Open Access Article
Vikram
Rathee
a,
Alessandro
Monti
b,
Marco E.
Rosti
b and
Amy Q.
Shen
*a
aMicro/Bio/Nanofluidics Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, 904-0495, Japan. E-mail: amy.shen@oist.jp; Tel: +81(0)989823374
bComplex Fluids and Flows Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, 904-0495, Japan
First published on 16th August 2021
Shear thickening in stable dense colloidal suspensions is a reversible phenomenon and no hysteresis is observed in the flow curve measurements. However, a reduction in the stability of colloids promotes particle aggregation and introduces a time dependent rheological response. In this work, by using a model colloidal system of hard spherical silica particles (average diameter of 415 nm) with varying particle volume fractions 0.2 ≤ ϕ ≤ 0.56, we study the effect of particle stability on the hysteresis of the shear thickening behavior of these suspensions. The particle stability is manipulated by adding a simple monovalent salt (sodium chloride) in the silica suspension with varying concentrations α ∈ [0,0.5] M. For repulsive and weakly attractive suspensions, the flow behavior is history independent and the shear thickening behavior does not exhibit hysteresis. However, significant hysteresis is observed in rheological measurements for strongly attractive suspensions, with shear history playing a critical role due to the dynamic nature of particle clusters, resulting in time dependent hysteresis behavior. By performing numerical simulations, we find that this hysteresis behavior arises due to the competition among shear, electrostatic repulsive, van der Waals attractive, and frictional contact forces. The critical shear stress (i.e., the onset of shear thickening) decreases with increasing salt concentrations, which can be captured by a scaling relationship based on the force balance between particle–particle contact force and electrostatic repulsive force. Our combined experimental and simulation results imply the formation of particle contacts in our sheared suspensions.
Owing to the practical importance of the yield stress properties in consumer products and industrial applications, different methods have been employed to achieve yield stress behavior in dilute colloidal suspensions.28,29 For example, an electrolyte solution can be added to the suspension to modify the particle surface potential, which makes the colloidal suspensions attractive and form particle clusters mediated by adhesive bonds. Solvent mediated attractive interaction due to hydrogen bonding between the particles can also lead to particle clusters.30 These strategies can induce stress bearing networks at ϕ < ϕj, which can break down and yield under applied shear.31–33
The sheared particle network undergoes complex microstructural evolution due to Brownian forces, attractive forces, hydrodynamic forces, contact forces, and particle volume fraction.4,11,13,14,16,34–38 Such complexity increases greatly in attractive dense suspension systems,39,40 where, for example, a history dependent yield stress and time dependent flow behavior with large hysteresis in shear viscosity values has been observed in ascending and descending flow curve measurements.41 This behavior is termed as thixotropy. Note that no hysteresis in ST regime has been observed in simple yield stress suspensions.42–45
ST behavior has not been well investigated in thixotropic yield stress suspensions. A typical signature of the thixotropic yield stress fluids is the hysteresis flow behavior where the hysteresis depends on the history and waiting time imposed at an applied shear stress or shear rate values during the ascending and descending flow curves. The difference between the adhesive contacts, which give rise to yield stress, and the mechanical contacts, which cause shear thickening, is that the former hinders the particle rotation while the latter promotes the particle sliding.46–48 This argument is true for non-Brownian particles having attractive interaction, but might become invalid for Brownian particles due to diffusion limited particle aggregations.49 Brownian suspensions in the presence of attractive depletion interaction (by adding polymers in the suspensions) have shown reduced shear thickening, but the hysteresis behavior is negligible with nearly reversible flow curves.50 However, using polymers to modulate the depletion interaction also leads to enhanced shear thickening.45 Therefore, it is difficult to distinguish the effect of polymers on shear thickening since both electrostatic interaction and depletion attraction contribute to the shear viscosity of the suspension. Considering these factors, adding simple electrolytes in a suspension is better suited to study yield stress and shear thickening behavior in attractive colloidal suspensions.
In this work, by combining experiments with simulations, we study the effect of particle interactions on the shear thickening behavior in a dense suspension of nearly monodisperse silica particles from repulsive to attractive regimes, by varying the particle volume fraction and the salt concentration of sodium chloride (thereby the interaction potential). We find that, for purely repulsive and weakly attractive suspensions, the ST behavior is reversible and their relative shear viscosities display either one (repulsive suspensions) or two different ST exponents (weakly attractive suspensions) with respect to shear stress for σ > σc. For strongly attractive suspensions, the thixotropic yield stress and hysteric behavior are observed during flow curve measurements. By performing temporal shear measurements, we show that an increase of η at low σ values is due to shear induced particle aggregation, which can lead to hysteresis. However, a growth of η at relatively higher σ values is caused by increased contribution from the frictional contacts, which remain reversible.
The particle interaction potential (ζ-potential) is adjusted by adding a simple monovalent salt (sodium chloride, NaCl) at a varying concentration α in the silica suspension. The ζ-potential of the suspension is measured by using a zeta-potential and particle size analyzer (ELSZ-2000, Otsuka Electronics). The measured ζ potential magnitude decreases with increasing α from ≈ −73 mV at α = 0.01 M, −33 mV at α = 0.1 M, −24.6 mV at α = 0.2 M, to −17.4 mV at α = 0.4 M.
000 Pa. Each stress value is applied for 5 s (unless stated otherwise), and the average shear viscosity η is calculated.
![]() | (1) |
i denote the mass and inertia tensor of the ith particle, while its translational and angular velocities are indicated with the symbols ui and wi, respectively. In eqn (1), FH,C,Eij and TH,C,Eij designate the forces and torques considered in the balance due to the particle–fluid and particle–particle interactions, i.e., the contributions due to hydrodynamics, inelastic contacts, and electrochemical (repulsive and attractive) potentials, respectively (the subscript j refers to one of the neighbouring particles for hydrodynamics NH, contacts NC or electrochemical potentials NE). In particular, the hydrodynamic contribution has been modelled as a combination of the Stokes drag and a short range lubrication force due to the relative motion of particles that squeezes the flowing fluid in the narrow gaps between them.11,53 The contribution of the contacts, instead, has been simulated using very stiff spring-dashpot systems, placed along the normal and tangential directions to mimic the friction due to the surface roughness,54 that start operating when the pair of particles becomes overlapped. Note that the tangential component of the contacts satisfies the Coulomb's law, |FCtangential| ≤ μC|FCnormal|, where μC = 0.5 is the friction coefficient set to match the experimental data. The high stiffness of the spring-dashpot systems is set to avoid large overlapping regions, thus replicating rigid particles.11 Finally, the electro-chemical contribution has been modelled as a combination of a distance-decaying repulsive force (with the distance-decay length scale set by the screening length) and an attractive force in the van der Waals form.44
Normalizing the set of eqn (1) by the reference viscous force, Fvis = ηs
a2 (where ηs is the viscosity of the solvent,
the shear-rate applied to the suspension and a is the typical radius of the particles), and applying the Buckingham theorem, we obtain three non-dimensional numbers. The first one is the Stokes number,
![]() | (2) |
![]() | (3) |
≫ 1 ∼ O(105).
The third dimensionless parameter refers to the time-scale introduced by the electrochemical contribution that restores the shear-rate dependency of non-Brownian suspensions,11
![]() | (4) |
0 = F/ηsa2 is the shear-rate due to the electrochemical contribution. Thus, by tuning F appropriately, shear-rate dependency can be imposed on the suspension.
The particle–particle and particle–fluid contributions described above cause mechanical stresses in the suspension that can be measured by means of the stresslet theory,55
![]() | (5) |
is the strain-rate tensor, Σ is the bulk stress tensor, V the volume of the domain and
H,C,Eij the stresslets due to the interactions. In this work we are mainly interested in the relative viscosity of the suspension, defined as ηr = Σ12/ηs
.
The Newton–Euler equations eqn (1) are numerically integrated with the modified velocity–Verlet scheme.56 The particles are immersed in a cubic box with side length L/a that depends on the volume fraction ϕ. At the edges of the domain, to mimic a plain shear flow, the Lees–Edwards boundary conditions57 are adopted to remove the effects of the walls. The interactions between the pair of particles are tackled via a highly parallelisable, fixed-radius near-neighbour algorithm. For more details on the numerical algorithm, please refer to Monti et al.58
59). At sufficiently high salt concentrations (α ≥ 0.3 M), the suspension becomes attractive and particles form aggregates.
By systematically varying ϕ and α, we perform detailed shear rheological characterizations to understand the shear thickening behavior for suspensions progressing from repulsive, weakly attractive, to strongly attractive states. To isolate the viscosity contribution from the particles, the relative viscosity ηR = η/ηs is used in the subsequent analysis, with η the shear viscosity of the suspension and ηs the solvent viscosity.
59). For the same ϕ values, the critical stress σc decreases with increasing α and ηR are also larger at salt-free conditions (α = 0 M) for σ ≤ σc, due to the electroviscous effects of diffuse layers.49,61 The values of σc lie in 370 ± 80 Pa at α = 0 M and ϕ < 0.56, in the CST regime. However, a major reduction in σc is observed in the DST regime (ϕ → ϕj).
Increasing the monovalent salt concentration to α > 0.2 M, the attractive contribution becomes more relevant. We next consider a suspension with ϕ = 0.45 and α = 0.3 M. Due to attractive particle interactions, the suspension forms discrete aggregates in equilibrium (see Fig. S2 in the ESI†
59) and becomes weakly attractive. The flow curve shows an initial shear thinning behavior with increasing σ, and transitions directly to the shear thickening regime (Fig. 2). Above σc ≈ 10 Pa, the ST exhibits two distinct ST exponents with β = 0.2 (σ > 10 Pa) and 0.1 (σ > 200 Pa), respectively. The reversible flow curve now displays a small hysteresis at σ < 100 Pa (hollow symbols in Fig. 2A) resulting from the dynamical nature of colloidal aggregates. To quantify if this hysteresis behavior is time dependent, we perform creep measurements where a constant σ is applied for 600 s (Fig. 2B). To ensure that the initial conditions are the same for all measurements, we apply a pre-shear stress of ∼ 5.4 Pa (corresponds to the minimum ηR shown in the flow curve) for 100 s. In the shear thickening regime of 30 ≤ σ ≤ 800 Pa, Fig. 2(B) shows that ηR is time independent and increases with increasing σ. However, at σ = 5 Pa and in particular σ = 0.5 Pa (corresponding to the shear thinning regime in the flow curve), some time dependent fluctuations in ηR are observed. This fluctuation is possibly caused by the dynamic aggregation and breakage of particle clusters under the competition between shear forces and interparticle attractive forces. The observed fluctuation could also be partially due to the noise and fluctuation at low shear stress values from stress controlled rheometers.
59). The forward flow curve measurement (filled symbols in Fig. 3(A)) exhibits similar behavior to those of ϕ = 0.45 and α = 0.3 M: shear thinning behavior is followed by shear thickening trend with increasing σ, displaying ηR ∝ σβ with three ST exponents of β ∼ 0.23, 0.01, and 0.1 respectively. During the reversible flow curve run, we observe a distinct hysteresis in η with reduced ηR values for σ < 200 Pa, showing two characteristic ST exponents of 0.26 and 0.1.
We now investigate whether the ST behavior is time dependent at ϕ = 0.46 and α = 0.4 M. During the forward flow curve measurement, we hold different waiting times (5 s, 60 s, 240 s, and 450 s) at each applied σ value. The relative viscosity ηR displays dramatic difference for different waiting times at σ < 75 Pa, likely caused by the shear induced aggregation kinetics. In particular, the flow procedure with the longest waiting time (filled circles in Fig. 3B) leads to the highest ηR and the strongest shear thinning behavior in the low σ < 75 Pa regime. However, a near constant value of ηR = 46 is observed at σ > 75 Pa, independent of the waiting times (see Fig. 3B). Note that the shear thickening behavior shown in Fig. 3(B) is based on the waiting time of 5 s protocol indicated in Section 2.2.
To verify this temporal behavior, we further perform creep measurements. To ensure the initial conditions being the same for all measurements, we have applied a pre-shear stress of ∼ 100 Pa for 100 s. This pre-shear stress value is selected since ηR values are independent of waiting times at 100 Pa (see Fig. 3(B)). Next, a constant shear stress at different values (5–750 Pa) is applied for roughly 600 s. In contrast to the case at smaller α (i.e., Fig. 2(B)), the creep measurements here show that ηR increases with time and does not reach a steady state in 600 s for σ < 100 Pa. At higher σ (i.e., 400 Pa and 750 Pa), smaller steady state values of ηR ∼ 34 are obtained, likely corresponding to the shear enhanced particle alignment under high shear stresses (Fig. 3(C)).
After pre-shearing the suspension at 100 Pa for 100 s, we apply σ = 1 Pa and observe that ηR continues to increase and never reaches a saturation value even after 6000 s (see Fig. 4A), indicating that the particle network grows over time, a signature of thixotropic yield stress behavior. To further confirm the value of the yield stress σy, in a new measurement, after pre-shearing the suspension at 100 Pa for 100 s, we decrease σ incrementally from 100 Pa, 50 Pa, 30 Pa, 20 Pa, 10 Pa, to 5 Pa. For each applied σ, a minimum strain is imposed to ensure that the suspension reaches steady state.62 We find that ηR reaches a steady state after 4000 s and fluctuates around a mean value of ηR ∼ 3400 at σ = 5 Pa (see Fig. 4B). We further decrease σ from 5 Pa to 1 Pa and observe that ηR diverges over time (Fig. 4(C)), shear rate
→ 0, implying that no flow is possible in the suspension at σ = 1 Pa, verifying that the suspension exhibits a yield stress value σy = 1 Pa. This also implies that the particles form a connected network and behave like a solid at σ ≤ 1 Pa.
We perform frequency sweep measurements at fixed strain amplitude of 1% on the suspension immediately after the forward flow curve measurements where σ is increased from 0.1 to 1000 Pa (after the pre-shear protocol) (Fig. 5A), or right after the suspension reaching the yielding conditions at σ = 1 Pa (Fig. 5B). Fig. 5A shows that the elastic modulus G′ is smaller than the viscous modulus G′′ during the frequency sweep, which implies that the network of particles is completely destroyed at high σ. However, after gradually decreasing the σ from 100 Pa to 1 Pa in the suspension (as shown in Fig. 4B and C), the subsequent frequency sweep measurements show that G′ > G′′, implying the elastic nature of the connected network of particles in the attractive suspensions (see Fig. 5B).
exp(−κh), with κ the inverse of the screening length,18ε the permittivity and ζ the measured zeta potential.
The Mason number Mn has been used in literature to capture the particle cluster size and breakup dynamics during flow, and ηR decreases with the reduction of the cluster size in the sheared suspensions. For yield stress fluids, the intermolecular forces can also be estimated as
, where C is a proportionality constant related to the bond breaking and yielding condition of the particle aggregates.63–66 The proportionality constant C can be estimated when the gradient dV/dh reaches the maximum, with V the DLVO pair potential of the suspension. In our work, C ∼ 0.01 if we assume the inter-particle separation to be 0.1 nm, the interaction potential energy to be 50 kT, with Mn = 0.06πϕ2σ/σy.36,41,63–68 For example, σ = 5 Pa corresponds to Mn = 0.2 and σy = 1 Pa corresponds to Mn = 0.04.
Fig. 6 shows how the relative viscosity ηR varies with increasing Mason number Mn under both transient (open symbols) and steady state (filled symbols) flow curve measurements corresponding to Fig. 4B. For Mn < 0.1, electrochemical forces Felectrochemical are dominant over shear forces Fshear so that attractive particle networks can be sustained.33,63,64 At 0.1 < Mn < 1.0, shear forces overcome attractive forces to break up the network into aggregates: the aggregate size decreases with increasing Mn, and the sheared suspension exhibits shear thinning behavior.66 In the steady state, at σ = 5 Pa, corresponding to Mn = 0.2, even though the particle network breaks up under applied σ (Fig. 6, close inverted triangles), the competition between the breakage and formation of the particle aggregates can lead to fluctuations in ηR ∼ 3400 (Fig. 4B). At Mn > 1, the network is broken into even smaller aggregates and individual particles, and ηR reduces further, with transient and steady flow curves overlapping each other.
, the friction coefficient μC, the non-dimensional shear-rate imposed
/
0, the ratio of the two screening lengths λs/a1, and the ratio between the repulsive and attractive forces FR/FA considered to mimic the effect of the salt concentration. In particular, for λs/a1 and FR/FA, the numbers in the bracket correspond to salt concentration α = 0 M (far left) and α = 0.4 M (far right). Note that for the non-dimensional groups, we have chosen the radius of the particles of the smallest phase as the reference radius
| N | L/a1 | a 2/a1 | V 2/V1 | St |
|
μ C |
/ 0 |
λ s/a1 | F R/FA |
|---|---|---|---|---|---|---|---|---|---|
| 512 | 17.376 | 1.09 | 1 | 10−3 | 4 × 105 | 0.5 | {10−5,…,10} | [0.0225,0.0065] | [19,1] |
Fig. 7 shows very good agreement of the relative viscosity ηR obtained from the simulation (symbols) and experimental results (solid curves). Note that the experimental value of ηR is lower than the simulated value at low σ ∼ 0.1 Pa, likely due to the noise and fluctuation at low shear stress values for stress controlled rheometers (Fig. 4).
![]() | ||
| Fig. 7 Relative viscosity (ηR) versus σ for ϕ = 0.46 and α = 0 and 0.4 M. Symbols represent numerical data at α = 0 and 0.4 M respectively, and solid lines represent experimental data. | ||
We also examine the temporal dependence of the rheological properties of the highly attractive suspension considered (ϕ = 0.46 and α = 0.4 M) to understand its hysteresis behavior. We first analyze the time-evolution of the relative viscosity contributions at different shear stress σ applied to the suspension. In particular, we consider the suspensions with applied σ = 0.2, 2, 10 and 100 Pa. Since the experimental measurements were obtained after an initial pre-shearing of the suspensions with σ = 100 Pa, an initial condition that mimics such pre-shearing has been applied to all the cases analysed. Fig. 9 illustrates the temporal evolution of ηR at different σ, highlighting contributions from the electrochemical (attractive and repulsive together) and contact forces related to the tangential friction11,58 in the dense suspension. In particular, a strong time-dependence of ηR is observed at low values of σ close to σy (see σ = 0.2 Pa, Fig. 9A). In this regime, the suspension needs very long time to adjust and reach the equilibrium condition. This is due to the continuous competition between electro-repulsive and attractive forces (red curve) that dominate the behavior of the suspension. The second contribution comes from the particle–particle contacts (green curve) which can be neglected at this low shear stress regime. This trend is consistent with the experimental observations shown in Fig. 3B with the 450 s waiting time.
When σ is increased beyond the yield stress, ηR (black curve) decreases (Fig. 7) and the waiting time required to reach the steady state reduces drastically. In this regime, the relative contributions from the electro-chemical forces (red curve) and the particle–particle contacts (green curve) exchange their roles, with the latter becoming increasingly important in relative terms of ηR, and the competition between the repulsive and attractive contributions weakens, making the suspension more stable (Fig. 8B–D).
Next, we compute the average size distribution of the particle clusters formed when the suspensions reach steady state and during the transient state, by sampling 100 snapshots over a short time window. Fig. 9 shows the distributions of particle clusters at the stationary (blue histogram) and the transient (orange histogram) state. In particular, we observe that at low σ = 0.2 Pa (Fig. 9A) the distribution of the typical cluster size in the transient regime is more fragmented, prompting the unstable behavior of the suspension due to the continuous competition between the repulsive and attractive forces that break and create aggregates of particles, respectively. At higher σ, where the suspension reaches the equilibrium state quickly and the competition between repulsive and attractive forces is almost absent, the transient and the converged distribution of the cluster size are comparable and no significant difference is observed (Fig. 9B–D).
A visual image of the clusters of particles obtained at low shear stress σ = 0.2 Pa is shown in Fig. 10, where panel A shows the maxi-cluster of particles formed when the steady state is reached and panel B shows the fragmentation effect during the transient state.
We are aware of the limitation on the maximum cluster size introduced by the number of particles used (N = 512), however the analysis carried out in this work is used mainly for qualitative confirmation and to provide an explanation of the experimental results.
For stable colloidal particles (i.e., α = 0 M and 0.29 ≤ ϕ ≤ 0.56), Fig. 11A shows that N1 becomes negative at σ > 600 Pa, which is well above σc ∼ 300 Pa shown in Fig. 1A. The particle contacts mediated by hydrodynamic or frictional forces need to be sufficiently large to induce the sign change of N1.19 At ϕ = 0.56 where discontinuous shear thickening (DST) is observed, N1 first turns negative with increasing σ, but abruptly changes to a positive value at σ ∼ 1900 Pa (see Fig. 11A), indicating that the frictional contact force becomes dominant in the DST regime for repulsive dense suspensions. This observation is similar to those reported for ST spherical colloidal particles by Royer et al.19
To illustrate the salt concentration effect, we keep the particle volume fraction ϕ = 0.46 fixed and vary the salt concentration α from 0–0.5 M. Even though the critical stress σc is decreased from 300 Pa to 15 Pa when α increases from 0 M to 0.3 M (Fig. 1B and 11C), the N1versus σ curves follow the same trend for all α (Fig. 11B), with N1 becoming negative when σ > 600 Pa.
We finally evaluate the dependence of σc on α. In the CST region (ϕ < 0.56), σc values do not vary significantly for different ϕ given a fixed α (Fig. 1), so plots in Fig. 11C hold (within errors of 20%) for all volume fractions ϕ. Since particle aggregations at α > 0.3 M change the flow curve behavior significantly, we restrict our attention to σc for α ≤ 0.3 M. Shear thickening was thought to be a shear rate-controlled phenomena in earlier studies and prior attempts have been made to estimate σc or
c dependence on the physio-chemical properties of the particles in the suspensions.25,69 Boersma et al. predicted the relationship between the critical shear rate
c and the repulsive force in a suspension69 and observed that the critical shear rate decreased with increasing ϕ. Later, Maranzano et al. predicted the critical stress σc based on the hydrodynamics model by changing the pH levels in the suspension to adjust the particle stability. However, their model simply assumed that the interparticle separation distance was smaller than their equilibrium separation distance, and the screening length and the surface potential in their suspension was not controlled. These predictions were based on the assumptions that the particles do not form any surface contacts. However, recent studies show that the colloidal particles can form physical contacts by overcoming electrostatic repulsive forces when σ ≥ σc where σc is determined by Frepulsive. For example, Comtet et al.18 proposed that at the onset of shear thickening, the particle–particle contact force balances the electrostatic repulsive force with
, with x a prefactor dependent on the friction coefficient, particle roughness and shape. Applying this scaling relationship, with the repulsive force calculated as Frepulsive = 2πεaζ2κ
exp(−κh) in our dense suspensions (0.29 ≤ ϕ ≤ 0.55), the best curve fit to our experimental data yields the prefactor x = 0.025 (open circles in Fig. 11C), showing good agreement between experimental and the scaling relationship. This further implies that the contact forces make important contributions to the shear thickening behavior in our sheared dense suspensions.
For 0.25 < ϕ ≤ 0.56 at α > 0.2 M, the attractive force dominates. Large particle clusters form and evolve continuously with increasing shear stress during the flow curve measurements. For moderate attractive force (i.e., ϕ ≥ 0.45 and α = 0.3 M), we observe two shear thickening exponents above σc and small hysteresis in ascending and descending flow curve measurements. For ϕ ≥ 0.45 at even higher salt concentration (i.e., α ≥ 0.4 M), the suspensions can form percolating networks with a measurable yield stress. In this regime, we observe strongly time-dependent behavior of the sheared suspension with large hysteresis.
In equilibrium, the cluster size and structure are determined by competing Brownian and inter-particle forces resulting in fractal-alike particle clusters.61 The size, shape and structure of these clusters change under applied shear stresses (or shear rates).8,31,70,71 For 0.25 < ϕ < 0.56 and α < 0.2 M, particles remain stable and flow curves show reversibility. In the case of moderately stable particles at α = 0.2 M, the interparticle separation in equilibrium remains sufficiently large such that the van der Waals forces do not play any significant role, and thus hysteresis is absent. At higher salt concentrations α > 0.2 M, the repulsive force becomes weaker so that the particle clusters can form which can reorganise under shear. The dynamic nature of the particle clusters causes hysteresis in flow curve measurements. This behavior is illustrated in the state diagram with a dividing line around α → 0.25 M that separates the hysteresis versus the no hysteresis domains in the shear thickening regime (at ϕ > 0.25), see Fig. 12. Note that the interparticle distance decreases significantly as ϕ → ϕj, so the particles can form clusters even when the suspension is moderately stable, i.e., at smaller salt concentrations.
Footnote |
| † Electronic supplementary information (ESI) available. See DOI: 10.1039/d1sm00971k |
| This journal is © The Royal Society of Chemistry 2021 |