1. Introduction
The addition of only a small amount of long-chain flexible polymers into fluid flow can lead to fascinating phenomena very different from its Newtonian counterpart, among which the polymer drag reduction in wall-bounded turbulence is the most intriguing (Procaccia, L'vov & Benzi Reference Procaccia, L'vov and Benzi2008; White & Mungal Reference White and Mungal2008; Xi Reference Xi2019; Marchioli & Campolo Reference Marchioli and Campolo2021; Saeed & Elbing Reference Saeed and Elbing2023). Under specific conditions, the drag reduction rate can even reach
$80\,\%$ (Virk et al. Reference Virk, Merrill, Mickley, Smith and Mollo-Christensen1967). Generally, the drag reduction rate increases with polymer concentration and approaches a maximum drag reduction (MDR) level, beyond which further increases in polymer concentration do not result in additional drag reduction. Virk, Mickley & Smith (Reference Virk, Mickley and Smith1970) summarized drag reduction experiments in pipe flow and found that at MDR, the friction factors from different datasets collapse and are limited by a unique asymptote:
$C_f=0.59Re^{-0.58}$ for
$Re\in [ 4\times 10^3,\ 4\times 10^4]$.
$Re$ is the Reynolds number based on the bulk velocity and pipe radius. In addition, at MDR, the mean streamwise velocity profile is also limited by an asymptote:
$u^+=11.7\ln y^+-17$, where
$u^+$ and
$y^+$ are the mean streamwise velocity and wall normal distance normalized by the inner scale. The underlying mechanisms of drag reduction and MDR asymptote in channel/pipe flows have remained an area of active research for decades. Xi (Reference Xi2019) proposed the hibernating/active turbulence picture to explain the universality of the asymptote. At MDR, the active state is reduced by polymers and the hibernating state becomes the dominated dynamics. Samanta et al. (Reference Samanta, Dubief, Holzner, Schäfer, Morozov, Wagner and Hof2013) found that when a polymer solution exhibits high elasticity, it will bypass the Newtonian turbulence and experiences elasto-inertial instability with the flow friction coinciding with the MDR asymptote. Their results suggest that the asymptotic state is not related to ordinary turbulence, but is associated with a new type of turbulent flow: elasto-inertial turbulence, where elasticity and inertia govern the flow dynamics (Choueiri, Lopez & Hof Reference Choueiri, Lopez and Hof2018).
Taylor–Couette (TC) flow, the flow between two concentric cylinders, is also a canonic wall-bounded flow. TC flow has acted as a paradigmatic system of the physics of fluids, such as instabilities (see Fardin, Perge & Taberlet (Reference Fardin, Perge and Taberlet2014) for a review) and high-Reynolds-number turbulence (see Grossmann, Lohse & Sun (Reference Grossmann, Lohse and Sun2016) for a review). Moreover, TC flow has a close analogy with pipe flow (Eckhardt, Grossmann & Lohse Reference Eckhardt, Grossmann and Lohse2001, Reference Eckhardt, Grossmann and Lohse2007). Compared with pipe/channel flow, TC flow has several advantages: (i) TC flow is a closed system, where the polymer concentration and flow properties (like temperature) can be well controlled; (ii) for closed systems, exact global balance relations between the driving and the dissipation can be derived.
TC flow laden with polymers is linearly unstable (Shaqfeh Reference Shaqfeh1996). On average, polymer molecules orient and stretch in the streamwise direction. The resulting hoop stresses create extra pressure rendering the basic Couette flow unstable. At low
$Re$ but high Weissenberg number, the elastic stress induces a spatially smooth and random in time turbulent flow, known as elastic turbulence (Groisman & Steinberg Reference Groisman and Steinberg2000, Reference Groisman and Steinberg2004). Elastic turbulence enhances the mixing but also significantly increases drag in the system (Groisman & Steinberg Reference Groisman and Steinberg2001, Reference Groisman and Steinberg2004). Polymers also alter the transition path of TC flow. In Newtonian fluid, with the increase of
$Re$, the flow experiences, in sequence, the azimuthal Couette flow, axisymmetric Taylor vortex flow, wavy Taylor vortex flow and turbulent Taylor vortex flow (Andereck, Liu & Swinney Reference Andereck, Liu and Swinney1986). In the viscoelastic fluid, many unique flow structures and patterns are reported, such as the diwhirls, rotating standing waves, disordered oscillations, flame patterns and ribbons (Groisman & Steinberg Reference Groisman and Steinberg1996, Reference Groisman and Steinberg1997; Dutcher & Muller Reference Dutcher and Muller2013; Latrache et al. Reference Latrache, Abcha, Crumeyrolle and Mutabazi2016; Lacassagne et al. Reference Lacassagne, Cagney, Gillissen and Balabani2020). These patterns are also reproduced in numerical simulations coupled with the finitely extensible nonlinear elastic-Peterlin model (Thomas, Sureshkumar & Khomami Reference Thomas, Sureshkumar and Khomami2006).
At higher
$Re$ and high Weissenberg number, both inertia and elasticity dominate the dynamics. The transition to elasto-inertial turbulence in TC flow has been summarized in a recent review (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2023). Moazzen et al. (Reference Moazzen, Lacassagne, Thomy and Bahrani2023) studied the effect of polymers on the drag of the system in elasto-inertial turbulence. Boulafentis et al. (Reference Boulafentis, Lacassagne, Cagney and Balabani2024) measured the coherent structures and found that the dominated structures are dynamically independent solitary vortex pairs. These vortex pairs are sustained by the hoop stress and can emerge when moving sufficiently close. The spatial spectra of the intensity fluctuations in the elasto-inertial turbulence have been found to follow a power-law scaling with its exponent approximately
$-7/3$ (Moazzen et al. Reference Moazzen, Lacassagne, Thomy and Bahrani2023) and
$-2.5$ (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024). These exponents are between
$-3$, scaling for a smooth velocity field (Groisman & Steinberg Reference Groisman and Steinberg2000), and
$-5/3$, scaling for high-
$Re$ fully developed turbulence. Thus, the degree of nonlinearity of the underlying flow field in elasto-inertial turbulence is lower than that in inertial turbulence (Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021).
At even higher
$Re$, the inertial effect becomes dominant. Using laser-induced fluorescence, Lee, Sengupta & Wei (Reference Lee, Sengupta and Wei1995) studied the effect of polymers on the formation of small-scale Görtler vortices near the inner cylinder. They observed that polymers suppress the formation of Görtler vortices in the Reynolds number range of
$1483\leq Re \leq 30\ 659$. Rajappan & McKinley (Reference Rajappan and McKinley2020) investigated the combined effect of superhydrophobic surfaces and polymer additives on the extent of drag reduction in TC flow. Polymers reduce about
$20\,\%$ of the drag, and a hydrophobic surface can enhance the drag reduction effect of polymers in the Reynolds number range of
$15\ 000\leq Re \leq 52\ 000$. Song et al. (Reference Song, Lin, Liu, Lu and Khomami2021a) numerically investigated the drag enhancement effect at a constant Weissenberg number for
$500\leq Re \leq 8000$. They divided the flow into two regimes: a low-Reynolds-number regime (
$Re \leq 1000$), where elastic force dominates and large-scale structures contribute to momentum transport, and a high-Reynolds-number regime (
$Re \geq 5000$), where inertial force dominates and small-scale near-wall structures govern the flow dynamics. Barbosa et al. (Reference Barbosa, Cussuol, Soares, Andrade and Khalil2022) studied the effect of polymer concentration on the magnitude of drag reduction rate. They found that the MDR can be overpassed when the polymer concentration is above the overlap concentration, and attribute this effect to polymer aggregation. In these studies, only the inner cylinder rotates. The Reynolds number is defined based on the velocity of the inner cylinder and the gap width between the two cylinders.
The aforementioned studies mainly focus on the effect of polymers on the stability and transition of TC flow. These flows are characterized by low Reynolds numbers. At high
$Re$, previous work employed visualization to study the effect of polymers on the Görtler vortices (Lee et al. Reference Lee, Sengupta and Wei1995) or torque measurement to study the drag modification (Rajappan & McKinley Reference Rajappan and McKinley2020; Barbosa et al. Reference Barbosa, Cussuol, Soares, Andrade and Khalil2022). Several fundamental questions remain: (i) whether the drag reduction behaviour and mechanism in TC flow are similar to those in rectilinear flows (such as channel/pipe flow); (ii) the MDR asymptote in TC flows; (iii) the velocity field including mean velocity and turbulence statistics. The first question addresses whether drag reduction behaviour and mechanisms are universal across different flow systems. In this study, we combine global torque measurement and local velocity field measurement to investigate polymer drag reduction in TC flow. The paper is organized as follows. In § 2, the experimental set-up, the working fluid, the measurement technique and the explored parameter space are described in detail. In § 3, we present our major results, including the friction-factor-Reynolds-number scaling, drag reduction contribution from the mean Taylor vortex and turbulence fluctuation, the effect of polymers on the energy distribution among different scales, and the velocity boundary layer. The paper ends with a summary and conclusion in § 3.
2. Experimental set-up and methodology
2.1. Experimental set-up and working fluid
The TC system is constructed from a commercial rheometer (Discovery Hybrid Rheometer, TA Instruments), as shown in figure 1(a). The inner cylinder, with a radius
$r_i=25$ mm and height
$L=100$ mm, is made of aluminium, which is anodized to form a black oxidation layer. This black surface can reduce unwanted reflection during the particle image velocimetry (PIV) and laser Doppler anemometer (LDA) measurements. The outer cylinder, with an inner radius
$r_o=35$ mm and height 110 mm, is made of Plexiglas, allowing for optical measurement. These two cylinders give a gap
$d=r_o-r_i=10$ mm, a radius ratio
$r_i/r_o=0.714$ and an aspect ratio
$\varGamma =L/d=10$. The outer cylinder is enclosed inside a cubic tank (not shown in figure 1a). The gap between the outer cylinder and the tank can be circulated with water from a refrigerator to control the temperature of the working fluid, which is set to be
$T=25\,^\circ$C. A PT100 thermocouple is used to measure the temperature, with its variation remaining within
$0.1$ K during the experiment.
(a) Sketch of the experimental set-up. The measurement techniques, including PIV and LDA, are also depicted. (b) Two measured instantaneous velocity components, which are denoted as
$u_\theta$ and
$u_r$ in the azimuthal and radial directions, respectively.

The inner cylinder is connected to the rheometer and driven by its motor at a constant angular velocity
$\omega _i$. The outer cylinder is fixed in our experiment. The working fluid exerts frictional drag on the surface of the inner cylinder, which can be read from the TA Instruments Trios software as torque data. The torque resolution of the rheometer is
$0.1\ {\rm nN}{\rm m}$. This torque consists of two parts: one is related to the TC flow and the other is related to the von Kámán flow formed between the ends of the two cylinders. The latter contribution can be estimated by using inner cylinder with different height, and the details can be found in previous studies (Wang et al. Reference Wang, Yi, Jiang and Sun2022; Yi et al. Reference Yi, Wang, van Vuren, Lohse, Risso, Toschi and Sun2022). The torque contributed by the TC flow is denoted as
$\tau$ in this study. For each solution, the torque measurement is repeated three times. Since the difference among repeated measurements is less than
$1\,\%$, we omit the error bar of torque in the following.
The working fluid we used is a mixture of deionized water and glycerol, whose weight fraction is
$w_g=25\,\%$. At
$T=25\,^\circ$C, the density and dynamic viscosity of the mixture are
$\rho _s=1.058\times 10^3\ {\rm kg}\ {\rm m}^{-3}$ and
$\mu _s=1.79\times 10^{-3}\ {\rm Ns}\ {\rm m}^{-2}$ (Wang et al. Reference Wang, Yi, Jiang and Sun2022). Polyacrylamide polymers with two molecular weights
$M_w$ are used: one with
$M_w=5\times 10^6$ (Sigma-Aldrich) and is denoted as pam5e6; another with
$M_w=2\times 10^7$ (Macklin), denoted as pam2e7. pam5e6 has been widely used in previous studies (Samanta et al. Reference Samanta, Dubief, Holzner, Schäfer, Morozov, Wagner and Hof2013; Choueiri et al. Reference Choueiri, Lopez and Hof2018). The preparation protocol of the polymer solution can be found in our previous studies (Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021; Zhang & Xi Reference Zhang and Xi2022; Peng, Zhang & Xi Reference Peng, Zhang and Xi2023). The dynamic viscosity of the polymer solutions is measured by the Discovery Hybrid Rheometer equipped with a cone-plate geometry, whose diameter is 40 mm and angle is
$2^\circ$. The shear viscosity
$\mu _{\dot {\gamma }}$ as a function of shear rate
$\dot {\gamma }$ for these two polymers is shown in Appendix A. The overlap concentration estimated from the zero-shear viscosity
$\mu _0$ is
$\phi ^*\approx 680$ ppm (parts per million by weight) for pam5e6 and
$\phi ^*\approx 17$ ppm for pam2e7 (see Appendix A). In this study, the used polymer concentration
$\phi$ in TC is in the range of
$0\leq \phi \leq 100$ ppm for pam5e6 and
$0\leq \phi \leq 4$ ppm for pam2e7. Hence, the polymer solution is in the dilute regime. The average shear rate in the gap of TC flow can be estimated to be
$\omega _ir_i/d$, which is larger than 80 s
$^{-1}$ in this study. When
$\dot {\gamma }>80$, the shear-thinning effect of both fluids is found to be neglected. We therefore average
$\mu _{\dot {\gamma }}$ over the range of
$80<\dot {\gamma }<400$ to obtain an averaged viscosity
$\mu$, and take this value as the viscosity of the working fluid (see Appendix A for more details). In the dilute regime, the relaxation time
$t_p$ of the polymer solution can be estimated from the Zimm model (Ouellette, Xu & Bodenschatz Reference Ouellette, Xu and Bodenschatz2009; Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021), i.e.
$t_p=N^{9/5}a^3\mu /(k_BT)$, where
$N$ is the number of monomers,
$a$ the length of one monomer and
$k_B$ the Boltzmann constant.
In TC turbulence, the control parameter is the Reynolds number, defined as
where
$\nu =\mu /\rho _s$ is the kinematic viscosity of the solution; or the Taylor number, defined as
When using viscoelastic fluid, two additional parameters are needed: the polymer concentration
$\phi$ and Weissenberg number defined as
The response parameter of the system is the dimensionless torque given by
which is related to the friction coefficient
$C_f$ by the relation
The drag reduction rate by polymers is defined as
where the subscripts
$_p$ and
$_n$ denote the viscoelastic case and Newtonian case, respectively. Please note that
$G_p$ and
$G_n$ are dimensionless torque measured at the same
$Re$ instead of at the same
$\omega _i$, namely, we increase
$\omega _i$ to keep
$Re$ the same in the viscoelastic case. The parameter space explored in the present study is shown in figure 2. The Reynolds number is in the range of
$4\times 10^3< Re<2.5\times 10^4$, and the corresponding Taylor number is in the range of
$2.8\times 10^7< Ta<7.5\times 10^8$. In this parameter range, the Taylor vortex is present and plays an important role in the flow dynamics (Ostilla-Mónico et al. Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014). The
$Wi$ for pam2e7 is nearly an order of magnitude larger than that for pam5e6. At high
$Re$, it is observed that the torque slowly increases with time, signifying that the polymer molecules are degraded by the high shear near the inner cylinder. We therefore use the first one minute torque data after it becomes statistically steady to obtain the average
$\tau$ for pam2e7.
Parameter space explored in this study. The green squares mark the parameters where we performed PIV and LDA measurements.

In viscoelastic flow, the elastic number
$El=Wi/Re$, which compares the elastic and inertial forces, is also an important parameter. In our study,
$El=Wi/Re=t_p\nu /d^2$ is independent of
$Re$;
$El=1.5\times 10^{-4}$ and
$1.3\times 10^{-3}$ for pam5e6 and pam2e7, respectively. According to Song et al. (Reference Song, Zhu, Lin, Liu and Khomami2023), TC turbulence with
$Re>4\times 10^3$ and
$El<0.01$ is in the inertia-dominated regime.
2.2. Methodology
We employ PIV to measure the velocity in the
$r$–
$\theta$ plane, where
$r$ and
$\theta$ are in the radial and azimuthal directions, respectively. Due to the presence of Taylor vortices, the turbulence statistics depend on the axial height (we refer to the axial direction as
$z$). The PIV measurements are conducted at 19 different heights which are separated by
$\Delta h=1.5$ mm. Thus, the height of the measurements spans a length of
$H=18\Delta h=27$ mm (figure 1a). The measurement length in units of gap width is
$H/d=2.7$, which is larger than the axial span of one pair of Taylor vortices observed in Newtonian turbulence (Froitzheim et al. Reference Froitzheim, Ezeta, Huisman, Merbold, Sun, Lohse and Egbers2019). Previous numerical simulations have reported that the Taylor vortex can be modified by polymers and its wavelength can either be decreased or increased (Song et al. Reference Song, Lin, Liu, Lu and Khomami2021a). We therefore measure a larger height to test whether it is the case.
The seeding particles used are polystyrene with an average diameter of 5
$\mathrm {\mu }$m and density
$1.05\times 10^3\ {\rm kg}\ {\rm m}^{-3}$. The seeding particles are illuminated by a laser sheet from a double-pulsed cavity laser (Beamtech Vlite200, 532 nm). The thickness of the laser sheet is
$\approx 1$ mm. The laser is mounted on a traverse system (fly-opt PSTV50-S57) which allows us to precisely change the location of the laser sheet along the vertical direction. The images of the seeding particles are recorded by an HiSense Zyla CMOS camera (2560–2160 pixel, 16 bit). The PIV is operated on a double frame mode at a frame rate
$f=4$ Hz, which is much smaller than the characteristic frequency of the flow
$\omega _i/(2{\rm \pi} )$, indicating that the measured velocity fields are independent of each other. At each height, we measure 500 velocity fields. During this time, the inner cylinder completes at least 1700 revolutions.
The captured image pairs are processed with a multi-pass algorithm. The interrogation window size is
$64\times 64$ and
$32\times 32$ in the first and second pass, respectively. In each pass, the overlap of the window is
$50\,\%$. This process yields velocity fields in Cartesian coordinate with a spacing of 0.11 mm between neighbour vectors. We then map the Cartesian velocity fields onto a polar grid using bilinear interpolation (Froitzheim et al. Reference Froitzheim, Ezeta, Huisman, Merbold, Sun, Lohse and Egbers2019). The measured velocity components in the radial and azimuthal directions are denoted as
$u_r$ and
$u_\theta$ (figure 1b), which are functions of
$r$,
$\theta$,
$z$ and time
$t$. Since TC turbulence is statistically stationary and axisymmetric, the velocity field can be decomposed into its mean and fluctuation parts by the following relation, e.g.
$u_\theta =\langle u_\theta \rangle _{t,\theta } + u^\prime _\theta$, where
$\langle \rangle _{t,\theta }$ indicates average operation over time and in the
$\theta$ direction. Here, we take
$u_\theta$ as an example, similar operations can be done for
$u_r$. The root-mean-square (r.m.s.) velocity of
$u^\prime _\theta$ is defined as
$\sigma _{u_\theta }=\sqrt {\langle u^{\prime 2}_\theta \rangle _{t,\theta,z}}$, where an additional average over the
$z$ direction across one pair of Taylor vortices is applied. In the following, averaging in the
$\theta$ direction is applied by default and therefore will be omitted in the subscript of the average operation.
We also perform LDA (TSI) measurement for its high temporal resolution. The LDA consists of two pairs of laser beams, and each pair measures one component of the velocity (see figure 1a). The LDA is mounted on a high-precision stage, which can move in the
$x$,
$y$ and
$z$ directions independently. The measurements are done for two radial locations:
$(r-r_i)/d=0.06$ and
$0.5$; and for three different heights: vortex centre, outflow and inflow. At the middle gap (
$(r-r_i)/d=0.5$), we measure both
$u_\theta$ and
$u_r$. Near the inner cylinder (
$(r-r_i)/d=0.06$), we can only measure
$u_\theta$ due to the block of the laser beam by the inner cylinder. For each experiment, we acquire
$1\times 10^6$ data points. The PIV and LDA measurements are performed at three Reynolds numbers (marked as green squares in figure 2) using the pam5e6 polymer due to its long-term stability.
In TC flow, the torque of the system is related to the angular velocity flux
$J^\omega$ by the relation
$G=\nu ^{-2}J^\omega$, where
$J^\omega$ is given by Eckhardt et al. (Reference Eckhardt, Grossmann and Lohse2007):
which is conserved along the radial direction. Here,
$\omega =u_\theta /r$ is the angular velocity. In the right-hand side of (2.7), the first and second terms are the advective and viscous diffusion contributions, respectively. In analogy to the definition of the dimensionless heat flux in Rayleigh–Bénard flow, Eckhardt et al. (Reference Eckhardt, Grossmann and Lohse2007) defined a Nusselt number as the ratio of
$J^\omega$ and its value
$J^\omega _{lam}=2\nu r_i^2 r_o^2 \omega _i/(r_o^2-r_i^2)$ for the laminar case, i.e.
$Nu_\omega =J^\omega /J^\omega _{lam}$.
In the range of Reynolds number investigated here, the TC turbulence is a combination of turbulent Taylor vortices and background fluctuations. Thus,
$r^3 \langle u_r\omega \rangle _{t,z}$ can be further decomposed into two parts (Brauckmann & Eckhardt Reference Brauckmann and Eckhardt2013), namely:
with the former one connected to the mean Taylor vortex motion and the latter one the turbulent fluctuations. The latter one can also be rewritten as
$r^3 \langle u^\prime _r\omega ^\prime \rangle _{t,z}=r^2 \langle u^\prime _r u_\theta ^\prime \rangle _{t,z}$, which is the Reynolds stress contribution to the angular velocity flux. We denote the dimensionless form of these two contributions as
$Nu_\omega ^{tv}=r^3 \langle \langle u_r\rangle_{t}\langle \omega \rangle_{t} \rangle_z/J^\omega _{lam}$ and
$Nu_\omega ^{rs}=r^2 \langle u^\prime _r u_\theta ^\prime \rangle _{t,z}/J^\omega _{lam}$.
3. Results
3.1. Drag reduction characteristics
Polymers reduce the drag of wall-bounded turbulence. In TC flow, this is reflected in the reduction of the dimensionless torque
$G$ compared with the Newtonian case. A comparison of
$G$ in the Newtonian case between this study and that of Lewis & Swinney (Reference Lewis and Swinney1999) is made in Appendix B. In figure 3(a), we compare
$G$ with increasing
$\phi$ for different
$Re$. To show the trend more clearly,
$G$ is scaled by
$Re^{-1.6}$ (Wang et al. Reference Wang, Yi, Jiang and Sun2022). In the range of
$Re$ investigated here, the local scaling exponent for
$G$ is not constant (Lewis & Swinney Reference Lewis and Swinney1999). Therefore,
$GRe^{-1.6}$ does not show a plateau in the Newtonian case (
$\phi =0$ ppm) for the whole range of
$Re$ investigated here. Instead,
$GRe^{-1.6}$ is nearly constant for
$Re$ in the range from
$9\times 10^3$ to
$2.5\times 10^4$, in line with previous studies (Wang et al. Reference Wang, Yi, Jiang and Sun2022; Yi et al. Reference Yi, Wang, van Vuren, Lohse, Risso, Toschi and Sun2022). With the increase of
$\phi$ and
$Re$,
$G$ for viscoelastic cases deviates more from their Newtonian counterparts, a feature consistent with what is reported in channel/pipe flows (Virk Reference Virk1975). Additionally, the difference between neighbouring
$\phi$ values diminishes as
$\phi$ increases, suggesting that the effects of polymers may saturate at high concentrations. To further strengthen this point, we show the drag reduction rate
$DR$ as a function of
$\phi$ for two typical
$Re$ in figure 3(b). Here,
$DR$ increases with
$\phi$ and approaches an asymptote at high concentration, which is referred to as MDR (Procaccia et al. Reference Procaccia, L'vov and Benzi2008; White & Mungal Reference White and Mungal2008).
(a) Dimensionless torque compensated by the
$1.6$ scaling,
$GRe^{-1.6}$, as a function of the Reynolds number
$Re$. (b) Drag reduction rate,
$DR$, for two typical
$Re$. The horizontal dashed lines represent the maximum drag reduction rate. The polymer we used here is pam5e6.

In channel/pipe flow, the friction factor
$C_f$ is generally used as the dimensionless drag of the flow system. Here, we also show
$C_f$ of the Newtonian case and viscoelastic cases at the MDR for both pam5e6 and pam2e7 in figure 4(a). At MDR, the two datasets nearly collapse and can be fitted with a power-law scaling
$C_f\sim Re^{-(0.58\pm 0.1)}$, which suggests that the friction factor at MDR in TC turbulence may also be universal, similar to what has been reported in pipe flows (Virk et al. Reference Virk, Mickley and Smith1970). In figure 4(b),
$C_f$ is compensated by the
$Re^{-0.58}$ scaling;
$C_fRe^{0.58}$ exhibits plateau-like behaviour for both pam5e6 and pam2e7, demonstrating the robustness of this scaling. For the two polymers used in this study, the
$Wi$ for pam2e7 is about ten times of the
$Wi$ for pam5e6 (see figure 2). However, the drag reduction rate differs by only about
$3\,\%$ at MDR. In experiments, the MDR asymptote generally is approached by increasing the polymer concentration; however, it is approached by the increase of Weissenberg number while keeping the concentration constant in numerical simulations (Procaccia et al. Reference Procaccia, L'vov and Benzi2008; Xi Reference Xi2019). The critical condition under which the MDR limit is realized is a function of both
$Wi$ and
$\phi$, and there is no systematic study on this up to now, due to the large parameter space. Studies in this direction would have important implication on practical application of polymer drag reduction.
(a) Friction factor,
$C_f$, for the Newtonian case and viscoelastic case under the maximum drag reduction condition. The dashed line has a slope of
$-0.58$. (b)
$C_f$ compensated by the
$Re^{-0.58}$ scaling.

Virk et al. (Reference Virk, Mickley and Smith1970) summarized previous studies about drag reduction by polymers in pipe flow, and found that
$C_f$ is limited by an asymptote described by the experimental correlation:
$C_f=0.59 Re^{-0.58}$ for
$4\times 10^3< Re<4\times 10^4$. While the scaling exponents are close in TC and pipe flows, the extent of drag reduction in TC is much smaller. For comparison, we compute the drag reduction rate for typical rectilinear flows, including pipe, channel and duct flows, under the MDR condition. Here, the drag reduction rate can be defined as
$DR=1-C_{f,p}/C_{f,n}$. The friction factors of the rectilinear flows for the Newtonian and viscoelastic cases are from Owolabi, Dennis & Poole (Reference Owolabi, Dennis and Poole2017). The drag reduction rates,
$DR$, are compared in figure 5. For rectilinear flows,
$DR$ can reach about
$60\,\%$ at
$Re\approx 1\times 10^4$ (Owolabi et al. Reference Owolabi, Dennis and Poole2017). However, in TC flow, the maximum drag reduction is about
$23\,\%$ at
$Re=2.2\times 10^4$. The level of drag reduction rate in our study is consistent with the results reported in numerical simulation (Lin et al. Reference Lin, Wan, Zhu, Liu, Lu and Khomami2022) and in experiments (Rajappan & McKinley Reference Rajappan and McKinley2020; Barbosa et al. Reference Barbosa, Cussuol, Soares, Andrade and Khalil2022). In their study, Rajappan & McKinley (Reference Rajappan and McKinley2020) used two kinds of polymers: polyacrylamide and polyethylene oxide.
Drag reduction rate at MDR for rectilinear flows (including pipe, channel and duct flows) and Taylor–Couette turbulence.

3.2. Drag reduction mechanism
In TC turbulence, the viscous diffusion dominates in the near-wall region, while the advective contribution dominates in the bulk region of the gap (Brauckmann & Eckhardt Reference Brauckmann and Eckhardt2013). Unfortunately, the diffusion term cannot be accurately measured due to the low spatial resolution of our PIV. For TC flow laden with polymers, along with the advective and viscous diffusion contributions in (2.7), polymer stress also contributes to the angular velocity flux (Song et al. Reference Song, Lin, Liu, Lu and Khomami2021a). However, this contribution cannot be directly measured experimentally. In channel and TC flows, the polymeric contribution to the drag is positive (Min et al. Reference Min, Yoo, Choi and Joseph2003; Song et al. Reference Song, Lin, Liu, Lu and Khomami2021a). One thus may conjecture that the advective contribution will be reduced by polymers in the bulk region. In this section, we attempt to reveal the underlying mechanism of the low drag reduction rate observed in TC turbulence by investigating the two parts of the advective contribution according to (2.8): the Reynolds shear stress and the Taylor vortex.
We first study the effect of polymers on the Reynolds shear stress. The radial dependence of the Reynolds shear stress, normalized by the inner cylinder velocity
$\omega _i r_i$, is shown in figure 6(a). In the near-wall region,
$\langle u^\prime _r u_\theta ^\prime \rangle _{t,z}$ exhibits local maximum and minimum with their intensity higher near the inner cylinder than near the outer one, similar to the streamwise vorticity fluctuations observed by Song et al. (Reference Song, Lin, Liu, Lu and Khomami2021a). In wall-bounded turbulence, the local maximum in the Reynolds shear stress is found to be closely connected with the near-wall quasi-streamwise vortices (Jiménez Reference Jiménez2018). In TC turbulence, the near-wall coherent structures are dominated by Görtler vortices (Dong Reference Dong2007). Dong (Reference Dong2007) also found that the intensity of Görtler vortices near the inner cylinder is stronger than near the outer one. The peaks near both cylinders decrease with the addition of polymers, suggesting that the Görtler vortices are suppressed by polymers, a phenomenon that was directly observed by laser induced fluorescence experiments (Lee et al. Reference Lee, Sengupta and Wei1995). This is also similar to the effects of polymers on channel flow, where polymers suppress the near-wall quasi-streamwise vortices and lead to a reduction of Reynolds shear stress (Min et al. Reference Min, Yoo, Choi and Joseph2003; Kim et al. Reference Kim, Adrian, Balachandar and Sureshkumar2008).
(a) Radial dependence of the Reynolds shear stress normalized by the velocity of the inner cylinder, i.e.
$\langle u^\prime _\theta u^\prime _r\rangle _{t,z}/(\omega _i r_i)^2$. (b) Radial dependence of the azimuthal (solid lines) and radial (dashed lines) r.m.s. velocity.
$Re=1.3\times 10^4$.

In figure 6(a), it can be seen that the Reynolds shear stress is suppressed by polymers, and the suppression effect strengthens as the polymer concentration increases. The decrease in Reynolds shear stress implies that the turbulent flow contributes less drag to TC flow laden with polymers according to (2.7) and (2.8). We also note that at MDR, the Reynolds shear stress in the viscoelastic case is smaller than in the Newtonian case, but still of the same order. This behaviour differs from what has been reported in channel/pipe flows, where the Reynolds shear stress is an order of magnitude smaller than in the Newtonian case and nearly vanishes at MDR (Warholic, Massah & Hanratty Reference Warholic, Massah and Hanratty1999; Choueiri et al. Reference Choueiri, Lopez and Hof2018). That is to say that the turbulent contribution to the drag can be ignored in channel/pipe flows when approaching the MDR asymptote.
The r.m.s. velocity in the azimuthal and radial directions,
$\sigma _{u_\theta }/(\omega _i r_i)$ and
$\sigma _{u_r}/(\omega _i r_i)$, are presented in figure 6(b). It can be observed that
$\sigma _{u_\theta }/(\omega _i r_i)$ and
$\sigma _{u_r}/(\omega _i r_i)$ are reduced across the gap by the polymers. Additionally, the reduction effect becomes stronger with the increase of polymer concentration.
We then study the effect of polymers on the Taylor vortex. As
$Re$ increases, the turbulent fluctuation grows with the strength of Taylor vortex diminishing when only the inner cylinder rotating (Ostilla-Mónico et al. Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014). Brauckmann & Eckhardt (Reference Brauckmann and Eckhardt2013) found that the Taylor vortex contributes to approximately half of the total angular velocity flux in the bulk region for
$Re\approx 2\times 10^4$. For smaller
$Re$, this contribution becomes even larger since the Taylor vortex occupies a larger fraction of the total energy in the flow. In figure 7, we compare the contour of the mean radial velocity from the Newtonian case in panel (a) and viscoelastic case at
$\phi =40$ ppm in panel (b) at
$Re = 1.3\times 10^4$. We observe that the wavelength and pattern of the Taylor vortex are nearly the same for these two cases. In the parameter space of this study, the inertia dominates the flow dynamics (Song et al. Reference Song, Zhu, Lin, Liu and Khomami2023), we thus can expect that the Taylor vortex, which is an inertial effect, would hardly be modified by polymers. This result is consistent with the numerical simulation of Song et al. (Reference Song, Lin, Liu, Lu and Khomami2021a), which reported that increasing fluid inertia (while keeping
$Wi$ constant) hinders the elastic effects.
Contours of the mean radial velocity,
$\langle u_r \rangle _t/(\omega _i r_i)$, for (a) the Newtonian case and (b) viscoelastic case at
$\phi =40$ ppm.
$Re=1.3\times 10^4$.

While the wavelength and pattern of the Taylor vortex are almost not altered by polymers, its strength is weakened. To quantify this, we adopt the method proposed by Froitzheim, Merbold & Egbers (Reference Froitzheim, Merbold and Egbers2017). The kinetic energy of the flow averaged over time and space reads
The first term is the angular base flow, the main contribution to
$E(r)$. To reveal the secondary flow and the turbulent fluctuations, we neglect this energy portion. Since the axial velocity is not measured, we only use the central region of the velocity to define the energy, with the assumption that the axial velocity of an idealized turbulent Taylor vortex vanishes in the centre of the gap (Froitzheim et al. Reference Froitzheim, Merbold and Egbers2017). To be specific, the energy related to the mean Taylor vortex and turbulent kinetic energy are given by
and they are normalized by
$(\omega _i r_i)^2$.
Figure 8 shows
$E_{tv}$ and
$E_{tke}$ as a function of polymer concentration. It is found that, while
$E_{tke}$ is strongly depressed by polymers with the increase of polymer concentration,
$E_{tv}$ is weakened by a less degree when
$\phi >4$ ppm. We conclude that the major contribution to drag reduction by polymers comes from the reduction in turbulence, with the weakened Taylor vortex playing a minor role in the current parameter regime of TC turbulence based on (2.8). The lower drag reduction rate in the current parameter regime of TC turbulence compared with channel/pipe/duct flows can be understood as follows: the angular velocity flux from the mean Taylor vortex motion is the dominating part of the system's drag, which is larger than
$50\,\%$ in the range of
$Re$ investigated in this study (Brauckmann & Eckhardt Reference Brauckmann and Eckhardt2013). However, polymers only have a minor effect on the Taylor vortex. In addition, due to the turbulent nature of the Taylor vortex, the Reynolds shear stress cannot be reduced to vanish at MDR as observed in figure 6(a).
Energy of the Taylor vortex,
$E_{tv}$, and turbulent kinetic energy,
$E_{tke}$, as a function of polymer concentration.
$Re=1.3\times 10^4$.

To understand how polymers suppress turbulence and weaken the Taylor vortex, we delve into the probability density functions (p.d.f.s) of the angular velocity flux from the Reynolds shear stress, i.e.
$r^3 u^\prime _r \omega ^\prime$, which is shown in figure 9(a). Both positive and negative tails of the p.d.f.s shrink with the increase of polymer concentration. When
$r^3 u^\prime _r \omega ^\prime$ are scaled by their mean value, as shown in figure 9(b), their positive tails nearly collapse with each other. However, the negative tails exhibit a slight shrinkage. In TC turbulence, the angular velocity plumes detached from the velocity boundary layer are observed to be the underlying structure of efficient angular velocity carriers (Brauckmann & Eckhardt Reference Brauckmann and Eckhardt2013; Ostilla-Mónico et al. Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014; Froitzheim et al. Reference Froitzheim, Ezeta, Huisman, Merbold, Sun, Lohse and Egbers2019). The shrinkage of the p.d.f.s in their tails may suggest that the angular velocity plumes are modified by polymers.
(a) Probability density functions of the angular velocity flux contributed by the Reynolds shear stress,
$r^3 u^\prime _r \omega ^\prime$, which is non-dimensionalized by the value for laminar flow
$J^\omega _{lam}$. Inset shows an enlarged view of the main plot near its peak in a linear–linear scale. (b) Probability density functions of
$r^3 u^\prime _r \omega ^\prime /J^\omega _{lam}$ scaled by its mean value, i.e.
$r^2 u^\prime _r u_\theta ^\prime /(J^\omega _{lam}Nu_\omega ^{rs})$. The probability density functions are calculated from data sampled on cylinder surface with
$(r-r_i)/d=0.5$ at
$Re=1.3\times 10^4$.

At
$Re=1.3\times 10^4$, the average slope boundary layer thickness in the gap can be estimated as
$\lambda _\omega =d/(2Nu_\omega )\approx 0.37$ mm, following the theory of Eckhardt et al. (Reference Eckhardt, Grossmann and Lohse2007). We plot in figure 10(a) the p.d.f.s of the normalized azimuthal velocity by their mean and standard deviation for data sampled on cylinder surface with
$(r-r_i)/d=0.03$, which is located inside the boundary layer. The p.d.f. is positively skewed since the plumes with higher speed dominate near the inner cylinder (Dong Reference Dong2007). Both tails of the p.d.f.s shrink with the increase of polymer concentration, as already observed in the p.d.f.s of
$r^3 u^\prime _r \omega$. The reduction in higher angular velocity fluctuations implies that polymers steady the velocity boundary layer.
(a) Probability density functions of the angular velocity,
$\omega$, sampled on cylinder surface
$(r-r_i)/d=0.03$. Here,
$\omega$ is normalized by its mean value and standard deviation in such a way that the integration of the curve with respect to the abscissa is
$1$. (b) Plume fraction
$F_p$ normalized by its value from the Newtonian case
$F_p(\phi =0)$.
$Re=1.3\times 10^4$.

Similar to the hot and cold plumes defined in Rayleigh–Bénard turbulence (Xie et al. Reference Xie, Huang, Funfschilling, Li, Ni and Xia2015), we define fast and slow plumes for TC turbulence. Specifically, they are defined as a time period when
$\pm [ \omega -\langle \omega \rangle _t ]>b\sigma _{\omega }$, with
$b=3$, ‘
$+$’ for fast plumes and ‘
$-$’ for slow ones. The plume fraction
$F_p$ is the probability of observing these plumes in a unit time. Plume fraction normalized by its Newtonian value, i.e.
$F_p/F_p(\phi =0)$, is shown in figure 10(b). For slow plumes,
$F_p/F_p(\phi =0)$ drops sharply with the addition of polymers; while for fast plumes,
$F_p/F_p(\phi =0)$ drops when
$\phi >4$ ppm. Polymers steady the velocity boundary layer, which results in a reduction of both the plume emission rate and the angular velocity flux (figure 9a). The reduction in the number of plumes detached from the boundary layer also feeds less energy to the bulk, leading to lower turbulent fluctuations (figure 6b). We tried different values of
$b$ from 2 to 3.5 and the normalized plume fraction based on different values of
$b$ shows similar trend as figure 10(b).
3.3. Velocity power spectra
In the previous section, we have shown that polymers reduce the turbulence fluctuations (figure 6b). Turbulent flows are characterized by fluctuations with a wide spectrum of length and time scales (Pope Reference Pope2000). The effect of polymers on the different scales of turbulence is investigated by the kinetic energy spectrum of the azimuthal velocity
$u_\theta ^\prime$. The time series of LDA measurements are linearly interpolated using twice the average acquisition frequency to create a time series with equal temporal spacing (Huisman, Lohse & Sun Reference Huisman, Lohse and Sun2013a). Discrete Fourier transformation is applied to the time series to obtain the spectrum, which is shown in figure 11(a) near the inner cylinder (
$(r-r_i)/d=0.06$) and in figure 11(b) at the middle gap (
$(r-r_i)/d=0.5$). In figure 11, we also plot the frequency corresponding to the gap width
$\langle u_\theta \rangle _t/d$ for the
$\phi =40$ ppm case as a reference. Near the inner cylinder, the energy content of turbulent fluctuations with scales smaller than
$d$ are strongly depressed, which is a result of the suppression of the nonlinear energy transfer mechanism by polymers (Ouellette et al. Reference Ouellette, Xu and Bodenschatz2009; Xi, Bodenschatz & Xu Reference Xi, Bodenschatz and Xu2013; Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021). In the near-wall region of TC turbulence, these small-scale structures are Görtler vortices, providing direct evidence to our observation in figure 6(a) that Görtler vortices are suppressed by polymers. For lower frequency, see
$f<20$ Hz or scale larger than
$5d$ (here, we invoke the Taylor-frozen hypothesis), the energy content is slightly enhanced. These large-scale structures are azimuthal velocity streaks (Dong Reference Dong2007). Polymers suppress the nonlinear energy transfer from large to small-scales (Ouellette et al. Reference Ouellette, Xu and Bodenschatz2009; Xi et al. Reference Xi, Bodenschatz and Xu2013; Valente, da Silva & Pinho Reference Valente, da Silva and Pinho2014; Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021), and the injected energy may accumulate in these large-scale structures. Thus, the effect of polymers on the turbulent fluctuations is a redistribution of energy among different scales: suppression of small-scale random fluctuation and energization of large-scale azimuthal velocity streaks. Since the former effect dominates, the overall effect of polymers is to suppress turbulence fluctuation in the near-wall region as shown in figure 6(b).
Power spectra for the azimuthal velocity
$u^\prime _\theta$ at (a)
$(r-r_i)/d=0.06$ and (b)
$(r-r_i)/d=0.5$. The vertical dashed lines denote
$\langle u_\theta \rangle _t/d$, which is the characteristic frequency corresponding to the gap width. The dash-dotted lines have a slope of
$-5/3$.
$Re=1.3\times 10^4$.

In wall-bounded turbulence, the velocity gradient is much higher near the wall than in the centre. The ability of turbulent flow to stretch polymer molecules at middle gap can be more precisely quantified by the local Weissenberg number
$Wi_{local}$, defined as
$Wi_{local}=t_p/t_\eta$, where
$t_\eta =\sqrt {\nu /\epsilon _d}$ is the Kolmogorov time scale at
$(r-r_i)/d=0.5$. Here,
$\epsilon _d\approx 0.1\tau \omega _i/[ {\rm \pi}(r_o^2-r_i^2)L\rho _s ]$ is the average energy dissipation rate in the bulk (Ezeta et al. Reference Ezeta, Huisman, Sun and Lohse2018). At
$Re=1.3\times 10^4$,
$Wi_{local}\approx 2.6$, which is smaller the coil–stretch transition Weissenberg number
$Wi_c=3\sim 4$ reported in homogeneous isotropic turbulence (Watanabe & Gotoh Reference Watanabe and Gotoh2010). Polymers cannot be stretched locally near the middle of the gap at
$Re=1.3\times 10^4$. From figure 11(b), we observe that all scales resolved here are inhibited by polymers. This inhibition instead reflects the effect of polymers on the velocity boundary layer: polymers steady the boundary layer and reduce the emission rate of velocity plume detached from the boundary layer, resulting in less energy feed into the bulk. To prove this, we show in figure 12(a) the normalized power spectra by
$\omega _i/(2{\rm \pi} )$ and the r.m.s. velocity (
$\sigma _{u_\theta }$ for azimuthal velocity
$u^\prime _\theta$ (solid lines) and
$\sigma _{u_r}$ for radial velocity
$u^\prime _r$ (dashed lines)). The integral of the normalized spectra with respect to
$f/(\omega _i/2{\rm \pi} )$ would be 1. The normalized spectra nearly overlap with each other, in support of the above discussion. At a higher Reynolds number
$Re=1.8\times 10^4$ (
$Wi_{local}\approx 4.1$), all the scales are suppressed by polymers as the
$Re=1.3\times 10^4$ case. However, the spectra cannot be collapsed by the above normalization (see figure 12b). Instead, the spectrum of
$\phi =40$ ppm changes its slope at an intermediate frequency, above and below which the energy content is increased and reduced, respectively, implying that the turbulent structures are altered in the bulk region at higher
$Re$ due to the larger
$Wi_{local}$.
(a) Power spectra normalized by
$\omega _i/(2{\rm \pi} )$ and r.m.s. velocity (
$\sigma _{u_\theta }$ for azimuthal velocity
$u^\prime _\theta$ (solid lines) and
$\sigma _{u_r}$ for radial velocity
$u^\prime _r$ (dashed lines)). Data shown here are from
$Re=1.3\times 10^4$. (b) Same as panel (a) but only for
$u^\prime _\theta$ at
$Re=1.8\times 10^4$. The LDA measurement position is at
$(r-r_i)/d=0.5$.

It is interesting to note that the spectra of
$u^\prime _r$ exhibit two power-law scalings with a
$-1$ slope for intermediate frequency and a
$-2.2$ slope for high frequency, and the cross-over happens at
$f_c/(\omega _i/2{\rm \pi} )) \approx 16$. This behaviour is reminiscent of what was observed in elastic turbulence (Groisman & Steinberg Reference Groisman and Steinberg2004). In elastic turbulence, where the elastic stress dominates the dynamics and inertia can be neglected, Groisman & Steinberg (Reference Groisman and Steinberg2004) reported that the spectra of
$u^\prime _r$ show respectively the
$-1.1$ and
$-2.2$ power-law decay for intermediate and high frequency in TC flow. However, in their flow, the cross-over between the two power-laws happens at
$f_c/(\omega _i/2{\rm \pi} ) \approx 0.45\pm 0.05$. The similar decay slope suggests that one cannot discriminate whether the flow is dominated by elasticity or inertia based solely on the spectrum of
$u^\prime _r$ in TC turbulence. The cross-over frequency is much higher in inertia-dominated flow than the elastic turbulence, indicating that much smaller scale structures are developed in the former case. The reason for the similar scaling observed in our study and in elastic turbulence is not known to us yet.
In elasto-inertial turbulence of TC flow, experiments have reported spatial spectra with decay exponents of
$-7/3$ (Moazzen et al. Reference Moazzen, Lacassagne, Thomy and Bahrani2023) and
$-2.6$ (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024), while numerical simulations have found exponents of
$-3$ (Lopez Reference Lopez2022) or less than
$-3$ (Song et al. Reference Song, Lin, Liu, Lu and Khomami2021a,Reference Song, Wan, Liu, Lu and Khomamib). For the Reynolds numbers investigated here, we do not observe a clear inertial scaling range with an exponent of
$-5/3$ for
$u^\prime _\theta$ in either the near-wall or bulk regions, as shown in figure 11, likely due to the Reynolds number not being high enough (Huisman et al. Reference Huisman, Lohse and Sun2013a). In TC flow with polymers, we also cannot assign a definitive scaling exponent to the spectra of
$u^\prime _\theta$ because of the limited scaling range. Further experiments are needed to characterize the spectra in viscoelastic TC flow across different parameter regimes.
3.4. Velocity boundary layer
In the previous sections, we discuss the effects of polymers on the Taylor vortex and turbulent fluctuations. The mean velocity profile near the wall is of great interest in the study of polymer drag reduction. Warholic et al. (Reference Warholic, Massah and Hanratty1999) divided the effect of polymers into two regimes based on the degree of drag reduction: low-drag-reduction regime, where the mean streamwise velocity profile for the viscoelastic case shifts upward in the logarithmic law layer with its slope unchanged; high-drag-reduction regime, where the slope of the mean streamwise velocity profile increases with drag reduction rate and approaches an empirical asymptote at MDR. The mean azimuthal velocity in TC flow is defined as
where
$\langle u_\theta (r) \rangle _{t,z}$ is the azimuthal component of the velocity and
$u_\theta (r_i)=\omega _i r_i$ is the azimuthal velocity of the inner cylinder. Here,
$y^+=(r-r_i)/\delta _\nu$ is the distance from the inner cylinder in unit of the viscous length scale
$\delta _\nu =\nu /\sqrt {\tau /(2{\rm \pi} \rho _s r_i^2L)}$.
Figure 13(a) shows
$u^+_\theta$ as a function of
$y^+$ for various polymer concentrations, along with the
$u^+_\theta =y^+$ profile in the viscous sublayer and the logarithmic law
$u^+_\theta =1/\kappa \ln y^++B$. For pipe, flat plate and channel flows,
$\kappa =0.40$ and
$B=5.0$ (Marusic et al. Reference Marusic, Monty, Hultmark and Smits2013). For channel flow laden with polymers,
$\kappa =1/11.7$ and
$B=-17$ at MDR (Virk et al. Reference Virk, Mickley and Smith1970). We observe that
$u^+_\theta$ is lower than the logarithmic law of Newtonian turbulence in a flat plate boundary layer, but it still exhibits a logarithmic shape for
$y^+>50$. The growth rate of
$u^+_\theta$ in the logarithmic layer is smaller than
$\kappa =0.40$. This is not surprising since the value
$\kappa =0.40$ was obtained for a zero pressure gradient boundary layer, which is not satisfied in TC turbulence (Huisman et al. Reference Huisman, Scharnowski, Cierpka, Kähler, Lohse and Sun2013b; Ostilla-Mónico et al. Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014). With the increase of polymer concentration,
$u^+_\theta$ shift upward, and their slopes are nearly the same as that of the Newtonian case. We stress that even at MDR, the slope of
$u^+_\theta$ is not changed by polymers.
(a) Azimuthal velocity profiles near the inner cylinder for varying polymer concentration at
$Re=1.3\times 10^4$. Here,
$u^+_\theta =y^+$ is the viscous sublayer profile. The logarithmic law of the wall
$u^+_\theta =1/\kappa \ln y^++B$ with typical values of
$\kappa =0.40$ and
$B=5.0$ is also shown for comparison. The logarithmic law with values of
$\kappa =1/11.7$ and
$B=-17$ is the maximum drag reduction asymptote reported in channel flow. (b) Parameter
$\kappa$ as a function of
$Re$ for both Newtonian case and viscoelastic case under the maximum drag reduction condition.
$\kappa$ from Huisman et al. (Reference Huisman, Scharnowski, Cierpka, Kähler, Lohse and Sun2013b) and Ostilla-Mónico et al. (Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014) are also shown.

We extract the slope of
$u^+_\theta$ at three different
$Re$. Least-square fitting is applied to
$u^+_\theta$ for
$y^+>50$, as suggested by Huisman et al. (Reference Huisman, Scharnowski, Cierpka, Kähler, Lohse and Sun2013b). Figure 13(b) shows
$\kappa$ for both Newtonian case and viscoelastic case at MDR, where the experimental results from Huisman et al. (Reference Huisman, Scharnowski, Cierpka, Kähler, Lohse and Sun2013b) and numerical results from Ostilla-Mónico et al. (Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014) are also shown for comparison. Here,
$\kappa$ decreases with
$Re$, and is in good agreement with previous results. Additionally,
$\kappa$ for the Newtonian case and viscoelastic case at MDR are nearly overlapping, suggesting that polymers do not alter the slope of the mean velocity in the current parameter range of TC turbulence.
4. Conclusions
We present an experimental study on polymer drag reduction in TC turbulence within the Reynolds number range of
$4 \times 10^3 < Re < 2.5 \times 10^4$. Two polyacrylamide polymers with different molecular weights are used with the corresponding elastic number smaller than
$0.01$. The flow thus is inertia-dominated. The system drag is measured by a rheometer with high precision. The local flow field is measured by planar PIV and LDA.
It is found that the drag reduction rate increases with polymer concentration and approaches the MDR limit. At MDR, the friction factor follows a power-law scaling with an effective exponent
$-0.58$, i.e.
$C_f \sim Re^{-0.58}$, which is close to the exponent reported in channel/pipe flows (Virk et al. Reference Virk, Mickley and Smith1970). However, the drag reduction rate in TC turbulence is about
$20\,\%$, which is much lower than that in rectilinear flows at similar Reynolds numbers. In addition, the Reynolds shear stress does not vanish, and the velocity profile only shifts upward with its slope in the logarithmic layer unchanged at MDR. These behaviours are reminiscent of the low-drag-reduction regime reported in channel flow (Warholic et al. Reference Warholic, Massah and Hanratty1999).
By separating the advective angular velocity flux into the mean flow (Taylor vortex) and turbulent flow (Reynolds shear stress) contributions, we show that the turbulent contribution is strongly reduced, while the mean flow contribution is only slightly reduced. In the current parameter range of TC turbulence, mean flow contribution accounts for more than
$50\,\%$ of the total angular velocity flux in the bulk (Brauckmann & Eckhardt Reference Brauckmann and Eckhardt2013). The lower drag reduction observed in TC turbulence thus originates from the observation that the major effect of polymers is on the turbulent flow and the mean flow is slightly affected. We further reveal that the above results can be traced to the fact that polymers steady the velocity boundary layer, and reduce the emission rate of both fast and slow plumes. The reduced number of intense plumes detached from the boundary layer in turn feeds less energy into the bulk turbulence, resulting in weaker turbulent fluctuations and Reynolds shear stress. In the near-wall region, polymers also redistribute the kinetic energy among different scales: the small-scale Görtler vortices are highly suppressed and the large-scale azimuthal velocity streaks become slightly more energetic.
At high Reynolds numbers, the effects of polymers are predominantly observed at the small scales of turbulence, as reported in previous studies (Ouellette et al. Reference Ouellette, Xu and Bodenschatz2009; Perlekar, Mitra & Pandit Reference Perlekar, Mitra and Pandit2010; Xi et al. Reference Xi, Bodenschatz and Xu2013; Zhang et al. Reference Zhang, Bodenschatz, Xu and Xi2021). We here show that the effects of polymers on the large-scale secondary flow structures is only marginal. When secondary flow structures are statistically persistent and dominate the global transport properties of the system, the drag reduction efficiency of polymers is diminished. At higher Reynolds numbers, the effects of Taylor vortex diminish and turbulent plumes dominate (Ostilla-Mónico et al. Reference Ostilla-Mónico, Van Der Poel, Verzicco, Grossmann and Lohse2014; Grossmann et al. Reference Grossmann, Lohse and Sun2016), we would expect that the effects of polymers and the drag reduction rate will become more pronounced. However, whether the vanishing Reynolds shear stress and asymptotic velocity profile can be observed at MDR remains unclear. The current work represents an advancement by identifying the similarities and differences in drag reduction mechanisms across different turbulence flow systems, setting the stage for further research into turbulent flows with polymer additives at high-Reynolds-number regimes in TC turbulence.
Funding
This work is financially supported by the National Natural Science Foundation of China under grant nos 11988102, 12125204, 12388101, 12402298 and 12402299, the New Cornerstone Science Foundation through the New Cornerstone Investigator Program, the XPLORER PRIZE and the 111 Project of China through grant no. B17037.
Declaration of interest
The authors report no conflict of interest.
Appendix A. Viscosity measurement
The shear viscosity of the polymer solutions is measured by the Discovery Hybrid Rheometer equipped with a cone-plate geometry, whose diameter is 40 mm and angle is
$2^\circ$. For pam5e6, the shear viscosity
$\mu _{\dot {\gamma }}$ is nearly constant and does not show the shear-thinning effect over the shear rate range of
$3<\dot {\gamma }<400$. For pam2e7,
$\mu _{\dot {\gamma }}$ slightly decreases with
$\dot {\gamma }$ and then approaches a constant when
$\dot {\gamma }>200$. To evaluate the extent of shear-thinning of the working fluids, we use the average gradient of the viscosity curve (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024),
whose value is less than or equal to 1. The smaller the value of
$n_\mu$, the more significant the shear-thinning effect of the working fluid. In TC flow, the average shear rate in the gap is
$\omega _ir_i/d$, which is larger than 80 s
$^{-1}$ in this study. Here,
$n_\mu$ is calculated from the viscosity curve in figure 14 over
$80<\dot {\gamma }<400$. For pam5e6,
$n_\mu >0.992$ at
$\phi \leq 100$ ppm. For pam2e7,
$n_\mu >0.978$ at
$\phi \leq 4$ ppm. At these values of
$n_\mu$, the shear-thinning effect can be practically neglected, as in many previous studies (Lacassagne et al. Reference Lacassagne, Cagney, Gillissen and Balabani2020; Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024). We thus average the measured shear viscosity over the range of
$80<\dot {\gamma }<400$ to obtain an averaged viscosity
$\mu$, i.e.
$\mu =\langle \mu _{\dot {\gamma }} \rangle _{80<\dot {\gamma }<400}$. This value is used as the viscosity of the working fluid in this study.
Shear viscosity
$\mu _{\dot {\gamma }}$ as a function of shear rate
$\dot {\gamma }$ for (a) pam5e6 and (b) pam2e7.

From the viscosity curves, the zero-shear viscosity
$\mu _0$ can be obtained (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024). For both polymers,
$\mu _0$ increases with polymer concentration, and can be fitted with a second-order polynomial, i.e.
$(\mu _0-\mu _s)/\mu _s=[ \mu ]{\cdot } \phi +A{\cdot } \phi ^2$, where
$\mu _s$ is the viscosity of the solvent,
$[ \mu ]$ the intrinsic viscosity and
$A$ a fitting parameter. The overlap concentration
$\phi ^*$ is estimated from
$[ \mu ]$ by the relation
$\phi ^*=0.77/[ \mu ]$ (Boulafentis et al. Reference Boulafentis, Lacassagne, Cagney and Balabani2024). Here,
$\phi ^*=680$ and 17 ppm for pam5e6 and pam2e7, respectively.
Appendix B. Dimensionless torque
The dimensionless torque,
$G$, measured in this study and from Lewis & Swinney (Reference Lewis and Swinney1999) are compared in figure 15. Here,
$G$ in the present study is slightly lower than that of Lewis & Swinney (Reference Lewis and Swinney1999), which can be attributed to the different radius ratio:
$r_i/r_o=0.714$ in this study and
$r_i/r_o=0.724$ reported by Lewis & Swinney (Reference Lewis and Swinney1999).
Comparison of the dimensionless torque,
$G$, from the present study in the Newtonian case and Lewis & Swinney (Reference Lewis and Swinney1999).

















































































