Self-similarity in the breakup of very dilute viscoelastic solutions

When pushed out of a syringe, polymer solutions form droplets attached by long and slender cylindrical filaments whose diameter decreases exponentially with time before eventually breaking. In the last stages of this process, a striking feature is the self-similarity of the solution shape near the end of the filament. This means that shapes at different times, if properly rescaled, collapse onto one universal shape. A theoretical description inspired by this similarity observation and based on the Oldroyd-B model was recently shown to disagree with existing experimental results. By revisiting these measurements and analysing the interface profiles of very diluted polyethylene oxide solutions at high temporal and spatial resolution, we show that they are very well described by the model.


Introduction
The formation of drops has become a paradigm for the study of singularities in fluid mechanics and beyond. The formation of a drop from an orifice leads to a new length scale, the diameter of the neck that connects the drop to the orifice, that goes to zero at a finite time when the drop breaks off from the orifice. As a result, pinch-off is described by a similarity solution that describes the time evolution and self-similar shape of the neck close to breakup. In the case of Newtonian fluids of both large and small viscosity, the neck diameter behaves like a power law as a function of the time to pinch-off. The interface is found to have a universal shape, and profiles at different times can be superimposed onto one another by rescaling the radial and axial coordinates by appropriate powers of the time distance to the singularity.
Beyond Newtonian fluids, much work has been dedicated to the formation of drops in non-Newtonian fluids such as polymer solutions. These fluids are characterized by a slow time scale λ on which the constituents relax. If one takes a dilute solution of a highmolecular weight polymer and tries to make a droplet (Middleman 1965;Goldin et al. 1969;Petrie & Denn 1976;Eggers 1997;Deblais et al. 2018), polymers become stretched in the extensional flow close to pinch-off, and long and slender filaments form in between drops, where previously power-law pinch-off would have been observed (Goldin et al. Figure 1. (a) Schematic of the experimental setup used to determine the interface shape of the polymer thread. A full-frame camera is combined with a flash light allowing short flash duration. The flash is triggered on the falling drop and a high accuracy delay line allows to shift the breakup event in time. (b) Typical photograph of a pendant drop of a very dilute polymer solution breaking from a syringe. A long tiny polymer thread connecting the two drops is formed. Scale bar is 1mm. (c) Self-similar thinning of the interface profiles obtained from the experimental work of Clasen et al. (2006) and the simulations of Turkoz et al. (2018). z0 is the axial location for which the profiles collapse best. The comparison reveals a discrepancy between the two. 1969; Bazilevskii et al. 1981;Entov & Yarin 1984;Wagner et al. 2004;Bhat et al. 2010;Clasen et al. 2006). In the case of a jet, this means a long sequence of almost circular drops form, connected by tiny threads, a phenomenon which has been called the "beadson-a-string" structure. In the case of a dripping faucet, one observes essentially the same phenomenon, except that here only a single drop forms, with perhaps a satellite drop in between (Wagner et al. 2004), connected to the faucet by a tiny thread, as seen in Fig. 1(a,b).
This and similar phenomena are often modelled using the so-called Oldroyd-B model (Bird et al. 1987), which describes the polymer relaxation with a single time scale λ. The stress is written as the sum of a polymeric contribution and that of the solvent.
The formation of filaments and their instabilities have by now become a benchmark problem for testing viscoelastic fluid mechanics (Anna & McKinley 2001;McKinley & Sridhar 2002;Furbank & Morris 2004;Suryo & Basaran 2006;Smith et al. 2010;Miskin & Jaeger 2012;Huisman et al. 2012). Instead of following a power law, the filament now thins exponentially with a rate set by the relaxation time of the polymer (Bazilevskii et al. 1981;Anna & McKinley 2001;Clasen et al. 2006). The profile is extremely uniform over the thread, and then merges smoothly with a neighboring drop on either side. It was proposed by Clasen et al. (2006) that the profile at this junction is once more a similarity solution, which yields a universal profile, if both the axial and the radial coordinate is rescaled with the thread radius, with an exponential rise toward the drop ( Fig. 1(c)). Clasen et al. (2006) were able to calculate the similarity profile using a lubrication approximation, in which the interface slope is assumed small; however, this assumption of small slopes is not satisfied throughout the profile. Indeed, while an experiment using long flexible polymers in a viscous solvent showed collapse to a selfsimilar profile, a comparison with the theoretical calculation revealed an axial length scale in the experiment that was about twice as short as the one obtained from lubrication theory ( Fig. 1(c)). At the time, this serious discrepancy in the self-similar profile was attributed to a failure of the lubrication approximation.
However, a recent full numerical simulation of the Oldroyd-B equations (Turkoz et al. 2018) showed the same discrepancy with the experimental data of Clasen et al. (2006), and rather agreed with the lubrication calculation ( Fig. 1(c)). This left the serious possibility that the source of the discrepancy was the Oldroyd-B model itself. One possibility was that the experimental fluid was described by more than a single length scale, the other that the finite extensibility of a real polymer has to be taken into account, as described for example by more elaborate models such as the FENE-P model (Bird et al. 1987), which also describes the shear-thinning behavior of a real polymeric fluid. More recently, the similarity theory of Clasen et al. (2006) was extended to a treatment of the full axisymmetric Oldroyd-B equations, and the similarity profile was calculated without any lubrication assumptions (Snoeijer et al. 2019). The robustness of these calculations was also underlined by the observation that the universal interface shape even holds for purely elastic filaments undergoing elastocapillary instabilities (Snoeijer et al. 2019;Eggers et al. 2020) that could be dubbed sausage-on-a-string instabilities (Mora et al. 2010). These findings suggest that the specific type of viscoelastic model for the polymer solution is not crucial for calculating the shape of the interface, and hence could not explain the discrepancy between theory and experiment.
To clear up these questions, we investigate the breakup of very dilute polymer solutions. We record the interface profiles of the filaments at an extremely high temporal and spatial resolution during the experiments, and compare the results with the newly developed similarity theory of Eggers et al. (2020), based on the full Oldroyd-B equations. Our results convincingly show that the experimental profiles all converge to a universal selfsimilar solution that in addition agrees excellently with theory.

Experiments
We experimentally study the extensional thinning and destabilization of filaments of long-chain polymer solutions in water at different concentrations (Fig. 2). The experiments are performed with polyethylene oxide (PEO) of a molecular weight (M w ) of 4±2×10 6 g mol −1 (Berman 1978) from Sigma-Aldrich (purity=0.98). Four concentrations C p between 10 and 120 wppm are obtained from successive dilutions of the initial batch. This range of concentrations is chosen to be well below the critical overlap concentration of polymer coils (Graessley 1980 (2017)). The concentrations we consider allow us to work in conditions that are very close to satisfying the Oldroyd-B model, as we will show below. A syringe pump supplies the polymer solutions to the needle tip (inner diameter h 0 =2 mm).
We used a Phantom V1 fast camera (frame rate 10,000 fps) to record the dynamics of the filament thinning at a high temporal resolution, and a full-frame camera (8256 × 4640 pix 2 , Nikon D850) equipped with a 5× microscope lens to obtain a very good spatial resolution of the polymer interface during its detachment. In order to be able to capture high-quality picture of the interface during the fast thinning (∼ms), we use a flash light (Vela one) with 1µs flash duration. The camera and the flash light are coupled to a trigger to which we can control the delay to the next breakup event with a very good accuracy. The delay between two flashes is controlled through a precise delay line (Digital Delay Generator, DG535 Stanford Research Systems) that delays the initial trigger pulse (the event is triggered electronically on each bottom edge of droplets falling in repetition) and allows a delay resolution from 5ps to 1000s. The limiting factor here is the flash duration of 1 µs of the flashlight itself. The shutter of the camera is open for a 'long' time during the flash, and so the read-out time of the camera is not a limiting step. This allow us to obtain a sequence of very highly resolved pictures (10 µm/pix). as one would obtain with a fast camera but with a much better resolution (see sequence of pictures in Fig. 3).
Typical results are shown in Figures 2 and 3. The thinning dynamics (Fig. 2) of PEO show an initial thinning similar to that of a low-viscosity Newtonian fluid. Subsequently, a very long and slender cylindrical filament is formed. In this elasto-capillary thinning regime, the dynamics slows down dramatically. Both the Oldroyd-B model and experiments show that in this regime the minimum neck radius h min as a function of time can be described as with λ 0 the longest relaxation time of the polymer solution (Anna & McKinley 2001;Amarouchene et al. 2001). In the range of (very) diluted concentrations studied here, λ 0 varies with concentration between ∼ 1 ms and 50 ms. Even though in the dilute limit a variation of the relaxation time is not predicted by theory, this is in fact commonly observed in experiments (Clasen et al. 2006). This is confirmed by plotting the dependence of the relaxation time λ 0 with the polymer concentration C p in figure 2(b) show that it follows a power law dependence as reported by Clasen et al. (2006) for dilute polymer solutions. Figure 3 shows a typical sequence of pictures of a droplet's interface, obtained using the combination of fast camera and triggered flash to increase the resolution. We use these photographs to extract the interface profile with a homemade algorithm in Matlab; this procedure allows determining the edges of the polymer solution as shown in Figure 3(f).

Results
In Figure 4(a) we show the evolution of the interface profile during filament thinning, near the drop that forms due to destabilization. Here, r is the filament radius and z the direction along the filament. In Figure 4(b), we plot the same profiles but rescaled by the minimum neck radius h min and with z 0 an adjustable parameter that represents the location for which the experimental profiles collapse best. As time progresses, the interface shape converges to a universal shape. Comparing this shape to the recent viscoelastic calculations using the Oldroyd-B model (Eggers et al. (2020))., we find an excellent agreement: the profiles converge to the same universal self-similar solution profile indicated by the black line. We quantify the threshold to the self-similar solution by measuring the distance ∆(t) of the experimental profiles to the self-similar curve at r/h min = exp(1). This quantity is shown in the inset of figure 4(b) and converges to a constant value at the moment where the elasto-capillary regime is reached. Since the onset is set by the elasto-capillary time, a dependence with the polymer concentration C p is expected. In Figure 4(c), we show the profiles after converging for different polymer concentrations. They fall onto each other, confirming the self-similarity of the polymer thread interface. Here, again, agreement with theory (black line) is excellent; this also confirm that the profile grows exponentially as discussed in details in Eggers et al. (2020).
Solving the constitutive equation for an Hookean dumbbell model, a quadratic dependence of the first normal stress difference N 1 on shear rateγ can be obtained (Bird et al. 1987): with T denoting temperature, ψ 1 is the first normal stress coefficient, n the number density of polymer molecules (Table 1) and k B Boltzmann's constant. The rheology measurements (Anton Paar, MCR 302) shown in Fig. 5(a) show that the increase in polymer concentration slightly influence the shear viscosity, in agreement with Einstein expansion, whereas it significantly affects the normal force ( Fig. 5(b)), from which we can extract the relaxation time of the solution λ 0,rheo = 7.4 ms (Lindner et al. 2003). This is a factor ∼ 2 below the value obtained with the relaxation time obtained from the pinch-off experiment (λ 0 = 14.5 ms). Such discrepancies have been already been reported before by Clasen et al. (2006). The primary reason for the discrepancy is that high molecular weight polymers invariably exhibit an significant polydispersity which may result in multiple timescales; the elongational flow is more sensitive to the longest time scale, whereas the shear flow probes an average time scale. We can nevertheless evaluate the ratio between these two time constants using the multimode Zimm model: where N is the total number of modes andσ is a measure of the hydrodynamics interaction andσ ≈ -0.4 in our dilute polymer solutions (Anna & McKinley 2001). This time quickly decays for the higher modes; evaluating the ratio between the two time scales for the five first modes gives a good approximation. In fact, this gives a value of λ 0,rheo /λ 0 ≈ 3 which is a good estimate of what we find experimentally and which is a value also reported experimentally by Liang & Mackley (1994).

Conclusions
In conclusion, we have studied the destabilisation of diluted polyethylene oxide (PEO) solution in water using a camera setup allowing us to visualize the droplet and filament shapes with very high resolution. We find that during destabilization of a polymer droplet initially attached to a capillary, the interface converges to a self-similar shape independent of time or polymer concentration, which agrees very well with the theoretical prediction of Eggers et al. (2020), based on the Oldroyd-B model. This contrasts with an earlier discrepancy, observed in Turkoz et al. (2018), between full numerical simulations of the Oldroy-B model and the earlier experiments of Clasen et al. (2006). Three main differences between our experiment and that of Clasen et al. (2006) can been mentioned that might explain the observed discrepancy. (i) In the work presented here we are now able to work at significantly higher spatial and temporal resolution than they could do at this time. (ii) The second important difference to note is the solvent used that is much more viscous than the one we use in our study (water) that also exhibit non-Newtonian behavior 'Boger fluid'. (iii) They used Caber device to impose an extensional deformation to their samples. For that purpose, they used endplates diameters up to 6mm (3 times