1 and Ke ≈ 9.5 μm2 s−1. Again, residual uncertainties on the precise value of K are frequently observed in several measurement techniques, including SPT.19

While the data for unstirred fluids confirmed, as expected, a normal diffusion with the predicted diffusion constant, any stirring action of stir bars was initially expected to enforce a superdiffusive or even ballistic signature in the MSD (1 < α ≤ 2), at least for short and possibly intermediate time scales. The observation of a synchronous rotation of stir bars (ESI, movie 1) even enforces this expectation. Yet, upon stirring the system, all curves for 〈r2(τ)〉t,E retained an almost linear scaling (Fig. 2b) albeit with an elevated prefactor, Ke. Notably, the variation of Ke was non-monotonous as 〈r2(τ)〉t,E for f = 5 Hz assumed values between the respective curves for f = 0 Hz and f = 1 Hz. Moreover, a pronounced bump in 〈r2(τ)〉t,E at early lag times appeared for f = 1 Hz due to the stirring.

To explore this in more detail, we next determined the mean scaling exponents and transport coefficients at each stirring frequency (i) via the PDFs p(α) and p(K) of TA-MSDs, yielding 〈α〉 and 〈K〉, and (ii) by fitting the ensemble-averaged MSD, yielding αe and Ke. As a result, we observed that αe only showed minor deviations from unity for all driving frequencies whereas 〈α〉 showed a significant excursion for intermediate values of f (Fig. 3a). This excursion of the short-time scaling exponent 〈α〉 is directly connected to the aforementioned visible bump in MSDs (cf.Fig. 2b), whose position τB was seen to migrate towards smaller lag times as τB ∼ 1/f for increasing f.

image file: d0cp03629c-f3.tif
Fig. 3 (a) In agreement with the qualitative impression from Fig. 2b, scaling exponents αe from 〈r2(τ)〉t,E (black circles) only show minor deviations from unity for all stirring frequencies. In contrast, extracting 〈α〉 from the PDF of individual TA-MSDs (red open diamonds) highlights a significant excursion at f ≈ 1 Hz. This excursion is related to a progressive shift of the bump in MSDs towards shorter lag times for increasing driving frequencies to which fitting of TA-MSDs is more sensitive than the ensemble-averaged 〈r2(τ)〉t,E. (b) Associated generalized transport coefficients, Ke (black circles) and 〈K〉 (red open diamonds), normalized to the respective values without stirring (denoted by superscript ‘(0)’), both show an almost twofold increase for intermediate driving frequencies that subsides for larger frequencies. The anticipated linear growth 〈K〉 ∼ f is well reproduced in the range f < 0.5 Hz (full red line, y = 1 +f with y = 〈K〉/〈K(0)〉 and 〈K(0)〉 denoting the value at f = 0 Hz). For frequencies f ≥ 0.6 Hz a power-law decrease due to tilting of stir bars (y = 1 + 0.5/f, red dashed line) is also in qualitative agreement with the expectations. See main text for details. Error bars indicate the standard deviation of the mean for 〈α〉 and 〈K〉, and the 95% confidence bounds of the fitting parameters αe and Ke.

Complementing this analysis, we also calculated the temporally varying apparent scaling parameter of the ensemble-averaged TA-MSDs, defined by α(τ) = d[thin space (1/6-em)]ln〈r2(τ)〉t,E/d[thin space (1/6-em)]ln[thin space (1/6-em)]τ. Please note that any meaningful power-law scaling needs to be defined over at least one order of magnitude, i.e. local variations in α(τ) rather reflect fluctuations that are superimposed to the general scaling exponent αe shown in Fig. 3a. Following previous work,20 we approximated the logarithmic derivative by a polynomial to obtain a smoothly varying function α(τ). As expected from our above analysis, marked fluctuations around a mean of approximately unity were visible in α(τ), especially for driving frequencies in the range f ∼ 1 Hz (see representative data in Fig. S1 in the ESI). These local variations, superimposed to the mean scaling exponent αe, arise mostly from the transient bump in MSDs (cf.Fig. 2b). Although one might expect a superdiffusive motion for very short time scales, neither MSDs nor α(τ) revealed consistent signatures for this, supposedly due to a too poor temporal resolution Δt.

For the generalized transport coefficients we also observed significantly increased values for intermediate values of f, this time for both, 〈K〉 and Ke (Fig. 3b). As compared to the case without any stirring, an almost two-fold larger mean transport coefficient was seen for intermediate stirring frequencies which eventually subsided for large f. Hence, despite an almost synchronous rotation of stir bars, the apparent mode of motion stayed roughly normally diffusive, but the apparent diffusion constant was significantly elevated due to the ambient stirring at intermediate driving frequencies. Considering the fluctuation–dissipation theorem, i.e. the Einstein–Stokes equation D = kBT/6πηR, this experimental finding indicates that the stirring adds an ambient active (and non-white) noise with an energy of the order kBT to the mere thermal motion of beads. The effective noise temperature experienced by beads is therefore up to twice the thermal energy.

The non-monotonous variation of 〈K〉 and Ke is at first glance surprising. However, inspecting the motion of stir bars in more detail reveals the underlying reason: balancing the magnetic and friction-induced torques leads to a progressive tilting of the rod-like stir bars with respect to the imaging plane for f > 0.5 Hz at the used magnetic field strength.17 Bearing in mind this experimental finding, and following previous arguments21 that consider only the mean distance between tracers and adjacent vortices (i.e. stir bars), one can obtain an approximate prediction of how 〈K〉 should vary with f by considering measured values of K to be proportional to the Peclet number,

image file: d0cp03629c-t2.tif(3)

Here, f = 1/T while [small script l]p denotes the stir bar's projected length in the imaging plane and D0 is the beads' thermal diffusion constant without any driving. For low driving frequencies, no tilting occurs and [small script l]p is simply the constant total length of the stir bar. Therefore, KPf is expected for frequencies f ≤ 0.5 Hz, which matches well the experimental data (see Fig. 3b). For larger frequencies, stir-bar tilting leads to a reduction of [small script l]p. Approximating the decrease as [small script l]p ∼ 1/f for f ≥ 0.5 Hz predicts KPf[small script l]p2∼ 1/f, which agrees qualitatively with the experimental data for 〈K〉 (Fig. 3b).

Aiming to retrieve even more information from MSDs, we also probed the so-called ‘ergodicity breaking parameter’,22,23E(τ) = 〈〈r2(τ)〉t2E/〈r2(τ)〉t,E2 − 1, that summarizes how much individual TA-MSDs fluctuate around their ensemble mean. For normal Brownian motion, E(τ) = 4τ/(3NΔt) is analytically predicted, whereas simulations reveal some finite-size deviations towards slightly larger values (ESI, Fig. S2). Without stirring, our experimental data were in favorable agreement with this expected behavior (ESI, Fig. S2); minor deviations for short lag times supposedly are remnants of a residual dynamic localization error.24 Upon stirring with a frequency in the range f ∼ 1 Hz, however, marked deviations over almost the entire range of lag times emerged (ESI, Fig. S2). Similar to transport in mucin hydrogels,25E(τ) assumed consistently larger values than predicted for normal diffusion at thermal equilibrium. Following this earlier report, we have used an empirical fit of the form E(τ) ∼τε to quantify these deviations. As a result, we observed that very low and very high stirring frequencies featured ε ≈ 1, whereas the intermediate regime f ∼ 1 Hz, in which an enhanced diffusion was observed, yielded values ε ≈ 0.7 (ESI, inset of Fig. S2). Therefore, stirring at intermediate frequencies induces a markedly higher spreading of individual TA-MSDs with respect to their ensemble average.

To further explore to which extent the apparently normally diffusive, but still driven trajectories deviate from mere thermal processes, we next inspected the PDF of normalized step increments. Previous reports have highlighted that spatiotemporally heterogeneous random walks, e.g. an intermittent switching between faster and slower diffusion (near and far away from a rotating stir bar), lead to characteristic signatures in this quantity,26–30 suggesting that the stirring could be detectable in this measure. We therefore calculated for each trajectory the time series of successive increments, Δxi = xi+1xi and Δyi = yi+1yi, and normalized these by their respective root-mean-square step length. For a homogenous normal random walk, the PDF of these normalized increments, p(χ), should follow a standard Gaussian. Since we did not observe systematic differences between x- and y-directions nor did we observe any asymmetry between positive and negative increment values, we combined all normalized increments into a single set and only considered their modulus. The resulting PDF, p(|χ|), also known as van Hove function,13 did not show any significant deviations from the expected Gaussian for any driving frequency (Fig. 4a). Performing the same analysis for increments taken within periods 2Δt, 3Δt etc. also did not show consistent deviations from a Gaussian (apart from the anticipated lower statistics). Nonetheless, it is tempting to speculate that image acquisition with higher frame rates, combined with a massive increase of trajectory data, could reveal diffusion heterogeneities that have been predicted for random-diffusivity models.31,32

image file: d0cp03629c-f4.tif
Fig. 4 (a) The PDF of the moduli of normalized increments, p(|χ|), follows the anticipated standard Gaussian (indicated by grey area) for all driving frequencies (color-coded). Therefore, no significant diffusional heterogeneity is observed due to stirring. (b) The autocorrelation of normalized squared increments, G(τ), hardly deviates from zero for all driving frequencies (color-coded). Comparing these experimental data to simulations (see ESI) suggests that particles experience a continuously varying, rather than a dichotomously switching, mobility along their (random) trajectory in the array of rotating stir bars. See main text for more details and discussion.

To complement the analysis via p(|χ|) and to explore in more detail whether stirring induces a varying mobility like in random-diffusivity models, we analyzed the ensemble-averaged autocorrelation function of fluctuations in the squared increments Δr2(t) = |r(t + Δt) − r(t)|2, defined as

image file: d0cp03629c-t3.tif(4)

If there is a change between two transport processes, a decay G(τ) ∼ exp(−λτ) is expected, with the characteristic rate λ being determined by the switching rates between the different modes of motion.29,33 A meaningful amplitude of the autocorrelation function is given by G(2Δt) since even homogenous random walks yield a nonzero value for Gt).29 Inspecting G(τ) for our experimental data revealed an almost vanishing amplitude of the autocorrelation function for all stirring frequencies and lag times (Fig. 4b). Moreover, the low values of G(τ) were plagued by strong fluctuations, which were even stronger when using individual trajectories. Disregarding the fluctuations in G(τ), still a slight dependence of the small but nonzero correlation amplitude on the stirring frequency was observable. We hypothesized that this can be explained by assuming particles to undergo a simple thermally driven diffusion far away from stir bars while diffusing with an enhanced mobility when stochastically entering the vicinity of a rotating stir bar. The enhancement factor for the mobility would then be determined by the stirring frequency as shown in Fig. 3b. Performing simulations along these lines (see ESI), we observed indeed that the autocorrelation amplitude increases with the enhancement factor. Yet, a dichotomous stochastic switching between two diffusion coefficients resulted in amplitudes of G(τ) that were at least fivefold larger than the experimentally observed ones (ESI, Fig. S3). However, allowing for a continuously varying diffusion coefficient resulted in markedly lower amplitudes that were consistent with our experimental data (ESI, Fig. S3). Thus, our experimental trajectories most likely do not encode a dichotomous stochastic switching between just two different modes of motion (e.g. far away and near to a stir bar), but rather they report on a continuously varying mobility along the (random) trajectory in the array of rotating stir bars.

As an additional step, we have also probed the power-spectral density (PSD) of individual trajectories, defined as

image file: d0cp03629c-t4.tif(5)
where the index k encodes x- and y-coordinates, and T = NΔt is the maximum time covered in the trajectory. Similar to TA-MSDs, also S(ν) shows fluctuations around the ensemble mean, 〈S(ν)〉E (Fig. 5a, inset). For the case of (fractional) Brownian motion processes, analytical predictions for 〈S(ν)〉E and for the fluctuations of individual S(ν) have recently been derived.34,35 In particular, a power-law scaling 〈S(ν)〉E ∼ 1/ν1+α is expected while normalized fluctuations of individual S(ν) around this average have universal PDFs that only depend on the dimensionality. Indeed, the ensemble-averaged PSD shows the anticipated scaling 〈S(ν)〉E ∼ 1/ν2 for all driving frequencies f (Fig. 5a). Notably, for intermediate stirring frequencies, an isolated peak at ν = f is seen in the PSD, i.e. a clear signature of the stirring is observed in this measure. Disregarding this particular frequency and collecting the fluctuations of S(ν) around 〈S(ν)〉E in the range ν ∈ [0.1,1] Hz for (i) only the x-coordinates, yielding normalized fluctuations A1, or (ii) for both coordinates, yielding A2, also shows a good agreement with the predicted universal PDFs p(A1) = exp(−A1) and p(A2) = A2[thin space (1/6-em)]exp(−A2) (see Fig. 5b). Deviations of p(A2) in the vicinity of the maximum are similar to previously published data on beads diffusing in agarose (cf.Fig. 3f in ref. 35), suggesting that A2 is more sensitive to additional noise in experimental data than A1. Including also data for νf was seen to lead to strong deviations, e.g. in a marked shift of the peak in p(A2) towards smaller values of A2.

image file: d0cp03629c-f5.tif
Fig. 5 (a) The ensemble-averaged power spectral densities (PSDs) at the indicated stirring frequencies f follow the prediction of normal Brownian motion (S(ν) ∼ 1/ν2). Additional, pronounced peaks at ν = f are seen for f = 0.5 Hz and f = 1.5 Hz, revealing that the system has been stirred. Inset: PSDs of individual trajectories (grey lines, f = 0 Hz) show marked fluctuations around the ensemble-average (black line). (b) Quantifying and normalizing these fluctuations for different stirring frequencies (color-coded as before) yields a very good agreement with analytical predictions for one- and two-dimensional trajectories (left and right plot, respectively). See main text for details.

Having observed that the PSD can capture at least part of the active stirring, we finally also inspected whether the normalized, ensemble- and time-averaged velocity autocorrelation function (VACF) is sensitive to the additional noise that is introduced by locally stirring the fluid. Being defined as

image file: d0cp03629c-t5.tif(6)
the VACF is predestined to report sensitively on altered random walk properties.36 Here, v(t) = [r(t + δt) − r(t)]/δt is the instantaneous velocity at time t, determined via the traveled distance within a period δt. As expected, a Markovian signature is observed in the VACF for f = 0 Hz (Fig. 6, black symbols), i.e. starting from unity a rapid decay to zero is seen for τ ≤ δt. This result reflects a memory-devoid normal random walk.36 For f > 0 Hz, however, clear oscillatory correlations are observed for larger time scales (Fig. 6, red and blue symbols). Unlike the ensemble mean, VACFs of individual trajectories show quite some fluctuations (ESI, Fig. S4), indicating that particles do not simply periodically rock back and forth but rather have a random additive component that is averaged to zero in Cv(τ). Retrieving the angular frequency ω0 of the VACF oscillation reveals an excellent agreement with the angular turning frequencies of stir bars in the range 1 rad s−1 ≤ 2πf ≤ 20 rad s−1 (Fig. 6, inset). Therefore, particles seem to undergo an almost normal diffusion with an elevated transport coefficient with the active noise from stirring being mostly encoded in the VACF.

image file: d0cp03629c-f6.tif
Fig. 6 Representative VACF curves [eqn (6) with δt = 5Δt] without stirring (black circles) and with stirring at the indicated frequencies (blue squares and red diamonds). As expected, no memory is observed in the VACF for the unstirred case (i.e. zero correlation for τ > 5Δt) whereas oscillatory correlations are observed upon stirring. Inset: The VACFs oscillation frequencies ω0 = 2π/T are in excellent agreement with the stir bars' angular driving frequencies ω = 2πf in the range 1 rad s−1ω ≤ 20 rad s−1.

In summary, we have reported here on how rotating miniaturized magnetic stir bars, embedded in aqueous microdroplets, can enhance the local diffusion of mesoscopic objects with a typical size on the 100 nm scale. Single-particle tracking on fluorescent beads allowed us to reveal that the diffusion characteristics remained almost normal, yet with an elevated diffusion coefficient and clear signatures of the stirring process in the power-spectral density and in the velocity autocorrelation function. At first glance, tracer motion in the array of almost synchronously rotating stir bars bears some similarities to reciprocal swimmers.37 Yet, for reciprocal swimmers the fluid experiences no net force and torque, whereas stir bars in our experiments clearly induce rotlets throughout the fluid, i.e. particles stochastically explore an array of vortices that locally accelerate them.

Due to the onset of an out-of-plane precessing of the stir bars (in our example at a stirring frequency f ∼ 1 Hz), only a twofold enhancement of diffusive transport could be achieved. The effective noise temperature applied to the fluid and picked up by the beads was therefore only up to twice the actual thermal temperature. It is worth emphasizing here that the introduced noise is by far not uncorrelated and white since our stir bars rotated almost synchronously. A simple transfer to the classical fluctuation–dissipation relation and naming a meaningful effective noise temperature on this basis is therefore only a crude way of characterizing the noise. This caveat may be addressed by decorrelating stirrers to achieve a turbulent state (see ref. 38 for an example).

In future applications, such as chemical synthesis in microdroplet reactors or in agitated non-equilibrium systems, a further enhancement of the introduced noise, besides a potentially desired decorrelation of stirrers, may be achieved by applying higher magnetic field strengths since the instability onset scales linearly with this parameter.39 Thus, a further enhancement of mixing by agitated diffusion in microdroplets and/or a controlled addition of nonequilibrium fluctuations with a tunable range are possible.

Conflicts of interest

There are no conflicts to declare.


The authors gratefully acknowledge financial support by the VolkswagenStiftung (Az. 92738).


  1. A. Theberge, F. Courtois, Y. Schaerli, M. Fischlechner, C. Abell, F. Hollfelder and W. Huck, Angew. Chem., Int. Ed., 2010, 49, 5846 Search PubMed.
  2. X. Fan and I. M. White, Nat. Photonics, 2011, 5, 591 Search PubMed.
  3. R. Seemann, M. Brinkmann, T. Pfohl and S. Herminghaus, Rep. Prog. Phys., 2012, 75, 016601 Search PubMed.
  4. E. K. Sackmann, A. L. Fulton and D. J. Beebe, Nature, 2014, 507, 181 Search PubMed.
  5. E. Sokolova, E. Spruijt, M. M. K. Hansen, E. Dubuc, J. Groen, V. Chokkalingam, A. Piruska, H. A. Heus and W. T. S. Huck, Proc. Natl. Acad. Sci. U. S. A., 2013, 110, 11692 Search PubMed.
  6. M. M. K. Hansen, L. H. H. Meijer, E. Spruijt, R. J. M. Maas, M. V. Rosquelles, J. Groen, H. A. Heus and W. T. S. Huck, Nat. Nanotechnol., 2016, 11, 191 Search PubMed.
  7. N. N. Deng, M. A. Vibhute, L. Zheng, H. Zhao, M. Yelleswarapu and W. T. S. Huck, J. Am. Chem. Soc., 2018, 140, 7399 Search PubMed.
  8. J. U. Shim, R. T. Ranasinghe, C. A. Smith, S. M. Ibrahim, F. Hollfelder, W. T. Huck, D. Klenerman and C. Abell, ACS Nano, 2013, 7, 5955 Search PubMed.
  9. A. Zinchenko, S. R. A. Devenish, B. Kintses, P.-Y. Colin, M. Fischlechner and F. Hollfelder, Anal. Chem., 2014, 86, 2526 Search PubMed.
  10. E. Abu Shah and K. Keren, eLife, 2014, 3, e01433 Search PubMed.
  11. M. C. Good, M. D. Vahey, A. Skandarajah, D. A. Fletcher and R. Heald, Science, 2013, 342, 856 Search PubMed.
  12. K. A. Ganzinger and P. Schwille, J. Cell Sci., 2019, 132, jcs227488 Search PubMed.
  13. F. Höfling and T. Franosch, Rep. Prog. Phys., 2013, 76, 046602 Search PubMed.
  14. M. Weiss, Int. Rev. Cell Mol. Biol., 2014, 307, 383 Search PubMed.
  15. U. Seifert, Rep. Prog. Phys., 2012, 75, 126001 Search PubMed.
  16. F. S. Gnesotto, F. Mura, J. Gladrow and C. P. Broedersz, Rep. Prog. Phys., 2018, 81, 066601 Search PubMed.
  17. P.-Y. Gires, M. Thampi and M. Weiss, Sci. Rep., 2020, 10, 10911 Search PubMed.
  18. M. Weiss, Phys. Rev. E, 2019, 100, 042125 Search PubMed.
  19. P. Struntz and M. Weiss, Phys. Chem. Chem. Phys., 2018, 20, 28910 Search PubMed.
  20. D. Ernst, J. Köhler and M. Weiss, Phys. Chem. Chem. Phys., 2014, 16, 7686 Search PubMed.
  21. M. N. Rosenbluth, H. L. Berk, I. Doxas and W. Horton, Phys. Fluids, 1987, 30, 2636 Search PubMed.
  22. W. Deng and E. Barkai, Phys. Rev. E: Stat., Nonlinear, Soft Matter Phys., 2009, 79, 011112 Search PubMed.
  23. M. Schwarzl, A. Godec and R. Metzler, Sci. Rep., 2017, 7, 3878 Search PubMed.
  24. M. Goulian and S. Simon, Biophys. J., 2000, 79, 2188 Search PubMed.
  25. A. G. Cherstvy, S. Thapa, C. E. Wagner and R. Metzler, Soft Matter, 2019, 15, 2526 Search PubMed.
  26. J.-H. Jeon, M. Javanainen, H. Martinez-Seara, R. Metzler and I. Vattulainen, Phys. Rev. X, 2016, 6, 021006 Search PubMed.
  27. W. He, H. Song, Y. Su, L. Geng, B. J. Ackerson, H. B. Peng and P. Tong, Nat. Commun., 2016, 7, 11701 Search PubMed.
  28. T. J. Lampo, S. Stylianidou, M. P. Backlund, P. A. Wiggins and A. J. Spakowitz, Biophys. J., 2017, 112, 532 Search PubMed.
  29. A. Sabri, X. Xu, D. Krapf and M. Weiss, Phys. Rev. Lett., 2020, 125, 058101 Search PubMed.
  30. W. Wang, F. Seno, I. M. Sokolov, A. V. Chechkin and R. Metzler, New J. Phys., 2020, 22, 083041 Search PubMed.
  31. M. V. Chubynsky and G. W. Slater, Phys. Rev. Lett., 2014, 113, 098302 Search PubMed.
  32. A. V. Chechkin, F. Seno, R. Metzler and I. M. Sokolov, Phys. Rev. X, 2017, 7, 021002 Search PubMed.
  33. J. Cao, Phys. Rev. E: Stat., Nonlinear, Soft Matter Phys., 2001, 63, 041101 Search PubMed.
  34. D. Krapf, E. Marinari, R. Metzler, G. Oshanin, X. Xu and A. Squarcini, New J. Phys., 2018, 20, 023029 Search PubMed.
  35. D. Krapf, N. Lukat, E. Marinari, R. Metzler, G. Oshanin, C. Selhuber-Unkel, A. Squarcini, L. Stadler, M. Weiss and X. Xu, Phys. Rev. X, 2019, 9, 011019 Search PubMed.
  36. R. Metzler, J.-H. Jeon, A. Cherstvy and E. Barkai, Phys. Chem. Chem. Phys., 2014, 16, 24128 Search PubMed.
  37. E. Lauga, Phys. Rev. Lett., 2011, 106, 178101 Search PubMed.
  38. G. Kokot, S. Das, R. G. Winkler, G. Gompper, I. S. Aranson and A. Snezhko, Proc. Natl. Acad. Sci. U. S. A., 2017, 114, 12870 Search PubMed.
  39. A. Ghosh, P. Mandal, S. Karmakar and A. Ghosh, Phys. Chem. Chem. Phys., 2013, 15, 10817 Search PubMed.


Electronic supplementary information (ESI) available. See DOI: 10.1039/d0cp03629c
These two authors contributed equally.

This journal is © the Owner Societies 2020