betaFeedback

WorksPhysics and Astronomy

Morphogenesis of bacterial colonies in liquid crystalline environments

Abstract

Natural bacterial habitats are often complex fluids with viscoelastic and anisotropic responses to stress; for example, they can take the form of liquid crystals (LCs), with elongated microscopic constituents that collectively align while still retaining the ability to flow. However, laboratory studies typically focus on cells in simple liquids or complex fluids with randomly-oriented constituents. Here, we show how interactions with LCs shape bacterial proliferation in multicellular colonies. Using experiments, we find that in a nematic LC, cells generically form aligned single-cell-wide "chains" as they reproduce. As these chains lengthen, they eventually buckle in a highly localized manner. By combining our measurements with a continuum mechanical theory, we demonstrate that this distinctive morphogenetic program emerges because cells are kept in alignment due to the LC's elasticity; as each chain lengthens, growth-induced viscous stresses along its contour eventually overcome the elasticity of the surrounding nematic, leading to buckling. Our work thus reveals and provides mechanistic insight into the previously-overlooked role of LCs in sculpting bacterial life in complex environments.

Introduction

Many bacterial habitats can exhibit liquid crystalline (LC) order; prominent examples include host mucus [1, 2, 3], extracellular polymeric substances in biofilms [4], and dense suspensions of filamentous phages [5, 6, 7]. However, laboratory studies of bacteria typically focus on cells in isotropic environments. As a result, whether and how inhabiting a liquid crystal influences bacterial behavior remains poorly understood.

Recent work indicates that the anisotropic rheology of LCs can dramatically alter how individual cells swim [8, 9, 10, 11, 12, 13, 14, 15] and aggregate with other cells [8, 16, 17, 18, 19, 20, 21, 22]. Nevertheless, despite growing recognition that an environment’s mechanical properties, in addition to its chemistry, can influence bacterial growth [23, 24, 25, 26, 27, 28], the possible influence of liquid crystalline order on another fundamental characteristic of bacterial life—proliferation through space in a multicellular colony [29]—remains underexplored. Addressing this gap in knowledge is important to establish mechanistic principles governing how proliferating active matter interacts with complex environments [30, 31]. It also has key biological implications; indeed, in nature, many bacteria are nonmotile or lose motility [32, 33, 34, 35, 36], but continue to proliferate in colonies. Colony morphology in turn influences how cells access nutrients, cooperate and compete with each other, resist external stressors (e.g., antibiotics, predators, phage), and cause infections in hosts [28, 29, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54]. Hence, we ask: Can an environment’s liquid crystalline order affect the morphology of proliferating bacterial colonies?

Here, we demonstrate that this is indeed the case, and we elucidate the underlying biophysical mechanisms. Through experiments, we find that rod-shaped nonmotile bacteria form single-cell-wide “chains” as they proliferate in a nematic LC, all aligned along the LC director—in stark contrast to forming a random dispersion as in the conventionally studied non-LC case. The chains elongate as proliferation continues, eventually buckling after reaching a critical length. This mechanical instability occurs over a much smaller length scale within the chain than the typical Euler buckling exhibited by long, slender filaments under compression. Using a continuum mechanical theory, we trace the origin of chain formation and buckling to the competition between LC elasticity, which tends to keep cells aligned along the LC director, and growth-induced compression from frictional stresses along each lengthening chain. This work thereby reveals how liquid crystalline environments can dramatically shape proliferating bacterial colonies and provides quantitative principles to predict and control these morphodynamics more broadly.

Experimental Results

Nonmotile bacteria form long, single-cell-wide chains as they proliferate in a nematic liquid crystal. Our experiments probe the proliferation of nonmotile bacteria in nutrient-rich aqueous solutions housed in thin, flat polydimethylsiloxane (PDMS) imaging channels bonded to underlying glass substrata. To start, we use Escherichia coli as a model bacterium; the cells constitutively express green fluorescent protein (GFP) in their cytoplasm, enabling direct visualization using fluorescence confocal microscopy. In LC-free solution, the cells continually grow, divide, separate from each other, and diffuse apart, eventually forming an isotropic dispersion [Fig. 1(a), Movie S1].

We observe dramatically different behavior upon addition of disodium chromoglycate (DSCG), a nontoxic lyotropic LC [57, 58], to the solution at a concentration c=15c = 15 w/w%. Under these conditions, the solution forms a nematic [59], in which the chromonic DSCG molecules self-assemble into elongated stacks that tend to collectively align along the same direction [60], which we prescribe by rubbing the glass substrate with a diamond lapping film [61]. In this case, instead of becoming randomly-oriented, cells are retained end-to-end as they proliferate, forming single-cell-wide “chains” aligned along their long axis in the direction of LC alignment [Fig. 1(b), first frame]. As proliferation progresses, the chains elongate, eventually buckling [second frame]—bending the surrounding LC in turn, as shown by the local reorientation of the LC director field n\mathbf{n} [Fig. 1(c), Movie S2], which we visualize using crossed polarizers placed around the sample in the microscope’s light path. Intriguingly, the chains form multiple, highly-localized, sharply-curved arcs as they buckle [arrows, second frame of Fig. 1(b)–(c)]—in stark contrast to the classic Euler buckling instability of a compressed long, slender filament, which instead results in the formation of a single, long-wavelength arc. As the chains continue to elongate, these localized buckles continue to grow and new buckling events arise as well [Movie S2], eventually creating an intertwined three-dimensional network of long, serpentine, multicellular, single-cell-wide chains [Fig. 1(d)].

To test the generality of this phenomenon, we perform the same experiment using nonmotile strains of two other species of bacteria: Vibrio cholerae and Pseudomonas aeruginosa. We again observe chain formation in both cases [Fig. S4(a)–(b)], indicating that this phenomenon arises across different cell types. Together, these results demonstrate that nonmotile bacterial colonies proliferating in nematic LCs generically form large, serpentine, single-cell-wide chains.

Nematic elasticity causes chain formation. Why do chains form? One possibility is that interactions with the surrounding LC alter how the individual cells grow, somehow causing them to proliferate in chains. However, bacterial growth rates directly measured using single-cell microscopy show no appreciable differences upon LC addition to the solution [Fig. S3(A)–(B)]. Moreover, the osmotic pressure of the LC is ≈2\approx2 kPa [Supplementary Information], far lower than the pressure at which bacterial physiology is typically altered, ∼100\sim100 kPa [62, 63, 64, 65]—further arguing against this possibility.

What other features of the LC drive chain formation? Close inspection of the LC director field n\mathbf{n} around a pair of cells during the nascent stage of chain formation provides a clue. The cells orient end-to-end along the imposed direction of LC alignment, with two opposing pairs of bright and dark regions flanking their outer poles apparent in the crossed polarizer micrograph [Fig. 1(e)–(f)]. These features are known as “boojums”, topological defects that form at the cell poles where the tangentially-anchored nematic director field must accommodate the curved spherocylindrical cell [66, 67]. The emergence of these point defects at the poles, rather than ring defects encircling the cell or complete disruption of the nematic order, indicates that the LC is tangentially anchored to the cell surface with a finite anchoring strength [8, 68]. That is, the elongated constituents of the LC tend to collectively align along both the direction imposed on the glass substrate as well as the surfaces of the cells. Deviations from this preferred alignment cost energy; following the so-called one-constant approximation, all such deviations are penalized by a single Frank elastic constant, KK. A minute rotation of the cell’s long axis away from the overall imposed direction of the nematic director by an angle θ=2∘\theta= 2^\circ requires Uel≈2π(lb/log⁡(2lb/a))Kθ2∼500kBTU_{\mathrm{el}} \approx2\pi(l_b/\log(2l_b/a))K\theta^2 \sim500k_{\mathrm{B}}T in energy [68– 70], where lbl_b and aa are the cell length and cross-sectional radius, respectively—far greater than the thermal energy kBTk_BT Supplementary Information. Therefore, the nematic elasticity of the LC forces cells to strongly align with it [8, 71] and continue to proliferate as aligned chains.

Two additional sets of experiments confirm this picture. First, we culture cells in nutrient-free solutions in which proliferation is arrested. When the solution does not contain LC, the cells are disconnected and randomly dispersed, with no orientational order [Fig. 2(a)(i)]; by contrast, in the LC, the cells orient along the imposed direction of LC alignment [Fig. 2(a)(ii)], consistent with our expectation and with previous observations of elongated ellipsoidal colloids in a nematic LC [72]. Second, given that the nematic elasticity driving chain formation is generated by the LC, we expect that after a chain has formed, removing the LC from the solution around it should allow the chain to disintegrate via random thermal motion of the cells. This is precisely what we observe upon flushing a suspension of chains with LC-free liquid [Fig. 2(b)], Movie S3. Taken altogether, these observations establish that nematic elasticity induced by the LC causes chains to form.

Non-proliferating bacteria in liquid-crystal-free fluid and nematic DSCG, followed by a time sequence of liquid-crystal removal

Figure 2. LC-mediated elastic forces are required for chain formation. (a) Non-proliferating E. coli cells (i) remain randomly dispersed in LC-free fluid but (ii) spontaneously align and form end-to-end aggregates in nematic DSCG (18 w/w%). Images show confocal micrographs with crossed-polarizer imaging overlaid in (ii) as in Figure 1(c),(Figure 1f). (b) Time sequence showing chain disintegration after LC removal; LC-free solvent is introduced at t=0t = 0. Cells remain in smaller side-by-side aggregates after LC removal, likely due to residual surface adhesin interactions [55, 56] that persist after cells are brought together by LC-induced elastic forces. See Movie S3 for full field of view and dynamics.

Chains buckle upon reaching a critical length. Having established the origin of chain formation, we next ask: Why do chains buckle as their constituent cells continue to proliferate [Fig. 1(b)–(c)]? And why does their buckling appear to be so different from the classic Euler buckling instability? To begin to address these questions, we first examine our experimental observations of chain buckling. Snapshots of the contour of a representative chain as it elongates are shown in Figure 3(a); it is initially straight, but buckles, becoming more tortuous, after elongating sufficiently. Direct measurement of the chain’s tortuosity, defined as its contour length LL normalized by its end-to-end distance, as it grows quantifies this observation: as shown in Figure 3(b), the tortuosity increases sharply as the chain elongates beyond a length ≈200\approx200 µm. We denote this critical buckling length by L∗L^*. Repeating this measurement for many other chains indicates that they consistently buckle at a critical length that decreases with increasing LC concentration (i.e., increasing nematic elasticity): L∗=170±90L^* = 170 \pm90 µm and 90±4090 \pm40 µm for experiments performed using c=15c = 15 w/w% and 18 w/w%, respectively [Fig. 3(c)]. In all cases, the buckling is far more localized than the long-wavelength arc formation characteristic of Euler buckling, as in Figure 1(b)–(Figure 1c).

Proliferating nonmotile bacteria form chain-like colonies in liquid crystalline environments

Figure 1. Proliferating nonmotile bacteria form chain-like colonies in liquid crystalline environments. (a) Fluorescent confocal microscopy time sequence of nonmotile E. coli proliferating in nutrient-rich LC-free fluid; cells form an isotropic dispersion as they grow, divide, and diffuse apart. (b) Same as in a, but in nematic DSCG solution (15 w/w%); cells instead remain aligned end-to-end, forming a single-cell-wide chain that eventually buckles. (c) Same as in b with crossed-polarizer imaging overlaid (directions indicated by A and P), revealing local reorientation of the LC director field nn accompanying chain buckling (bright regions). The overall far-field orientation imposed on the director by the underlying substrate is shown by the double-headed arrow. (d) Three-dimensional confocal reconstruction of a chain-like colony after 520 min, showing the serpentine, single-cell-wide internal structure. (e) Schematic of the LC director profile around a rod-shaped bacterium with weak tangential anchoring; boojums (topological defects) form at the cell poles where the director field must accommodate the curved cell surface. (f) Fluorescent confocal micrograph with crossed-polarizer imaging overlaid (directions indicated by A and P) of two cells dispersed in nematic DSCG solution. The image shows the local reorientation of the LC director field nn around the cell, with the emergence of boojums at the cell poles. The overall far-field orientation imposed on the director by the underlying substrate is shown by the double-headed arrow. Experiments showed are performed using a microfluidic device 25 µm in height, 25 mm in width, and 22 mm in length.

Successive centerline configurations, tortuosity as a function of contour length, and measured buckling lengths

Figure 3. Characterization of chain buckling. (a) Successive centerline configurations of a buckling chain in 18 w/w% DSCG, with colors indicating time. The chain initially grows straight before developing highly localized, sharply curved buckles. (b) Tortuosity (contour length divided by end-to-end distance) as a function of contour length for the chain in a; the sharp increase marks the onset of buckling at the critical length L∗L^*. (c) Measured buckling lengths for chains in 15 w/w% (light blue, n=13n = 13) and 18 w/w% (light orange, n=38n = 38) DSCG. Horizontal lines indicate means; the critical buckling length decreases with increasing LC concentration (** indicates p≤0.01p \le0.01, two-sample tt-test).

Mathematical Modeling

Force balance of a growing continuum chain. Our experiments highlight the pivotal role of nematic elasticity in causing chains to form and influencing how they buckle. Indeed, while the bacterial chains are composed of discrete, unattached cells, the nematic elasticity of the surrounding LC effectively couples adjacent cells together—causing deformations of a given chain to be mechanically correlated over distances much larger than a single cell length. Hence, to shed light on the physics underlying chain growth and buckling, we turn to mathematical modeling. In particular, we describe each chain as a continuous, slender, elastic filament [Fig. 4] of fixed radius aa and a length that grows exponentially in time tt, L(t)=L(0)eαtL(t)=L(0)e^{\alpha t}, where α\alpha is the single-cell growth rate. The chain centerline is parameterized by the arclength s∈[−L(t)/2,L(t)/2]s \in[-L(t)/2,L(t)/2] as x0(s,t)\boldsymbol{x}_0(s,t), with centerline unit tangent vector s^=∂sx0\hat{\boldsymbol{s}}=\partial_s\boldsymbol{x}_0. The material velocity at each point, V(s,t)\boldsymbol{V}(s,t), can be decomposed as V=αss^+v(s,t)\boldsymbol{V}=\alpha s\hat{\boldsymbol{s}}+\boldsymbol{v}(s,t), where the first term reflects tangential growth.

The force per unit length acting on the chain at each station (point along the centerline) is then given by

f(s,t)=fviscous(s,t)+fLC(s,t)+∂s[T(s,t)s^],(1)\boldsymbol{f}(s,t)=\boldsymbol{f}^{\mathrm{viscous}}(s,t)+\boldsymbol{f}^{\mathrm{LC}}(s,t)+\partial_s\left[T(s,t)\hat{\boldsymbol{s}}\right], \tag*{(1)}

where fviscous\boldsymbol{f}^{\mathrm{viscous}} is the viscous drag per unit length induced by the LC, fLC\boldsymbol{f}^{\mathrm{LC}} is the elastic restoring force per unit length from the deformed nematic, and TT is the internal tension (or compression if T<0T<0) arising from bacterial growth. Because viscous damping is much larger than inertial effects in our experiments, f≡0\boldsymbol{f}\equiv0 at all times [73]; since the bacterial chain grows very slowly compared to the other time scales governing the LC flow, the LC is assumed to be in quasi-static equilibrium at all times. Moreover, because the chain is slender, we can use the resistive force approximation, which assumes the drag at each point depends only on the local chain velocity and orientation, to calculate fviscous=−[ζ∥s^s^+ζ⊥(I−s^s^)]⋅V\boldsymbol{f}^{\mathrm{viscous}}=-[\zeta_{\parallel}\hat{\boldsymbol{s}}\hat{\boldsymbol{s}}+\zeta_{\perp}(\mathbf{I}-\hat{\boldsymbol{s}}\hat{\boldsymbol{s}})]\cdot\boldsymbol{V}; the drag coefficients ζ∥\zeta_{\parallel} and ζ⊥\zeta_{\perp} penalize motion in the chain tangent and normal directions, respectively. Eq. (1) can then be decomposed into tangential and normal components as

−ζ∥(v∥+αs)+∂sT=0,(2)-\zeta_{\parallel}(v_{\parallel}+\alpha s)+\partial_sT=0, \tag*{(2)}
−ζ⊥v⊥+(I−s^s^)⋅(fLC+T∂ss^)=0,(3)-\zeta_{\perp}\boldsymbol{v}_{\perp}+(\mathbf{I}-\hat{\boldsymbol{s}}\hat{\boldsymbol{s}})\cdot\left(\boldsymbol{f}^{\mathrm{LC}}+T\partial_s\hat{\boldsymbol{s}}\right)=0, \tag*{(3)}

where the material velocity v=v∥s^+v⊥\boldsymbol{v} = v_{\parallel}\hat{\boldsymbol{s}}+\boldsymbol{v}_{\perp} with s^⋅v⊥=0\hat{\boldsymbol{s}}\cdot\boldsymbol{v}_{\perp}=0, and s^⋅fLC=0\hat{\boldsymbol{s}}\cdot\boldsymbol{f}^{\mathrm{LC}}=0 [detailed in Supplementary Information]. Assuming small chain displacements for simplicity, we approximate the deformed position as x0(s,t)≈sx^+u(s,t)y^\boldsymbol{x}_0(s,t)\approx s\hat{\boldsymbol{x}}+u(s,t)\hat{\boldsymbol{y}}, where x^\hat{\boldsymbol{x}}, y^\hat{\boldsymbol{y}}, and z^\hat{\boldsymbol{z}} denote the unit Cartesian vectors and uu is the vertical displacement. We then have v⊥≈∂tuy^\boldsymbol{v}_{\perp}\approx\partial_tu\hat{\boldsymbol{y}} and s^≈x^+∂suy^\hat{\boldsymbol{s}}\approx\hat{\boldsymbol{x}}+\partial_su\hat{\boldsymbol{y}}, giving

ζ⊥∂tu=T(s,t)∂ssu+y^⋅fLC(s,t).(4)\zeta_{\perp}\partial_tu=T(s,t)\partial_{ss}u+\hat{\boldsymbol{y}}\cdot\boldsymbol{f}^{\mathrm{LC}}(s,t). \tag*{(4)}

To close this system of equations, we compute the tension TT and LC restoring force fLC\boldsymbol{f}^{\mathrm{LC}} for the two principal cases that arise in the experiments: when the chain is (i) free at both ends (“free-free”) or (ii) fixed at one end, e.g., due to pinning on the glass substrate and free at the other (“fixed-free”). Details are provided in the Supplementary Information.

In brief, integrating Eq. (13a) yields the tension T(s,t)T(s,t), which depends on α\alpha, ζ∥\zeta_{\parallel}, L(t)L(t), and the endpoint tension TLT_L. In both cases (i) and (ii), as growth progresses, the tension eventually becomes negative (i.e., compressive)—introducing the possibility of buckling [74]. However, unlike in classical Euler buckling, the load is distributed along the length of the chain due to the viscous drag, just as in the buckling of sedimenting filaments [75] and filaments placed in compressional flows [76, 77, 78]. To compute the LC elastic restoring force, we assume that the LC is deep in the nematic state with director field n(x,t)\boldsymbol{n}(x,t) and ∣n∣=1|\boldsymbol{n}|=1 [79]. With n→x^\boldsymbol{n}\to\hat{\boldsymbol{x}} as ∣x∣→∞|x|\to\infty, we assume for simplicity that n(x)≈x^+p(x)\boldsymbol{n}(x)\approx\hat{\boldsymbol{x}}+\boldsymbol{p}(x) with x^⋅p=0\hat{\boldsymbol{x}}\cdot\boldsymbol{p}=0, for a small perturbation in the molecular orientation p=O(∂su)\boldsymbol{p}=O(\partial_s\boldsymbol{u}). Using the one-constant approximation for the Frank elastic energy, ∇2p=0\nabla^2\boldsymbol{p}=0 [79], enabling us to compute the LC traction (surface force per unit area) acting on the chain. We thereby compute fLC(s,t)\boldsymbol{f}^{\mathrm{LC}}(s,t), which depends on KK, aa, L(t)L(t), and the anchoring strength WW quantifying the energetic penalty for LC molecules to deviate from tangential alignment with the chain surface.

Parameters characterizing bacterial chain growth in a liquid crystal

To distill the relevant physics from our theoretical model, we first reduce the number of parameters in Eqs. (1)–(27) by scaling lengths and time by the constant chain radius aa and inverse growth rate α−1\alpha^{-1}, respectively. Using these natural scales, we obtain four key dimensionless parameters that govern chain growth:

  • The chain aspect ratio, Λ≡L/(2a)\Lambda\equiv L/(2a), which increases to ≫1\gg1 as growth progresses.

  • The drag anisotropy ratio, χ≡ζ⊥/ζ∥\chi\equiv\zeta_{\perp}/\zeta_{\parallel}. Approximating the chain as a filament moving in an isotropic fluid a distance hh from a wall, with h≪Lh\ll L [80], ζ∥≈[2π/log⁡(2h/a)]μ∥\zeta_{\parallel}\approx[2\pi/\log(2h/a)]\mu_{\parallel} and χ≡ζ⊥/ζ∥≈2μ⊥/μ∥\chi\equiv\zeta_{\perp}/\zeta_{\parallel}\approx2\mu_{\perp}/\mu_{\parallel} for the LC perpendicular and parallel viscosities μ⊥\mu_{\perp} and μ∥\mu_{\parallel}, respectively. Typical viscosity ratios for DSCG range from μ⊥/μ∥≈2\mu_{\perp}/\mu_{\parallel}\approx2 to 66 [81, 82], so we fix χ≈5\chi\approx5.

  • The dimensionless anchoring strength, w≡aW/Kw\equiv aW/K. We estimate that our experiments probe w∼10−2w\sim10^{-2}–10010^{0} [70, 83, 84] (Methods), consistent with the weak tangential anchoring indicated by the micrograph in Figure 1(f).

  • The Ericksen number, Er≡a2ζ∥α/(2πK)\mathrm{Er}\equiv a^2\zeta_{\parallel}\alpha/(2\pi K), which compares viscous forces introduced by chain growth to elastic forces arising from LC deformation. We estimate (Methods) that our experiments probe Er≈3.0×10−7\mathrm{Er}\approx3.0\times10^{-7} (c=15 w/w%c=15\ \mathrm{w/w}\%) and 7.7×10−77.7\times10^{-7} (c=18 w/w%c=18\ \mathrm{w/w}\%). Er≪1\mathrm{Er}\ll1, which highlights the dominant role of nematic LC elasticity in controlling chain growth.

As detailed in the Supplementary Information, we neglect the endpoint tension TLT_L, because it is orders of magnitude smaller than the viscous stress a2ζ∥αa^2\zeta_{\parallel}\alpha, thereby simplifying our system of equations further.

Competition between LC elasticity and growth-induced stresses along the chain contour drives localized buckling. Having nondimensionalized the governing equations, we predict if and when a growing chain will buckle by performing a linear stability analysis of the straight chain state [detailed in the Supplementary Information]. That is, we ask: at what chain length (or equivalently, aspect ratio Λ\Lambda) does a small perturbation to the straight configuration begin to grow rather than decay? To address this question using our model, we perturb the chain centerline with a small displacement u=aeστU(ξ)u=a e^{\sigma\tau}U(\xi) [Figure 4], where ξ≡2s/L(t)∈[−1,1]\xi\equiv2s/L(t)\in[-1,1] and τ≡αt\tau\equiv\alpha t; here, σ\sigma is the growth rate of the perturbation. We then expand the non-dimensional displacement U(ξ,τ)U(\xi,\tau) as a Fourier series with time-dependent coefficients and substitute into the dimensionless form of Eq. (27), with Λ(τ)\Lambda(\tau) assumed to be frozen at its instantaneous value ≈Λ\approx\Lambda. This procedure yields an eigenvalue problem for the growth rate σk\sigma_k of each Fourier mode U^k\hat{U}_k as a function of Λ\Lambda:

Continuum model of a growing bacterial chain in a nematic LC

Figure 4. Continuum model of a growing bacterial chain in a nematic LC. The chain is treated as a slender filament of fixed radius aa and growing length L(t)L(t). The transverse (y^\hat{y}) displacement u(s,t)u(s,t) is parameterized by the arclength ss. Growth generates internal compression T(s,t)T(s,t) along the chain contour due to viscous drag from the surrounding LC, while chain deflections are resisted by the elastic restoring force from the deformed nematic director field (wavy lines, bottom).

γ(χ)σkU^k=(χ2+qk23−14−qk3Er⋅Λ3Zk)U^k+∑m≠kJkmU^m,(5)\gamma(\chi)\sigma_k\hat{U}_k=\left(\frac{\chi}{2}+\frac{q_k^2}{3}-\frac{1}{4}-\frac{q_k^3}{\mathrm{Er}\cdot\Lambda^3 Z_k}\right)\hat{U}_k+\sum_{m\ne k}J_{km}\hat{U}_m, \tag*{(5)}

for the free-free case, where ZkZ_k quantifies how strongly the LC resists chain deformations at mode kk and JkmJ_{km} are coupling coefficients between modes, both derived in the Supplementary Information. The fixed-free case yields an analogous eigenvalue problem with modified mode functions and coupling coefficients. Chain buckling then occurs when any of the modes becomes unstable (σk>0\sigma_k>0), which defines the critical buckling aspect ratio Λ∗\Lambda^* corresponding to the critical length L∗L^*.

Our numerical solutions of Λ∗\Lambda^* from Eq. (4) are shown in Figure 5 (solid curves) for the exemplary free-free case. The results for the fixed-free case are similar in magnitude, but shifted slightly lower by a factor of ≈1.5\approx1.5, because a fixed end cannot relieve growth-induced compression by moving outward, unlike a free end—which is reflected in differences in T(s,t)T(s,t) between the two cases [Supplementary Information]. Inspired by the solution when the Fourier modes are decoupled (Jkm=0J_{km}=0), we also approximate our results as

Theoretical predictions for the critical buckling length

Figure 5. Theoretical predictions for the critical buckling length. (a) Dimensionless critical buckling length Λ∗≡L∗/(2a)\Lambda^* \equiv L^*/(2a) as a function of the Ericksen number Er⁡≡a2∣α∣/(2πK)\operatorname{Er} \equiv a^2|\alpha|/(2\pi K) for different values of the dimensionless anchoring strength w≡aW/Kw \equiv aW/K. Solid curves show numerical solutions of the linear stability analysis Eq. (4); dashed curves show the analytical approximation Eq. (5). The scaling Λ∗∼Er⁡−1/2\Lambda^* \sim\operatorname{Er}^{-1/2} reflects the balance between growth-induced viscous compression and LC elastic restoring forces. (b) Λ∗\Lambda^* as a function of ww for different values of Er⁡\operatorname{Er}. For weak anchoring (w≪1w \ll1), the director can slip relative to the bacterial cell surface, reducing the effective elastic coupling and allowing buckling at shorter chain lengths; for strong anchoring (w≫1w \gg1), the coupling saturates and Λ∗\Lambda^* becomes independent of ww. Both panels show predictions for the free-free boundary conditions and χ≈5\chi\approx5 [Methods]. Symbols show experimentally measured buckling lengths at 15 w/w% (filled circle) and 18 w/w% (open circle) DSCG, plotted at their corresponding Er⁡\operatorname{Er} values; comparison with theory yields inferred anchoring strengths w≈0.01w \approx0.01, corresponding to W≈2.6×10−7 J/m2W \approx2.6 \times10^{-7}\,\mathrm{J/m^2}.

Λ∗≈a1Er−1log⁡(a2Er−1)+4/w,(6)\Lambda^*\approx\sqrt{\frac{a_1\mathrm{Er}^{-1}}{\log(a_2\mathrm{Er}^{-1})+4/w}}, \tag*{(6)}

which provides a useful analytical expression for Λ∗(Er,w)\Lambda^*(\mathrm{Er},w), shown by the dashed curves; here, {a1,a2}\{a_1,a_2\} are numerical constants equal to {2.91,16.2}\{2.91,16.2\} and {1.26,0.664}\{1.26,0.664\} for the free-free and fixed-free cases, respectively.

To test our theoretical predictions, we compare the measured critical buckling lengths [Figure 3(c)] with those predicted by our theory. As shown by the symbols in Figure 5(b), which indicate our measured values at their corresponding Ericksen numbers (Er⁡≈3.0×10−7\operatorname{Er} \approx3.0 \times10^{-7} and ≈7.7×10−7\approx7.7 \times10^{-7}, respectively), we find good agreement with the free-free theoretical predictions with dimensionless anchoring strengths w≈1.1×10−2w \approx1.1 \times10^{-2} and 0.73×10−20.73 \times10^{-2}, respectively; comparing to the fixed-free prediction instead yields slightly larger values, w≈3.0×10−2w \approx3.0 \times10^{-2} and 2.0×10−22.0 \times10^{-2}, respectively. These values correspond to dimensional anchoring strengths W≈2.6W \approx2.6–2.8×10−7 J/m22.8 \times10^{-7}\,\mathrm{J/m^2} and 7.27.2–7.6×10−7 J/m27.6 \times10^{-7}\,\mathrm{J/m^2}, respectively; because our experiments likely include chains with a mixture of boundary conditions, the true anchoring strength likely lies within these ranges [Supplementary Information]. Nevertheless, these values are in excellent agreement with previous studies of bacteria and similar inclusions in nematic LCs, which found W≈10−7W \approx10^{-7}–10−5 J/m210^{-5}\,\mathrm{J/m^2} [70, 83, 84].

The good quantitative agreement between theory and experiment validates our central physical picture: that the competition between LC nematic elasticity and growth-induced viscous stresses along the chain contour drives buckling.

periment validates our central physical picture: that the competition between LC nematic elasticity and growth-induced viscous stresses along the chain contour drives buckling. Figure 5 reveals how these two effects control the critical buckling length. Consider first the dependence on the Ericksen number, Er⁡≡a2∣α∣/(2πK)\operatorname{Er} \equiv a^2|\alpha|/(2\pi K). The compressive viscous drag force ∼∣α∣L2\sim|\alpha|L^2 increases as the chain grows, eventually exceeding the critical LC elastic restoring force ∼K\sim K at L=L∗L = L^*. Balancing these forces at buckling yields Λ∗∼Er⁡−1/2\Lambda^* \sim\operatorname{Er}^{-1/2}, as shown by the curves in Figure 5(a); larger Er⁡\operatorname{Er} (faster growth or weaker LC elasticity) leads to buckling at shorter chain lengths. At the small Ericksen numbers relevant to our experiments (Er⁡∼10−7\operatorname{Er} \sim10^{-7}), LC elasticity dominates, permitting chains to grow to hundreds of cells before the accumulated compression triggers a buckling instability. Next, consider the dependence on anchoring strength. For weak anchoring (w≪1w \ll1) such as in our experiments (w∼10−2w \sim10^{-2}), the director can slip relative to the bacterial cell surface, reducing the effective elastic coupling between chain deflections and the surrounding LC. This weakened coupling lowers the restoring force resisting buckling, which thus occurs at a relatively low accumulated compressive stress and thus for shorter chains—hence, in this regime, Λ∗\Lambda^* grows with ww, as shown in Figure 5(b). By contrast, for strong anchoring (w≫1w \gg1), the LC director couples to the bacterial surface tightly and the elastic coupling saturates; further increases in ww provide no additional coupling, so Λ∗\Lambda^* becomes independent of ww, as also shown in Figure 5(b).

The theoretical analysis also explains why chain buckling differs so strikingly from classical Euler buckling. In Euler buckling, an end-applied compressive load destabilizes a single, long-wavelength mode spanning the entire filament. For a bacterial chain in an LC, by contrast, compression is not applied at the endpoints but is generated by viscous drag continuously along the chain as it grows. Simultaneously, the surrounding nematic acts as a distributed elastic foundation, penalizing deflections at all wavelengths. Together, these effects promote instability at shorter wavelengths, producing the highly localized, sharply curved buckles observed in our experiments rather than a single gentle arc. Some experimental realizations, like that shown in Figure 3, exhibit larger-wavelength deformations reminiscent of the wrinkling-to-folding transition in elastic media under large compressive loads [85]; exploring this connection further using a fully nonlinear theory will be an interesting direction for future work.

Discussion

By combining experiments and continuum mechanical theory, we have shown that liquid crystalline environments can shape bacterial colony morphology through purely physical interactions. Nonmotile bacteria proliferating in a nematic LC form long, serpentine, single-cell-wide chains aligned along the director—a morphology that arises because LC elasticity keeps proliferating cells oriented end-to-end. As these chains elongate, growth-induced viscous stresses accumulate along their contours, eventually overcoming the elastic penalty imposed by the surrounding nematic and triggering highly-localized buckling that starkly contrasts to the long-wavelength bending typical of classical Euler buckling—conceptually similar to how the elastic cytoskeleton suppresses bending of microtubules inside cells [86]. In addition to explaining all our experimental observations, our theoretical analysis provides quantitative principles that describe the dependence of chain buckling on the Ericksen number and LC–cell surface anchoring strength, motivating future studies that explore variations in these parameters further. Indeed, while our theory quantitatively captures the decrease in critical buckling length with increasing LC concentration, the spread in measured L∗L^{*} values suggests additional variability—likely from cell-to-cell differences in anchoring strength, and possibly from imperfection sensitivity or subcritical effects that could be disentangled through analysis of a fully nonlinear model.

Intriguingly, bacterial colony morphologies reminiscent of the chains studied here have been observed in various biological contexts. However, their origin is typically attributed to specific adhesive interactions holding cells together [87, 88, 89, 90]. Here, we reveal a fundamentally distinct mechanism by which proliferating bacteria can form such long, single-cell-wide chains: through mechanical interactions with a surrounding liquid crystal. This finding, which complements our previous study of “cables” formed by cells proliferating in polymer solutions [91], highlights how purely physical interactions with surrounding complex fluids—such as host mucus [1, 2, 3], extracellular polymeric substances in biofilms [4], and dense suspensions of filamentous phages [5, 6, 7]—can profoundly shape growing bacterial colonies.

What are the biological implications of chain formation? Our findings suggest that LC-mediated chain formation could influence how nonmotile bacteria explore space, access nutrients, and interact with external stressors. Long chains have large surface-area-to-volume ratios, potentially increasing susceptibility to antibiotics or phages [5, 6, 7, 51, 52, 53, 54]; alternatively, the extended chain morphology could enable colonies to reach new environmental niches [88, 92] or nutrient sources [49, 50]. The buckling instability itself may serve as a mechanical signal that triggers phenotypic changes in the cells themselves; indeed, our theory suggests that the compressive force in chains reaches ∼1 pN\sim1\,\mathrm{pN} at the onset of buckling—comparable to the forces that drive bacterial mechanosensitive responses [93], such as biofilm formation. Investigating these possibilities, both in quiescent conditions and under flow [94, 95], will be an important direction for future work.

Our work also opens new directions in soft matter physics. While colloidal assembly in LCs has been extensively studied [59, 72, 96, 97], our results highlight that qualitatively new behaviors emerge when the constituent particles can proliferate [30]. The interplay between growth-generated stresses, LC elasticity, and surface anchoring creates a rich design space for controlling colony morphology, and potentially for engineering living materials with prescribed architectures. Extending our framework to other anisotropic environments, such as cholesteric LCs or nematic gels with tunable elasticity, could reveal additional morphogenetic programs accessible through physical interactions alone [98].

Acknowledgments

We thank Z. Gitai and J. Shaevitz for useful discussions, and the laboratories of B. Austin, B. Bassler, and Z. Gitai for providing strains of E. coli, V. cholerae, and P. aeruginosa, respectively.

Funding: S.E.S. acknowledges support from NSF grant DMS-2527011. N.S.W. acknowledges support from NSF Center for the Physics of Biological Function grant PHY-1734030. This research was also supported by the National Institute of General Medical Sciences (NIGMS) of the National Institutes of Health (NIH) under Award Number R01 GM082938. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. S.S.D. acknowledges support from NSF grants CBET-1941716, DMR-2011750, and EF-2124863 as well as the Camille Dreyfus Teacher-Scholar and Pew Biomedical Scholars Programs.

Author contributions: S.G.L.C. and S.S.D. designed the overall research project; S.G.L.C. performed all experiments and experimental analyses; T.G.J.C. and S.E.S. developed the theoretical model and performed all calculations; S.G.L.C., T.G.J.C., S.E.S., N.S.W., and S.S.D. analyzed data and wrote the paper.

Competing interests: The authors declare that they have no competing interests.

Methods

Preparation of cells

Escherichia coli

We incubated an overnight culture of E. coli strain W3110 in 2 w/w% Lennox Lysogeny Broth (LB) at 30 °C. Next, we inoculated 20 μL from the overnight culture in 2 mL of LB for 3 hours such that, at the time of imaging, bacteria would reach mid-exponential phase. The E. coli constitutively express green fluorescent protein (GFP) throughout their cytoplasm and have a deletion of the flagellar regulatory gene flhDC that renders them nonmotile.

Pseudomonas aeruginosa

We incubated an overnight culture of P. aeruginosa strain PA01 in LB, supplemented with carbenicillin (200 µg/mL) to preserve the strain’s fluorescence, at 37 °C. Next, we inoculated 20 µL from the overnight culture in 2 mL of LB for 3 hours. We used a strain with a double fliC and pilA deletion that renders the cells nonmotile, and verified that these cells are nonmotile using direct visualization.

Vibrio cholerae

We incubated an overnight culture of V. cholerae O1 biovar El Tor strain C6706 in LB at 37 °C. Next, we inoculated 20 µL from the overnight culture in 2 mL of LB for 3 hours at 37 °C. We used a strain with several gene deletions: It has a deletion of pomA, which renders the cells nonmotile, and we verified that these cells are nonmotile using direct visualization. It also has deletions of rbmA, bap1, rbmC, and vpsL, which renders the cells as nonbiofilm formers.

Preparation of liquid crystal solution

We purchased liquid crystal disodium chromoglycate (DSCG) from Thermo Fisher Scientific. DSCG stock solutions were prepared in LB and put in a spinning rotor until full dissolution was observed. The DSCG solution was then filtered using a 5 µm membrane from GE Healthcare life sciences. To obtain liquid crystal solutions at the desired concentrations we diluted the stock solutions with the same solvent.

Imaging of bacteria in liquid crystal solution

coli in liquid crystal

To image fluorescent E. coli we used custom built rectangular PDMS devices with dimensions 2 mm in width, 22 mm in length, and 25 µm in height. The device was filled with the liquid crystal solution containing a well mixed bacterial population at a concentration ccell≈105c_{\mathrm{cell}} \approx10^{5}–10610^{6} cells/mL. Next, we sealed inlet and outlet ports of the PDMS channel with paraffin oil to minimize evaporation. Finally, we imaged the dish with a 20× magnification air objective [Fig. 1(a)–(c)] or 60× magnification oil objective with added 1.5× zoom [Fig. 1(f)] in a Nikon AXR inverted laser-scanning confocal microscope with the stage maintained at 30±130 \pm1 °C. We recorded images throughout the sample every 1–10 min over ∼ 6–14 hours. To obtain the images in Figure 1(c),(Figure 1f), we inserted a linear polarizer on the lightpath towards the transmittance detector.

To investigate the reversibility of chain formation [Fig. 2B], we used a transparent-walled glass-bottom petri dish sealed with an overlying PDMS slab. We additionally pierced the PDMS slab with three 20-gauge needles, sealed using optical glue, and connected to a Harvard Apparatus 11 Elite syringe pump to provide fluid in/outflow. In particular, one needle acts as an inlet for injection of the test solution containing ≈ 4×1054 \times10^{5} cells/mL suspended in 15 w/w% DSCG solution; the second acts as an inlet for injection of DSCG-free LB media; and the third acts as the outlet. We then imaged the sealed chamber from below using a Nikon A1R+ inverted laser scanning confocal microscope with the stage maintained at 30±130 \pm1 °C as before, first under quiescent no-flow conditions with the chamber filled with the test solution containing cells and LC. Once chains have formed, we then pump LC-free fluid through at a flow rate of 2.5 µL/min to remove LC-containing solution.

To determine the onset of chain buckling [Fig. 3], we (i) smooth the confocal micrographs of growing chains using a 2D Gaussian filter, (ii) binarize the images using Otsu’s method of thresholding [99], (iii) skeletonize the individual chains using the medial axis transform and thereby determine their backbones, and (iv) directly measure the tortuosity of each identified chain backbone.

aeruginosa and V. cholerae in liquid crystal

We filled individual wells of a glass-bottom 96-well plate with 100 µL of the solution and 0.1 µL of the inoculum of cells such that their initial concentration is ≈ 10610^{6} cells/mL. We stored the plate in a static 30 °C incubator and imaged the plate at different time points: For P. aeruginosa, we imaged at 120 and 840 min after the start of incubation, and for V. cholerae, we imaged at 180 and 360 min after incubation as shown in Figure S4A-B. For all images taken, we used a 20× air objective mounted in a Nikon A1R+ inverted laser scanning confocal microscope with the stage maintained at room temperature.

Nonmotile P. aeruginosa and V. cholerae forming chains as they proliferate in liquid crystalline solutions

Figure S4. Nonmotile P. aeruginosa and V. cholerae form chains as they proliferate in liquid crystalline solutions. (A) Time sequence of nonmotile P. aeruginosa proliferating in nutrient-rich fluid without (top) or with (bottom) 18 w/w% DSCG. (B) Same as in A but with nonmotile V. cholerae. The time indicates the duration elapsed after the 96-well plate containing the samples is inserted into the static incubator. The images were taken at different positions of the same well at the different time points.

Supplementary Information

Osmotic pressure of nematic liquid crystal solutions

To estimate the osmotic pressure of nematic liquid crystals (LCs) we use the expression [100]

Πosm=3npkBT,(7)\Pi_{\mathrm{osm}} = 3n_p k_B T, \tag*{(7)}

where npn_p is the number density of nematic liquid crystal molecules, kBk_B is Boltzmann’s constant, and TT is the absolute temperature. We estimate the number density as np≈4ϕ/(L2Dπ)n_p \approx4\phi/(L^2D\pi) [100, 101], where ϕ\phi is the fractional volume occupied by the LC particles, DD is the average diameter of the LC particles, and ll is the average length of the LC particles. Inputting the values of Table S1 renders Πosm=2000\Pi_{\mathrm{osm}} = 2000 Pa.

Parameter descriptionSymbolValue
Cell growth rateα\alpha2×10−4 sec−12 \times10^{-4}\,\mathrm{sec}^{-1}
DSCG parallel viscosityμ∥\mu_{\parallel}0.4 Pa⋅sec0.4\,\mathrm{Pa}\cdot\mathrm{sec} [106] (18 w/w%); 0.1 Pa⋅sec0.1\,\mathrm{Pa}\cdot\mathrm{sec} [106] (15 w/w%)
DSCG viscosity ratioμ⊥/μ∥\mu_{\perp}/\mu_{\parallel}2–6 (measured, cf. [81, 82])
Cell radiusaa1 μm1\,\mu\mathrm{m}
LC - bacteria surface anchoring energyWW10−5 J/m210^{-5}\,\mathrm{J}/\mathrm{m}^{2} [70]; 10−6 J/m210^{-6}\,\mathrm{J}/\mathrm{m}^{2} [83]; 0.2×10−5 J/m20.2 \times10^{-5}\,\mathrm{J}/\mathrm{m}^{2} [84]
DSCG average Frank elastic constantKK18.66 pN18.66\,\mathrm{pN} [9] (18 w/w%); 11.91 pN11.91\,\mathrm{pN} [9] (15 w/w%)
Volume fractionϕ\phi0.115 [107]
Average LC molecule diameterDD1.6 nm1.6\,\mathrm{nm} [107]
Average LC molecule lengthll20 nm20\,\mathrm{nm} [107]
Cell lengthlbl_b11–4 μm4\,\mu\mathrm{m}
Buckling lengthL∗L^{*}87±42 μm87 \pm42\,\mu\mathrm{m} (18 w/w% DSCG); 166±93 μm166 \pm93\,\mu\mathrm{m} (15 w/w% DSCG)
cell height from surfacehh1.0±0.4 μm1.0 \pm0.4\,\mu\mathrm{m} (18 w/w% DSCG); 1.0±0.7 μm1.0 \pm0.7\,\mu\mathrm{m} (15 w/w% DSCG)

Table S1. Summary of measurements and parameters used in theoretical calculations.

Estimation of chain distance from a surface

We observe that buckling occurs in chains that have at least one endpoint transiently pinned, suggesting that bacterial chains come into contact with the surfaces of the imaging device (the chain’s endpoints may come into contact with the glass bottom surface [102] or with the top PDMS surface [103]). To incorporate these surface effects into the theoretical calculations, we assume that the filament is growing a distance, hh, from a surface (Figure S2). To estimate hh experimentally, we track the zz-coordinate of a chain’s endpoints throughout the experiment. Next, we calculate the standard deviation of the zz-coordinate time series. The standard deviation yields an estimate of the chain’s fluctuations near the surface of the imaging device, providing an estimate of a chain’s average distance from the surface. We repeated this procedure for multiple buckling chains. The averages of these measurements are shown in Table S1 for different DSCG concentrations.

Schematic depicting the geometry to estimate the distance between a bacterial chain and a surface

Figure S2. Schematic depicting the geometry to estimate the distance between a bacterial chain and a surface. The schematic shows a bacterial chain (green spherocylinders) within the custom PDMS device. The bacterial chain is a distance hh from the glass bottom surface. To estimate hh we track the endpoints of the bacterial chain (indicated here as red and blue dots). Next, we take the standard deviation of the zz-position time series to get an estimate of the chain’s fluctuations near the surface of the imaging device. The average of the fluctuations yields an estimate of hh for a single chain. We repeat this process for multiple buckling chains to obtain the average values shown in Table S1.

Mathematical modeling

Slender filament growth and kinematics

The bacterial chain will be treated as a continuum filament of length L(t)L(t), which starts at t=0t = 0 in a straight configuration with centerline X0(S)\mathbf{X}_0(S) parameterized by the arclength S∈[−L(0)/2,L(0)/2]S \in[-L(0)/2,L(0)/2] (the reference frame). At time tt, the centerline position is written as x0(s(S,t),t)\mathbf{x}_0(s(S,t),t), where s∈[−L(t)/2,L(t)/2]s \in[-L(t)/2,L(t)/2] is the current arclength parameter and ∣∂sx0∣=1\lvert\partial_s\mathbf{x}_0\rvert= 1.

Assuming a uniform growth law for each unit of bacterial length, we have ∂s/∂S=α\partial s/\partial S = \alpha and s˙≡∂ts(S,t)=αs(S,t)\dot{s} \equiv\partial_t s(S,t) = \alpha s(S,t). Hence, with s(S,0)=Ss(S,0) = S, we have naturally that s(S,t)=Sexp⁡(αt)s(S,t) = S\exp(\alpha t).

The material velocity, denoted by V(S,t)\mathbf{V}(S,t), of the filament centerline is given by the time rate of change of the current position for fixed material coordinate SS:

V(S,t)=dx0(s(S,t),t)dt∣S=∂tx0∣S+s˙ ∂sx0=:v(s(S,t),t)+αs(S,t)s^(s(S,t),t),(8)\mathbf{V}(S,t) = \left.\frac{\mathrm{d}\mathbf{x}_0(s(S,t),t)}{\mathrm{d}t}\right|_S = \left.\partial_t\mathbf{x}_0\right|_S + \dot{s}\,\partial_s\mathbf{x}_0 =: \mathbf{v}(s(S,t),t) + \alpha s(S,t)\hat{\mathbf{s}}(s(S,t),t), \tag*{(8)}

where s^≡∂sx0(s(S,t),t)\hat{\mathbf{s}} \equiv\partial_s\mathbf{x}_0(s(S,t),t) is the unit tangent vector along the centerline. We recognize the final two terms as the rigid body motion component and the growth component, respectively. We further decompose the rigid body motion into its tangential and normal parts. At time tt,

v(s,t)=v∥(s,t)s^(s,t)+v⊥(s,t),(9)\mathbf{v}(s,t) = v_{\parallel}(s,t)\hat{\mathbf{s}}(s,t) + \mathbf{v}_{\perp}(s,t), \tag*{(9)}

where s^⋅v⊥=0\hat{\mathbf{s}}\cdot\mathbf{v}_{\perp} = 0. Continuity demands that ∣∂sx0∣=1\lvert\partial_s\mathbf{x}_0\rvert= 1 at all times, which implies that

0=12ddt∣∂sx0∣2=∂sx0⋅∂stx0=s^⋅∂s(∂tx0)=s^⋅∂sv.(10)0 = \frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}\lvert\partial_s\mathbf{x}_0\rvert^2 = \partial_s\mathbf{x}_0\cdot\partial_{st}\mathbf{x}_0 = \hat{\mathbf{s}}\cdot\partial_s(\partial_t\mathbf{x}_0) = \hat{\mathbf{s}}\cdot\partial_s\mathbf{v}. \tag*{(10)}

As expected, the rigid body component of the filament velocity must be constant in the tangential direction, v∥(s,t)=v∥(t)v_{\parallel}(s,t) = v_{\parallel}(t), or else the filament would rupture.

Filament geometry

For a small planar displacement of the filament, uu, we write the centerline position as

x0(s,t)=sx^+u(s,t)y^+O(Lus2),(11)\boldsymbol{x}_0(s,t)=s\hat{\boldsymbol{x}}+u(s,t)\hat{\boldsymbol{y}}+O(Lu_s^2), \tag*{(11)}

where {x^,y^,z^}\{\hat{\boldsymbol{x}},\hat{\boldsymbol{y}},\hat{\boldsymbol{z}}\} denote the unit Cartesian vectors, with x^\hat{\boldsymbol{x}} pointing along the length of the undeformed filament. To study the initial instability we will assume that ∣us∣≪1|u_s|\ll1. The centerline unit tangent vector, s^=∂sx0\hat{\boldsymbol{s}}=\partial_s\boldsymbol{x}_0, is accompanied by normal and binormal unit vectors along the centerline, {n^,z^}\{\hat{\boldsymbol{n}},\hat{\boldsymbol{z}}\}, where

s^=x^+usy^+O(us2),(12)\hat{\boldsymbol{s}}=\hat{\boldsymbol{x}}+u_s\hat{\boldsymbol{y}}+O(u_s^2), \tag*{(12)}
n^=−usx^+y^+O(us2),(13)\hat{\boldsymbol{n}}=-u_s\hat{\boldsymbol{x}}+\hat{\boldsymbol{y}}+O(u_s^2), \tag*{(13)}

forming an orthonormal triad.

Each cross-section of the filament is assumed to be circular with diameter 2a2a. The filament surface, denoted by ∂D\partial D (neglecting the filament ends), is then parameterized by (s,η)∈[−L(t)/2,L(t)/2]×[0,2π)(s,\eta)\in[-L(t)/2,L(t)/2]\times[0,2\pi), as

x(s,η)=x0(s)+ae^⊥(s,η),(14)\boldsymbol{x}(s,\eta)=\boldsymbol{x}_0(s)+a\hat{\boldsymbol{e}}_\perp(s,\eta), \tag*{(14)}

where e^⊥(s,η)=cos⁡ηn^(s)+sin⁡ηz^\hat{\boldsymbol{e}}_\perp(s,\eta)=\cos\eta\hat{\boldsymbol{n}}(s)+\sin\eta\hat{\boldsymbol{z}}. At each point on the surface, x(s,η)∈∂D\boldsymbol{x}(s,\eta)\in\partial D, we have another orthonormal triad {ξ^,ζ^,ν^}\{\hat{\boldsymbol{\xi}},\hat{\boldsymbol{\zeta}},\hat{\boldsymbol{\nu}}\}, composed of two unit tangent vectors on the surface, ξ^=xs/∣xs∣\hat{\boldsymbol{\xi}}=\boldsymbol{x}_s/|\boldsymbol{x}_s| and ζ^=xη/∣xη∣\hat{\boldsymbol{\zeta}}=\boldsymbol{x}_\eta/|\boldsymbol{x}_\eta|, and the unit normal vector, which points into the LC, ν^=ζ^×ξ^/∣ζ^×ξ^∣\hat{\boldsymbol{\nu}}=\hat{\boldsymbol{\zeta}}\times\hat{\boldsymbol{\xi}}/|\hat{\boldsymbol{\zeta}}\times\hat{\boldsymbol{\xi}}|. With the centerline displacement assumed to be small, these are given by

x(s,η)=sx^+ar^+O(Lus),(15)\boldsymbol{x}(s,\eta)=s\hat{\boldsymbol{x}}+a\hat{\boldsymbol{r}}+O(Lu_s), \tag*{(15)}
ζ^(s,η)=−sin⁡ηn^(s)+cos⁡ηz^+O(us2),(16)\hat{\boldsymbol{\zeta}}(s,\eta)=-\sin\eta\hat{\boldsymbol{n}}(s)+\cos\eta\hat{\boldsymbol{z}}+O(u_s^2), \tag*{(16)}
ξ^(s,η)=s^(s)+O(us2),(17)\hat{\boldsymbol{\xi}}(s,\eta)=\hat{\boldsymbol{s}}(s)+O(u_s^2), \tag*{(17)}
ν^(s,η)=cos⁡ηn^(s)+sin⁡ηz^+O(us2).(18)\hat{\boldsymbol{\nu}}(s,\eta)=\cos\eta\hat{\boldsymbol{n}}(s)+\sin\eta\hat{\boldsymbol{z}}+O(u_s^2). \tag*{(18)}

where r^(η)=cos⁡ηy^+sin⁡ηz^\hat{\boldsymbol{r}}(\eta)=\cos\eta\hat{\boldsymbol{y}}+\sin\eta\hat{\boldsymbol{z}} and η^(η)=−sin⁡ηy^+cos⁡ηz^\hat{\boldsymbol{\eta}}(\eta)=-\sin\eta\hat{\boldsymbol{y}}+\cos\eta\hat{\boldsymbol{z}} are the radial and azimuthal coordinate vectors, respectively. Later we will use the fact that ζ^=η^+O(us)\hat{\boldsymbol{\zeta}}=\hat{\boldsymbol{\eta}}+O(u_s), ν^=r^+O(us)\hat{\boldsymbol{\nu}}=\hat{\boldsymbol{r}}+O(u_s), and s^=x^+O(us)\hat{\boldsymbol{s}}=\hat{\boldsymbol{x}}+O(u_s), i.e. recovering cylindrical coordinates at leading-order.

Force balance and filament dynamics

Force balance is considered at each segment along the filament. At a station ss, in the limit of large viscous damping relative to the inertial effects (i.e., Stokes flow), the resultant force (per unit length) acting on the filament is zero at all times. We decompose this force as

f(s,t)=fviscous(s,t)+∂sF(s,t)+fLC(s,t)=0,(19)\boldsymbol{f}(s,t)=\boldsymbol{f}^{\mathrm{viscous}}(s,t)+\partial_s\boldsymbol{F}(s,t)+\boldsymbol{f}^{\mathrm{LC}}(s,t)=0, \tag*{(19)}

where fviscous\boldsymbol{f}^{\mathrm{viscous}} is the viscous drag, ∂sF\partial_s\boldsymbol{F} is the differential elastic force, and fLC\boldsymbol{f}^{\mathrm{LC}} is the LC restoring force. We shall compute each of these terms in turn.

First is the viscous drag, which we model using a resistive force approximation,

fviscous=−R⋅V,(20)\boldsymbol{f}^{\mathrm{viscous}}=-\mathcal{R}\cdot\boldsymbol{V}, \tag*{(20)}

where R\mathcal{R} is a resistance tensor,

R=ζ∥s^s^+ζ⊥(I−s^s^).(21)\mathcal{R}=\zeta_{\parallel}\hat{\boldsymbol{s}}\hat{\boldsymbol{s}}+\zeta_{\perp}(\boldsymbol{I}-\hat{\boldsymbol{s}}\hat{\boldsymbol{s}}). \tag*{(21)}

The drag coefficients ζ∥\zeta_{\parallel} and ζ⊥\zeta_{\perp} penalize motion in the filament tangent and normal directions, respectively. In terms of the rigid body motion and growth components of the filament velocity, we may write

fviscous=−[ζ∥s^s^+ζ⊥(I−s^s^)]⋅(v+αss^)=−ζ∥(v∥+αs)s^−ζ⊥v⊥.(22)\boldsymbol{f}^{\mathrm{viscous}}=-\left[\zeta_{\parallel}\hat{\boldsymbol{s}}\hat{\boldsymbol{s}}+\zeta_{\perp}(\boldsymbol{I}-\hat{\boldsymbol{s}}\hat{\boldsymbol{s}})\right]\cdot(\boldsymbol{v}+\alpha s\hat{\boldsymbol{s}})=-\zeta_{\parallel}(v_{\parallel}+\alpha s)\hat{\boldsymbol{s}}-\zeta_{\perp}\boldsymbol{v}_{\perp}. \tag*{(22)}

For reference, in a Newtonian fluid with viscosity μ\mu, we would have ζ∥≈2πμ/log⁡(L/2a)\zeta_{\parallel} \approx2\pi\mu/\log(L/2a) and ζ⊥≈2ζ∥\zeta_{\perp} \approx2\zeta_{\parallel} at leading-order in 1/log⁡(L/2a)1/\log(L/2a), assumed small; or, we instead have ζ∥≈2πμ/log⁡(2h/a)\zeta_{\parallel} \approx2\pi\mu/\log(2h/a) and ζ⊥≈2ζ∥\zeta_{\perp} \approx2\zeta_{\parallel} if the filament is moving a distance h≪Lh \ll L away from a wall [80].

Next is the differential elastic force due to the connectivity of the filament itself. We assume that the filament cannot support a bending moment; hence, from the angular momentum balance, we can write the force density as a tangential piece alone, F(s,t)=T(s,t)s^\boldsymbol{F}(s,t) = T(s,t)\hat{\boldsymbol{s}}. With this, we can decompose the force balance, Eq. (19), into its tangential and normal components,

−ζ∥(v∥+αs)+∂sT=0,(23)-\zeta_{\parallel}(v_{\parallel}+\alpha s)+\partial_s T=0, \tag*{(23)}
−ζ⊥v⊥+(I−s^s^)⋅(fLC+T∂sss^)=0,(24)-\zeta_{\perp}v_{\perp}+(\boldsymbol{I}-\hat{\boldsymbol{s}}\hat{\boldsymbol{s}})\cdot(\boldsymbol{f}^{\mathrm{LC}}+T\partial_{ss}\hat{\boldsymbol{s}})=0, \tag*{(24)}

respectively, where we have used that s^⋅fLC=0\hat{\boldsymbol{s}}\cdot\boldsymbol{f}^{\mathrm{LC}}=0 (see §S5 G). If both ends of the filament are free, the end tensions must equate, T(±L/2)=TL=O(w2)T(\pm L/2)=T_L=O(w^2), with w=aW/Kw=aW/K the dimensionless anchoring strength. Integrating the tangential force balance, Eq. (13a), thus yields v∥=0v_{\parallel}=0, as well as an expression for the tension,

T(s,t)=TL−ζ∥α2(L(t)24−s2).(25)T(s,t)=T_L-\frac{\zeta_{\parallel}\alpha}{2}\left(\frac{L(t)^2}{4}-s^2\right). \tag*{(25)}

This baseline tension can be negative (compression), which has the potential to buckle the filament. This compression can be even more pronounced if one end of the filament meets additional resistance, due to surface interactions, for instance.

Consider the extreme case when one endpoint (e.g., the one at s=−L/2s=-L/2) is pinned (hinged), which requires an additional tangential force F\boldsymbol{F} there. With V(−L/2,t)=[v∥(t)−αL(t)/2]x^+v⊥(−L/2,t)=0\boldsymbol{V}(-L/2,t)=[v_{\parallel}(t)-\alpha L(t)/2]\hat{\boldsymbol{x}}+v_{\perp}(-L/2,t)=0, we find that v∥(t)=αL/2v_{\parallel}(t)=\alpha L/2 and v⊥(−L/2,t)=0v_{\perp}(-L/2,t)=0. Integrating (13a), we find that −ζ∥αL2/2+TL−T−L=0-\zeta_{\parallel}\alpha L^2/2+T_L-T_{-L}=0 for T±L≡T(±L/2)T_{\pm L}\equiv T(\pm L/2). Using the endpoint force balance, F+T−L=0\boldsymbol{F}+T_{-L}=0, we find that the required tangential force is T−L=−F=TL−ζ∥αL2/4T_{-L}=-F=T_L-\zeta_{\parallel}\alpha L^2/4. In this pinned problem the baseline tension is given instead by

T(s,t)=TL−αζ∥2(L(t)2−s)(3L(t)2+s).(26)T(s,t)=T_L-\frac{\alpha\zeta_{\parallel}}{2}\left(\frac{L(t)}{2}-s\right)\left(\frac{3L(t)}{2}+s\right). \tag*{(26)}

For the free filament, from (14), the tension is smallest (and becomes negative, as a compression) at the midpoint, s=0s=0. For the pinned filament, from (15) the tension is smallest at the point of pinning, s=−L/2s=-L/2, and it can be up to 4 times larger in the absence of TLT_L (i.e. −αζL2/8-\alpha\zeta L^2/8 vs. −αζL2/2-\alpha\zeta L^2/2).

In either case, using v⊥=uty^[1+O(us)]\boldsymbol{v}_{\perp}=u_t\hat{\boldsymbol{y}}[1+O(u_s)], Eq. (13b) may be written as

ζ⊥ut=T(s)uss+y^⋅fLC(s),(27)\zeta_{\perp}u_t=T(s)u_{ss}+\hat{\boldsymbol{y}}\cdot\boldsymbol{f}^{\mathrm{LC}}(s), \tag*{(27)}

where T(s)T(s) is given in Eq. (25), for free-free ends, or Eq. (26), for fixed-free ends. In the free-free case, the boundary conditions are selected to match the LC director field far away (assumed to be the x^\hat{\boldsymbol{x}}-direction), i.e., us(±L/2)=0u_s(\pm L/2)=0. In the fixed-free case, the boundary conditions instead take the form u(−L/2)=us(L/2)=0u(-L/2)=u_s(L/2)=0.

LC director field and traction

Before we can determine the traction on the filament due to the surrounding LC environment, we need to determine the LC configuration. Fortunately, the system is evolving slowly, so we can determine the LC field in a quasi-static manner, i.e., solving an equilibrium problem for each filament configuration. Here, we let L=L(t)L=L(t) and drop all arguments of time since the LC director field is determined instantaneously at any given time tt.

Far from the bacterial chain, the LC director field is assumed to be uniform in the x^\hat{\boldsymbol{x}} direction. Elsewhere, at a position x\boldsymbol{x} in the fluid, the LC director field is represented as n(x)=x^+p(x)+O(us2)\boldsymbol{n}(\boldsymbol{x})=\hat{\boldsymbol{x}}+\boldsymbol{p}(\boldsymbol{x})+O(u_s^2) with p⋅x^=0\boldsymbol{p}\cdot\hat{\boldsymbol{x}}=0, where p=O(us)\boldsymbol{p}=O(u_s) in order to have a balance between the deflection and the director field. With this approximation, the Ericksen torque stress tensor and projected molecular field are given by

Π=K∇n=K∇p[1+O(us)],(28)\boldsymbol{\Pi}=K\nabla\boldsymbol{n}=K\nabla\boldsymbol{p}[1+O(u_s)], \tag*{(28)}
h=P(n)⋅(∇⋅Π)=K∇2p[1+O(us)].(29)\boldsymbol{h}=\boldsymbol{P}(\boldsymbol{n})\cdot(\nabla\cdot\boldsymbol{\Pi})=K\nabla^2\boldsymbol{p}[1+O(u_s)]. \tag*{(29)}

respectively, for the single (Frank) elastic constant KK and projection operator P(n)≡I−nn\mathbf{P}(\mathbf{n}) \equiv\mathbf{I}-\mathbf{n}\mathbf{n}, see Ref. [79]. At equilibrium, h=0\mathbf{h}=0 and so ∇2p=0\nabla^2\mathbf{p}=0 with p→0\mathbf{p}\to0 as ∣x∣→∞|\mathbf{x}|\to\infty. On the boundary, x∈∂D\mathbf{x}\in\partial D, we have the projected surface molecular field

hsurf=P(n)⋅[ν^⋅Π−∂Fsurf∂n]=[K∂νp−P(n)⋅∂Fsurf∂n][1+O(us)],(30)\mathbf{h}_{\mathrm{surf}}=\mathbf{P}(\mathbf{n})\cdot\left[\hat{\boldsymbol{\nu}}\cdot\boldsymbol{\Pi}-\frac{\partial F_{\mathrm{surf}}}{\partial\mathbf{n}}\right]=\left[K\partial_{\nu}\mathbf{p}-\mathbf{P}(\mathbf{n})\cdot\frac{\partial F_{\mathrm{surf}}}{\partial\mathbf{n}}\right]\left[1+O(u_s)\right], \tag*{(30)}

where FsurfF_{\mathrm{surf}} is a surface energy that describes how the director field is anchored to the bacteria (to be defined), see Ref. [104]. At equilibrium, hsurf(x)=0\mathbf{h}_{\mathrm{surf}}(\mathbf{x})=0 for x∈∂D\mathbf{x}\in\partial D. Note from Eq. (8a) that hsurf(x)=hsurf(sx^+ar^)+O(u∇hs)\mathbf{h}_{\mathrm{surf}}(\mathbf{x})=\mathbf{h}_{\mathrm{surf}}(s\hat{\mathbf{x}}+a\hat{\mathbf{r}})+O(u\nabla h_s) for x∈∂D\mathbf{x}\in\partial D, so ∂D\partial D is effectively a cylinder of radius aa aligned with x^\hat{\mathbf{x}}. We thus introduce a cylindrical polar coordinate system (s,r,η)∈[−L/2,L/2]×[a,∞)×[0,2π)(s,r,\eta)\in[-L/2,L/2]\times[a,\infty)\times[0,2\pi).

Surface anchoring and director field

On the filament surface, the LC is assumed to satisfy weak (i.e. finite strength) degenerate tangential anchoring conditions, with surface energy

Fsurf=W2(n⋅ν^)2,(31)F_{\mathrm{surf}}=\frac{W}{2}(\mathbf{n}\cdot\hat{\boldsymbol{\nu}})^2, \tag*{(31)}

for the unit normal vector ν^=r^−uscos⁡ηx^+O(us2)\hat{\boldsymbol{\nu}}=\hat{\mathbf{r}}-u_s\cos\eta\hat{\mathbf{x}}+O(u_s^2) and anchoring strength WW, see Ref. [105]. Since n⋅ν^=p⋅r^−uscos⁡η+O(us2)\mathbf{n}\cdot\hat{\boldsymbol{\nu}}=\mathbf{p}\cdot\hat{\mathbf{r}}-u_s\cos\eta+O(u_s^2), the surface molecular field is

hsurf=[K∂rp−W(p⋅r^−uscos⁡η)r^][1+O(us)].(32)\mathbf{h}_{\mathrm{surf}}=\left[K\partial_r\mathbf{p}-W(\mathbf{p}\cdot\hat{\mathbf{r}}-u_s\cos\eta)\hat{\mathbf{r}}\right]\left[1+O(u_s)\right]. \tag*{(32)}

Furthermore, since p⋅x^=0\mathbf{p}\cdot\hat{\mathbf{x}}=0, we can write p=ϕ1(x)y^+ϕ2(x)z^\mathbf{p}=\phi_1(\mathbf{x})\hat{\mathbf{y}}+\phi_2(\mathbf{x})\hat{\mathbf{z}}, for two scalar functions ϕ1=O(us)\phi_1=O(u_s) and ϕ2=O(us)\phi_2=O(u_s), which satisfy ∇2ϕ1=∇2ϕ2=0\nabla^2\phi_1=\nabla^2\phi_2=0 subject to ϕ1(x),ϕ2(x)→0\phi_1(\mathbf{x}),\phi_2(\mathbf{x})\to0 as ∣x∣→∞|\mathbf{x}|\to\infty and

K∂rϕ1=W[(ϕ1−us)cos⁡η+ϕ2sin⁡η]cos⁡η,(33)K\partial_r\phi_1=W[(\phi_1-u_s)\cos\eta+\phi_2\sin\eta]\cos\eta, \tag*{(33)}
K∂rϕ2=W[(ϕ1−us)cos⁡η+ϕ2sin⁡η]sin⁡η,(34)K\partial_r\phi_2=W[(\phi_1-u_s)\cos\eta+\phi_2\sin\eta]\sin\eta, \tag*{(34)}

on r=ar=a.

a. Free-free case: In the case of two free-filament ends, with us(±L/2)=0u_s(\pm L/2)=0, we can write uu as a cosine series:

u(s,t)=∑k=1∞u^k(t)cos⁡(qk(2s+L)L),(35)u(s,t)=\sum_{k=1}^{\infty}\hat{u}_k(t)\cos\left(\frac{q_k(2s+L)}{L}\right), \tag*{(35)}

where qk=kπ/2q_k=k\pi/2. Using separation of variables, we pose general series representations for ϕ1\phi_1 and ϕ2\phi_2 of the form:

ϕ1(x)=∑k=1∞−2qku^kLsin⁡(qk(2s+L)L)[∑m=−∞∞Km(2qkr/L)Km(2qka/L)akmeimη],(36)\phi_1(\mathbf{x})=\sum_{k=1}^{\infty}\frac{-2q_k\hat{u}_k}{L}\sin\left(\frac{q_k(2s+L)}{L}\right)\left[\sum_{m=-\infty}^{\infty}\frac{K_m(2q_kr/L)}{K_m(2q_ka/L)}a_k^m e^{im\eta}\right], \tag*{(36)}
ϕ2(x)=∑k=1∞−2qku^kLsin⁡(qk(2s+L)L)[∑m=−∞∞Km(2qkr/L)Km(2qka/L)bkmeimη],(37)\phi_2(\mathbf{x})=\sum_{k=1}^{\infty}\frac{-2q_k\hat{u}_k}{L}\sin\left(\frac{q_k(2s+L)}{L}\right)\left[\sum_{m=-\infty}^{\infty}\frac{K_m(2q_kr/L)}{K_m(2q_ka/L)}b_k^m e^{im\eta}\right], \tag*{(37)}

where Km(⋅)K_m(\cdot) is the mmth modified Bessel function of the second kind and akm=ak−m‾a_k^m=\overline{a_k^{-m}} and bkm=bk−m‾b_k^m=\overline{b_k^{-m}} are complex constants, which are to be determined now.

The boundary conditions on r=ar=a impose the conditions

∑m=−∞∞eimη[Skmwakm+(akmcos⁡η+bkmsin⁡η)cos⁡η]=cos⁡2η,(38)\sum_{m=-\infty}^{\infty}e^{im\eta}\left[\frac{S_k^m}{w}a_k^m+(a_k^m\cos\eta+b_k^m\sin\eta)\cos\eta\right]=\cos^2\eta, \tag*{(38)}
∑m=−∞∞eimη[Skmwbkm+(akmcos⁡η+bkmsin⁡η)sin⁡η]=cos⁡ηsin⁡η,(39)\sum_{m=-\infty}^{\infty}e^{im\eta}\left[\frac{S_k^m}{w}b_k^m+(a_k^m\cos\eta+b_k^m\sin\eta)\sin\eta\right]=\cos\eta\sin\eta, \tag*{(39)}

where w=aW/Kw=aW/K is the dimensionless anchoring strength, Skm=Sm(2qka/L)S_k^m=S_m(2q_ka/L), and

Sm(x)=−xKm′(x)Km(x).(40)S_m(x)=-\frac{xK_m'(x)}{K_m(x)}. \tag*{(40)}

Replacing the trigonometric functions with exponentials and evaluating coefficients yields the conditions

2akm[1+2wSkm]+akm−2+akm+2−ibkm−2+ibkm+2=2δ0m+δ2m+δ−2m,(41)2a_k^m\left[1+\frac{2}{w}S_k^m\right]+a_k^{m-2}+a_k^{m+2}-ib_k^{m-2}+ib_k^{m+2}=2\delta_{0m}+\delta_{2m}+\delta_{-2m}, \tag*{(41)}
2ibkm[1+2wSkm]+akm−2−akm+2−ibkm−2−ibkm+2=δ2m−δ−2m.(42)2ib_k^m\left[1+\frac{2}{w}S_k^m\right]+a_k^{m-2}-a_k^{m+2}-ib_k^{m-2}-ib_k^{m+2}=\delta_{2m}-\delta_{-2m}. \tag*{(42)}

where δij\delta_{ij} is the Kronecker delta function. Solving these recurrence relations yields akm=bkm=0a_k^m=b_k^m=0 for m∉{−2,0,2}m\notin\{-2,0,2\}, bk0=0b_k^0=0, ibk2=ibk−2=ak0=ak2ib_k^2=ib_k^{-2}=a_k^0=a_k^2,

ak0=Sk2Sk0+Sk2+2wSk0Sk2,andak2=Sk0/2Sk0+Sk2+2wSk0Sk2.(43)a_k^0=\frac{S_k^2}{S_k^0+S_k^2+\frac{2}{w}S_k^0S_k^2},\qquad\text{and}\qquad a_k^2=\frac{S_k^0/2}{S_k^0+S_k^2+\frac{2}{w}S_k^0S_k^2}. \tag*{(43)}

Altogether, we have the solution

ϕ1(x)=∑k=1∞−2qku^kLsin⁡(qk(2s+L)L)[AkK0(2qkr/L)K0(2qka/L)+Bkcos⁡(2η)K2(2qkr/L)K2(2qka/L)],(44)\phi_1(\mathbf{x})=\sum_{k=1}^{\infty}\frac{-2q_k\hat{u}_k}{L}\sin\left(\frac{q_k(2s+L)}{L}\right)\left[A_k\frac{K_0(2q_kr/L)}{K_0(2q_ka/L)}+B_k\cos(2\eta)\frac{K_2(2q_kr/L)}{K_2(2q_ka/L)}\right], \tag*{(44)}
ϕ2(x)=∑k=1∞−2qku^kLsin⁡(qk(2s+L)L)[Bksin⁡(2η)K2(2qkr/L)K2(2qka/L)],(45)\phi_2(\mathbf{x})=\sum_{k=1}^{\infty}\frac{-2q_k\hat{u}_k}{L}\sin\left(\frac{q_k(2s+L)}{L}\right)\left[B_k\sin(2\eta)\frac{K_2(2q_kr/L)}{K_2(2q_ka/L)}\right], \tag*{(45)}

where

Ak≡1/S0(2qka/L)1/S2(2qka/L)+1/S0(2qka/L)+2/wandBk≡1/S2(2qka/L)1/S2(2qka/L)+1/S0(2qka/L)+2/w.(46)A_k\equiv\frac{1/S_0(2q_ka/L)}{1/S_2(2q_ka/L)+1/S_0(2q_ka/L)+2/w}\qquad\text{and}\qquad B_k\equiv\frac{1/S_2(2q_ka/L)}{1/S_2(2q_ka/L)+1/S_0(2q_ka/L)+2/w}. \tag*{(46)}

b. One free-end: In the case of one fixed and one free-filament ends, with u(−L/2)=0u(-L/2)=0 and us(L/2)=0u_s(L/2)=0, respectively, we can instead write uu as the series

u(s,t)=∑k=1∞u^k(t)sin⁡(Qk(2s+L)L),(47)u(s,t)=\sum_{k=1}^{\infty}\hat{u}_k(t)\sin\left(\frac{Q_k(2s+L)}{L}\right), \tag*{(47)}

for Qk=(2k−1)π/4Q_k=(2k-1)\pi/4. In this case, we find

ϕ1(x)=∑k=1∞2Qku^kLcos⁡(Qk(2s+L)L)[CkK0(2Qkr/L)K0(2Qka/L)+Dkcos⁡(2η)K2(2Qkr/L)K2(2Qka/L)],(48)\phi_1(\mathbf{x})=\sum_{k=1}^{\infty}\frac{2Q_k\hat{u}_k}{L}\cos\left(\frac{Q_k(2s+L)}{L}\right)\left[C_k\frac{K_0(2Q_kr/L)}{K_0(2Q_ka/L)}+D_k\cos(2\eta)\frac{K_2(2Q_kr/L)}{K_2(2Q_ka/L)}\right], \tag*{(48)}
ϕ2(x)=∑k=1∞2Qku^kLcos⁡(Qk(2s+L)L)[Dksin⁡(2η)K2(2Qkr/L)K2(2Qka/L)],(49)\phi_2(\mathbf{x})=\sum_{k=1}^{\infty}\frac{2Q_k\hat{u}_k}{L}\cos\left(\frac{Q_k(2s+L)}{L}\right)\left[D_k\sin(2\eta)\frac{K_2(2Q_kr/L)}{K_2(2Q_ka/L)}\right], \tag*{(49)}

where

Ck≡1/S0(2Qka/L)1/S2(2Qka/L)+1/S0(2Qka/L)+2/wandDk≡1/S2(2Qka/L)1/S2(2Qka/L)+1/S0(2Qka/L)+2/w.(50)C_k\equiv\frac{1/S_0(2Q_ka/L)}{1/S_2(2Q_ka/L)+1/S_0(2Q_ka/L)+2/w}\qquad\text{and}\qquad D_k\equiv\frac{1/S_2(2Q_ka/L)}{1/S_2(2Q_ka/L)+1/S_0(2Q_ka/L)+2/w}. \tag*{(50)}
Surface traction

To compute the traction on the filament due to the LC, we will need the Ericksen stress tensor and the surface stress tensor, given by

T=K2∥∇n∥2I−K∇n⋅∇nT,(51)\mathbf{T}=\frac{K}{2}\lVert\nabla\mathbf{n}\rVert^2\mathbf{I}-K\nabla\mathbf{n}\cdot\nabla\mathbf{n}^{T}, \tag*{(51)}
Tsurf=FsurfP(ν^)−P(ν^)⋅∂Fsurf∂ν^ν^,(52)\mathbf{T}_{\mathrm{surf}}=\mathcal{F}_{\mathrm{surf}}\mathbf{P}(\hat{\boldsymbol{\nu}})-\mathbf{P}(\hat{\boldsymbol{\nu}})\cdot\frac{\partial\mathcal{F}_{\mathrm{surf}}}{\partial\hat{\boldsymbol{\nu}}}\hat{\boldsymbol{\nu}}, \tag*{(52)}

respectively, see [79, 104].

The first part of the traction is associated with the Ericksen stress T\mathbf{T}, and is given by f(s,η)=ν^⋅Tf(s,\eta)=\hat{\boldsymbol{\nu}}\cdot\mathbf{T}. But this traction enters at O(p2)=O(us2)O(p^2)=O(u_s^2), and so we neglect it. The second part of the traction is associated with the surface stress Tsurf\mathbf{T}_{\mathrm{surf}}, and is given by fsurf(s,η)=∇s⋅Tsurf\mathbf{f}_{\mathrm{surf}}(s,\eta)=\nabla_s\cdot\mathbf{T}_{\mathrm{surf}}, where ∇s≡P(ν^)⋅∇\nabla_s\equiv\mathbf{P}(\hat{\boldsymbol{\nu}})\cdot\nabla with P(ν^)≡I−ν^ν^\mathbf{P}(\hat{\boldsymbol{\nu}})\equiv\mathbf{I}-\hat{\boldsymbol{\nu}}\hat{\boldsymbol{\nu}}, a projection operator for the surface normal direction. Since ν^=r^+O(us)\hat{\boldsymbol{\nu}}=\hat{\mathbf{r}}+\mathcal{O}(u_s), we note that

∇s=P(ν^)⋅∇=(x^∂s+1rη^∂η)[1+O(us)].(53)\nabla_s=\mathbf{P}(\hat{\boldsymbol{\nu}})\cdot\nabla=\left(\hat{\mathbf{x}}\partial_s+\frac{1}{r}\hat{\boldsymbol{\eta}}\partial_\eta\right)\left[1+\mathcal{O}(u_s)\right]. \tag*{(53)}

For the weak degenerate tangential anchoring conditions considered above, we have the surface stress

Tsurf=W2(n⋅ν^)2P(ν^)−W(n⋅ν^)P(ν^)⋅nν^,(54)\mathbf{T}_{\mathrm{surf}}=\frac{W}{2}(\mathbf{n}\cdot\hat{\boldsymbol{\nu}})^2\mathbf{P}(\hat{\boldsymbol{\nu}})-W(\mathbf{n}\cdot\hat{\boldsymbol{\nu}})\mathbf{P}(\hat{\boldsymbol{\nu}})\cdot\mathbf{n}\hat{\boldsymbol{\nu}}, \tag*{(54)}

for ν^=r^−uscos⁡ηx^+O(us2)\hat{\boldsymbol{\nu}}=\hat{\mathbf{r}}-u_s\cos\eta\hat{\mathbf{x}}+\mathcal{O}(u_s^2) and n=x^+ϕ1y^+ϕ2z^+O(us2)\mathbf{n}=\hat{\mathbf{x}}+\phi_1\hat{\mathbf{y}}+\phi_2\hat{\mathbf{z}}+\mathcal{O}(u_s^2). It follows that n⋅ν^=(ϕ1−us)cos⁡η+ϕ2sin⁡η+O(us2)\mathbf{n}\cdot\hat{\boldsymbol{\nu}}=(\phi_1-u_s)\cos\eta+\phi_2\sin\eta+\mathcal{O}(u_s^2), and so,

Tsurf=−W[(ϕ1−us)cos⁡η+ϕ2sin⁡η]x^r^[1+O(us)]=−K∂r[ϕ1x^y^+ϕ2x^z^][1+O(us)],(55)\mathbf{T}_{\mathrm{surf}}=-W[(\phi_1-u_s)\cos\eta+\phi_2\sin\eta]\hat{\mathbf{x}}\hat{\mathbf{r}}[1+\mathcal{O}(u_s)]=-K\partial_r[\phi_1\hat{\mathbf{x}}\hat{\mathbf{y}}+\phi_2\hat{\mathbf{x}}\hat{\mathbf{z}}][1+\mathcal{O}(u_s)], \tag*{(55)}

where we have used the boundary condition (20). Using (34) yields the surface traction:

fsurf(s,η)=∇s⋅Tsurf=−K[∂s∂rϕ1y^+∂s∂rϕ2z^][1+O(us)].(56)\mathbf{f}_{\mathrm{surf}}(s,\eta)=\nabla_s\cdot\mathbf{T}_{\mathrm{surf}}=-K[\partial_s\partial_r\phi_1\hat{\mathbf{y}}+\partial_s\partial_r\phi_2\hat{\mathbf{z}}][1+\mathcal{O}(u_s)]. \tag*{(56)}

Finally, the force (per length) acting on the filament at each station ss, i.e. fLC(s)\mathbf{f}^{\mathrm{LC}}(s), is found by integrating the total surface traction, f(s,η)+fsurf(s,η)\mathbf{f}(s,\eta)+\mathbf{f}_{\mathrm{surf}}(s,\eta), over the azimuthal angle, η\eta:

fLC(s)=∫02π[f(s,η)+fsurf(s,η)]a dη=−aK[(∫02π∂s∂rϕ1 dη)y^+(∫02π∂s∂rϕ2 dη)z^][1+O(us)].(57)\mathbf{f}^{\mathrm{LC}}(s)=\int_0^{2\pi}[\mathbf{f}(s,\eta)+\mathbf{f}_{\mathrm{surf}}(s,\eta)]a\,\mathrm{d}\eta=-aK\left[\left(\int_0^{2\pi}\partial_s\partial_r\phi_1\,\mathrm{d}\eta\right)\hat{\mathbf{y}}+\left(\int_0^{2\pi}\partial_s\partial_r\phi_2\,\mathrm{d}\eta\right)\hat{\mathbf{z}}\right][1+\mathcal{O}(u_s)]. \tag*{(57)}

Note that s^⋅fLC(s)=O(us2)\hat{\mathbf{s}}\cdot\mathbf{f}^{\mathrm{LC}}(s)=\mathcal{O}(u_s^2), as used earlier.

a. Two free ends: For two free ends, we have from (28),

∂s∂rϕ1=∑k=1∞(2qkL)2u^kacos⁡(qk(2s+L)L)[AkS0(2qkaL)+Bkcos⁡(2η)S2(2qkaL)],(58)\partial_s\partial_r\phi_1=\sum_{k=1}^{\infty}\left(\frac{2q_k}{L}\right)^2\frac{\hat{u}_k}{a}\cos\left(\frac{q_k(2s+L)}{L}\right)\left[A_kS_0\left(\frac{2q_ka}{L}\right)+B_k\cos(2\eta)S_2\left(\frac{2q_ka}{L}\right)\right], \tag*{(58)}
∂s∂rϕ2=∑k=1∞(2qkL)2u^kacos⁡(qk(2s+L)L)[Bksin⁡(2η)S2(2qkaL)],(59)\partial_s\partial_r\phi_2=\sum_{k=1}^{\infty}\left(\frac{2q_k}{L}\right)^2\frac{\hat{u}_k}{a}\cos\left(\frac{q_k(2s+L)}{L}\right)\left[B_k\sin(2\eta)S_2\left(\frac{2q_ka}{L}\right)\right], \tag*{(59)}

on r=ar=a, for qk=πk/2q_k=\pi k/2. Using that cos⁡(2η)\cos(2\eta) and sin⁡(2η)\sin(2\eta) have zero mean, and inserting the expressions in (29), the LC restoring force is found to be

fLC(s)∼−16πaKL3y^∑k=1∞qk3u^k[4qkawL−K0(2qka/L)K0′(2qka/L)−K2(2qka/L)K2′(2qka/L)]−1cos⁡(qk(2s+L)L).(60)\mathbf{f}^{\mathrm{LC}}(s)\sim-\frac{16\pi aK}{L^3}\hat{\mathbf{y}}\sum_{k=1}^{\infty}q_k^3\hat{u}_k\left[\frac{4q_ka}{wL}-\frac{K_0(2q_ka/L)}{K_0'(2q_ka/L)}-\frac{K_2(2q_ka/L)}{K_2'(2q_ka/L)}\right]^{-1}\cos\left(\frac{q_k(2s+L)}{L}\right). \tag*{(60)}

b. Fixed-free ends: For only one free-end, the restoring force is similarly found to be (40), but with the substitutions cos⁡↦sin⁡\cos\mapsto\sin and qk↦Qk=(2k−1)π/4q_k\mapsto Q_k=(2k-1)\pi/4.

Filament dynamics and buckling length estimates

The filament dynamics in Eq. (27) are now fully determined, and we can begin to analyze them. Let us at this stage non-dimensionalize the problem by defining a dimensionless time, τ≡αt\tau\equiv\alpha t, dimensionless length, ξ≡2s/L(t)\xi\equiv2s/L(t), where ξ∈[−1,1]\xi\in[-1,1], and dimensionless displacement U(ξ,τ)≡u(s,t)/aU(\xi,\tau)\equiv u(s,t)/a. We also introduce the dimensionless filament length Λ(τ)≡L(t)/(2a)\Lambda(\tau)\equiv L(t)/(2a), dimensionless viscosity χ≡ζ⊥/ζ∥\chi\equiv\zeta_\perp/\zeta_\parallel, dimensionless anchoring strength w≡aW/Kw\equiv aW/K, and Ericksen number Er≡a2ζ∥α/(2πK)\mathrm{Er}\equiv a^2\zeta_\parallel\alpha/(2\pi K).

From the uniform growth law, L′(t)=αL(t)L'(t)=\alpha L(t) (see §S5 D), we have that

∂tu(s,t)=aα[∂τU(ξ,τ)−ξ∂ξU(ξ,τ)].(61)\partial_tu(s,t)=a\alpha[\partial_\tau U(\xi,\tau)-\xi\partial_\xi U(\xi,\tau)]. \tag*{(61)}

and so the filament dynamics, i.e., Eq. (27), is governed by the dimensionless equation

χ(Uτ−ξUξ)=T~(ξ)Uξξ+Er−1FLC(ξ,τ),(62)\chi(U_\tau-\xi U_\xi)=\widetilde{T}(\xi)U_{\xi\xi}+\mathrm{Er}^{-1}F^{\mathrm{LC}}(\xi,\tau), \tag*{(62)}

for the dimensionless tension

T~(ξ)≡T(s,t)αζ∥L(t)2/4={−(1−ξ2)/2free-free,−(1−ξ)(3+ξ)/2fixed-free,(63)\widetilde{T}(\xi) \equiv\frac{T(s,t)}{\alpha\zeta_{\parallel} L(t)^2/4} = \begin{cases} -(1-\xi^2)/2 & \text{free-free},\\ -(1-\xi)(3+\xi)/2 & \text{fixed-free}, \end{cases} \tag*{(63)}

where we have dropped the end-force TLT_L as lower order, the dimensionless LC restoring force

FLC(ξ,τ)≡FLC(s,t)2πK/a={−∑k=1∞qk3UkΛ3[2qkwΛ−K0(qk/Λ)K0′(qk/Λ)−K2(qk/Λ)K2′(qk/Λ)]−1cos⁡[qk(1+ξ)]free-free,−∑k=1∞Qk3UkΛ3[2QkwΛ−K0(Qk/Λ)K0′(Qk/Λ)−K2(Qk/Λ)K2′(Qk/Λ)]−1sin⁡[Qk(1+ξ)]fixed-free,(64)F^{\mathrm{LC}}(\xi,\tau) \equiv\frac{F^{\mathrm{LC}}(s,t)}{2\pi K/a} = \begin{cases} -\sum_{k=1}^{\infty}\frac{q_k^3U_k}{\Lambda^3}\left[\frac{2q_k}{w\Lambda}-\frac{K_0(q_k/\Lambda)}{K_0'(q_k/\Lambda)}-\frac{K_2(q_k/\Lambda)}{K_2'(q_k/\Lambda)}\right]^{-1}\cos[q_k(1+\xi)] & \text{free-free},\\ -\sum_{k=1}^{\infty}\frac{Q_k^3U_k}{\Lambda^3}\left[\frac{2Q_k}{w\Lambda}-\frac{K_0(Q_k/\Lambda)}{K_0'(Q_k/\Lambda)}-\frac{K_2(Q_k/\Lambda)}{K_2'(Q_k/\Lambda)}\right]^{-1}\sin[Q_k(1+\xi)] & \text{fixed-free}, \end{cases} \tag*{(64)}

with qk=kπ/2q_k=k\pi/2, Qk=(2k−1)π/4Q_k=(2k-1)\pi/4, and the Fourier series

U(ξ,τ)={∑k=1∞Uk(τ)cos⁡[qk(1+ξ)]free-free,∑k=1∞Uk(τ)sin⁡[Qk(1+ξ)]fixed-free.(65)U(\xi,\tau)= \begin{cases} \sum_{k=1}^{\infty}U_k(\tau)\cos[q_k(1+\xi)] & \text{free-free},\\ \sum_{k=1}^{\infty}U_k(\tau)\sin[Q_k(1+\xi)] & \text{fixed-free}. \end{cases} \tag*{(65)}

with coefficients Uk(τ)≡uk(t)/aU_k(\tau)\equiv u_k(t)/a.

Two free-ends

In the free-free case, inserting the Fourier series for U(ξ,τ)U(\xi,\tau) and using the orthogonality of the basis functions yields

χUk′(τ)={χ2+qk23−14−qk3Er⁡Λ3[2qkwΛ−K0(qk/Λ)K0′(qk/Λ)−K2(qk/Λ)K2′(qk/Λ)]−1}Uk(τ)+∑m≠kJkmUm(τ),(66)\chi U_k'(\tau)=\left\{\frac{\chi}{2}+\frac{q_k^2}{3}-\frac{1}{4}-\frac{q_k^3}{\operatorname{Er}\Lambda^3}\left[\frac{2q_k}{w\Lambda}-\frac{K_0(q_k/\Lambda)}{K_0'(q_k/\Lambda)}-\frac{K_2(q_k/\Lambda)}{K_2'(q_k/\Lambda)}\right]^{-1}\right\}U_k(\tau)+\sum_{m\ne k}J_{km}U_m(\tau), \tag*{(66)}

for qk=πk/2q_k=\pi k/2 and

Jkm=∫−11{qm22(1−ξ2)cos⁡[qm(1+ξ)]−χqmξsin⁡[qm(1+ξ)]}cos⁡[qk(1+ξ)] dξ=−[1+(−1)k+m]m2[m2(1−χ)+k2(1+χ)](k2−m2)2,(67)\begin{aligned} J_{km}=\int_{-1}^{1}\left\{\frac{q_m^2}{2}(1-\xi^2)\cos[q_m(1+\xi)]-\chi q_m\xi\sin[q_m(1+\xi)]\right\}\cos[q_k(1+\xi)]\,\mathrm{d}\xi\\ &=-[1+(-1)^{k+m}]\frac{m^2[m^2(1-\chi)+k^2(1+\chi)]}{(k^2-m^2)^2}, \tag*{(67)} \end{aligned}

when m≠km\ne k. Here, we see that the even and odd modes about ξ=0\xi=0 have uncoupled. Since there is fairly rapid decay in kk away from k=mk=m, a first approximation is to ignore the summation in (66). Doing this yields

χUk′(τ)≈{χ2+qk23−14−qk3Er⁡Λ3[2qkwΛ−K0(qk/Λ)K0′(qk/Λ)−K2(qk/Λ)K2′(qk/Λ)]−1}Uk(τ).(68)\chi U_k'(\tau)\approx\left\{\frac{\chi}{2}+\frac{q_k^2}{3}-\frac{1}{4}-\frac{q_k^3}{\operatorname{Er}\Lambda^3}\left[\frac{2q_k}{w\Lambda}-\frac{K_0(q_k/\Lambda)}{K_0'(q_k/\Lambda)}-\frac{K_2(q_k/\Lambda)}{K_2'(q_k/\Lambda)}\right]^{-1}\right\}U_k(\tau). \tag*{(68)}

Although Λ(τ)\Lambda(\tau) is generally time-dependent, to find the buckling lengths, we shall freeze it at its instantaneous value, Λ(τ)=Λ\Lambda(\tau)=\Lambda, and pose the ansatz Uk(τ)=eσkτU^kU_k(\tau)=e^{\sigma_k\tau}\widehat{U}_k, yielding the growth rate of the kkth mode:

σk≈12+1χ{qk23−14−qk3Er⁡Λ3[2qkwΛ−K0(qk/Λ)K0′(qk/Λ)−K2(qk/Λ)K2′(qk/Λ)]−1}.(69)\sigma_k\approx\frac{1}{2}+\frac{1}{\chi}\left\{\frac{q_k^2}{3}-\frac{1}{4}-\frac{q_k^3}{\operatorname{Er}\Lambda^3}\left[\frac{2q_k}{w\Lambda}-\frac{K_0(q_k/\Lambda)}{K_0'(q_k/\Lambda)}-\frac{K_2(q_k/\Lambda)}{K_2'(q_k/\Lambda)}\right]^{-1}\right\}. \tag*{(69)}

It follows that the kkth mode becomes unstable when Λ>Λk\Lambda>\Lambda_k where Λk\Lambda_k solves

χ2+qk23−14−qk3Er⁡Λk3[2qkwΛk−K0(qk/Λk)K0′(qk/Λk)−K2(qk/Λk)K2′(qk/Λk)]−1≈0.(70)\frac{\chi}{2}+\frac{q_k^2}{3}-\frac{1}{4}-\frac{q_k^3}{\operatorname{Er}\Lambda_k^3}\left[\frac{2q_k}{w\Lambda_k}-\frac{K_0(q_k/\Lambda_k)}{K_0'(q_k/\Lambda_k)}-\frac{K_2(q_k/\Lambda_k)}{K_2'(q_k/\Lambda_k)}\right]^{-1}\approx0. \tag*{(70)}

For small Er⁡\operatorname{Er} (and thus large Λk\Lambda_k), we note that

Λk3qk3[2qkwΛk−K0(qk/Λk)K0′(qk/Λ)−K2(qk/Λk)K2′(qk/Λk)]=Λk22qk2(1−2γ+4w+log⁡4Λk2qk2)+O(1),(71)\frac{\Lambda_k^3}{q_k^3}\left[\frac{2q_k}{w\Lambda_k}-\frac{K_0(q_k/\Lambda_k)}{K_0'(q_k/\Lambda)}-\frac{K_2(q_k/\Lambda_k)}{K_2'(q_k/\Lambda_k)}\right]=\frac{\Lambda_k^2}{2q_k^2}\left(1-2\gamma+\frac{4}{w}+\log\frac{4\Lambda_k^2}{q_k^2}\right)+O(1), \tag*{(71)}

as qk/Λk→0q_k/\Lambda_k \to0, where γ\gamma is the Euler–Mascheroni constant. At leading-order in Er⁡−1\operatorname{Er}^{-1}, assumed large, we have that

Λk22qk2(1−2γ+4w+log⁡4Λk2qk2)≈4Er⁡−12χ−1+4qk2/3.(72)\frac{\Lambda_k^2}{2q_k^2}\left(1-2\gamma+\frac{4}{w}+\log\frac{4\Lambda_k^2}{q_k^2}\right)\approx\frac{4\operatorname{Er}^{-1}}{2\chi-1+4q_k^2/3}. \tag*{(72)}

which finally gives the approximation

Λk≈(2π2k2Er⁡−12χ−1+π2k2/3)1/2[W(32Er⁡−1e1−2γ+4/w2χ−1+π2k2/3)]−1/2,(73)\Lambda_k\approx\left(\frac{2\pi^2 k^2\operatorname{Er}^{-1}}{2\chi-1+\pi^2k^2/3}\right)^{1/2}\left[W\left(\frac{32\operatorname{Er}^{-1}e^{1-2\gamma+4/w}}{2\chi-1+\pi^2k^2/3}\right)\right]^{-1/2}, \tag*{(73)}

where W(z)W(z) is the Lambert WW function (the solution to Wexp⁡W=zW\exp W=z). Using the rudimentary approximation of W(z)≈log⁡zW(z)\approx\log z for large zz, this expression can be approximated further as

Λk≈(2π2k2Er⁡−12χ−1+π2k2/3)1/2[log⁡(32Er⁡−12χ−1+π2k2/3)+1−2γ+4w]−1/2.(74)\Lambda_k\approx\left(\frac{2\pi^2 k^2\operatorname{Er}^{-1}}{2\chi-1+\pi^2k^2/3}\right)^{1/2}\left[\log\left(\frac{32\operatorname{Er}^{-1}}{2\chi-1+\pi^2k^2/3}\right)+1-2\gamma+\frac{4}{w}\right]^{-1/2}. \tag*{(74)}

The buckling length of the filament is then given by the smallest critical length, Λ∗=min⁡kΛk\Lambda^*=\min_k\Lambda_k. For χ=5\chi=5, Er⁡=10−6\operatorname{Er}=10^{-6}, and w=10−2w=10^{-2}–10010^0, we find that the smallest mode is k=1k=1, which gives the estimate to dimensionless buckling Λ∗≈62.2\Lambda^*\approx62.2–294294. This approximation is compared to numerical solutions of the coupled system, Eq. (66), in the main text Figure 5. Although the numerical results are not quantitatively recovered, the approximation does recover the qualitative behavior. It thus inspires the fitting expression presented in the main text:

Λ∗=(a1Er⁡−1log⁡(a2Er⁡−1)+4/w)1/2,(75)\Lambda^*=\left(\frac{a_1\operatorname{Er}^{-1}}{\log(a_2\operatorname{Er}^{-1})+4/w}\right)^{1/2}, \tag*{(75)}

for fitting parameters aia_i.

Fixed-free ends

In the fixed-free case, inserting the Fourier series for U(ξ,τ)U(\xi,\tau) and using the orthogonality of the basis functions yields a similar expression to before:

χUk′(τ)={4Qk23−14−Qk3Er⁡Λ3[2QkwΛ−K0(Qk/Λ)K0′(Qk/Λ)−K2(Qk/Λ)K2′(Qk/Λ)]−1}Uk(τ)+∑m≠kIkmUm(τ),(76)\chi U_k'(\tau)=\left\{\frac{4Q_k^2}{3}-\frac{1}{4}-\frac{Q_k^3}{\operatorname{Er}\Lambda^3}\left[\frac{2Q_k}{w\Lambda}-\frac{K_0(Q_k/\Lambda)}{K_0'(Q_k/\Lambda)}-\frac{K_2(Q_k/\Lambda)}{K_2'(Q_k/\Lambda)}\right]^{-1}\right\}U_k(\tau)+\sum_{m\ne k}I_{km}U_m(\tau), \tag*{(76)}

for Qk=π(2k−1)/4Q_k=\pi(2k-1)/4 and

Ikm=∫−11{Qm22(1−ξ)(3+ξ)sin⁡[Qm(1+ξ)]+χQmξcos⁡[Qm(1+ξ)]}sin⁡[Qk(1+ξ)] dξ=−(−1)k+m(2m−1)2[(2k−1)2+(2m−1)2]8(k−m)2(k+m−1)2−(2m−1)[2k−1+(−1)k+m(2m−1)]χ4(k−m)(k+m−1),(77)\begin{aligned} I_{km}=\int_{-1}^{1}\left\{\frac{Q_m^2}{2}(1-\xi)(3+\xi)\sin[Q_m(1+\xi)]+\chi Q_m\xi\cos[Q_m(1+\xi)]\right\}\sin[Q_k(1+\xi)]\,\mathrm{d}\xi\\ &=-\frac{(-1)^{k+m}(2m-1)^2\left[(2k-1)^2+(2m-1)^2\right]}{8(k-m)^2(k+m-1)^2}-\frac{(2m-1)\left[2k-1+(-1)^{k+m}(2m-1)\right]\chi}{4(k-m)(k+m-1)}, \tag*{(77)} \end{aligned}

when m≠km\ne k.

Again, omitting the summation in Eq. (76) and taking the ansatz Uk(τ)=eσkτU^kU_k(\tau)=e^{\sigma_k\tau}\hat{U}_k yields an approximation for the growth rate of the kkth mode,

σk≈1χ{4Qk23−14−Qk3Er⁡Λ3[2QkwΛ−K0(Qk/Λ)K0′(Qk/Λ)−K2(Qk/Λ)K2′(Qk/Λ)]−1}.(78)\sigma_k\approx\frac{1}{\chi}\left\{\frac{4Q_k^2}{3}-\frac{1}{4}-\frac{Q_k^3}{\operatorname{Er}\Lambda^3}\left[\frac{2Q_k}{w\Lambda}-\frac{K_0(Q_k/\Lambda)}{K_0'(Q_k/\Lambda)}-\frac{K_2(Q_k/\Lambda)}{K_2'(Q_k/\Lambda)}\right]^{-1}\right\}. \tag*{(78)}
Analytical approximation for the dimensionless buckling length as a function of the Ericksen number

Figure S1. The analytical approximation [Eq. 55] for the dimensionless buckling length, Λ∗\Lambda^*, as a function of the Ericksen number, Er\mathrm{Er}, for free-free (dashed curves) and fixed-free (solid curves) boundary conditions. A fixed end reduces the buckling length by a factor of ≈1.5\approx1.5.

For small Er\mathrm{Er}, making the same approximations as before yields the fact that the kkth mode becomes unstable when Λ>Λk\Lambda> \Lambda_k, where

Λk≈((2k−1)2π2Er−1/2π2(2k−1)2/3−1)1/2[W(32Er−1π2(2k−1)2/3−1e1−2γ+4/w)]−1/2.(79)\Lambda_k \approx\left(\frac{(2k-1)^2\pi^2\mathrm{Er}^{-1/2}}{\pi^2(2k-1)^{2/3}-1}\right)^{1/2}\left[W\left(\frac{32\mathrm{Er}^{-1}}{\pi^2(2k-1)^{2/3}-1}e^{1-2\gamma+4/w}\right)\right]^{-1/2}. \tag*{(79)}

or, with the additional rudimentary approximation,

Λk≈((2k−1)2π2Er−1/2π2(2k−1)2/3−1)1/2[log⁡(32Er−1π2(2k−1)2/3−1)+1−2γ+4w]−1/2.(80)\Lambda_k \approx\left(\frac{(2k-1)^2\pi^2\mathrm{Er}^{-1/2}}{\pi^2(2k-1)^{2/3}-1}\right)^{1/2}\left[\log\left(\frac{32\mathrm{Er}^{-1}}{\pi^2(2k-1)^{2/3}-1}\right)+1-2\gamma+\frac{4}{w}\right]^{-1/2}. \tag*{(80)}

For χ=5\chi= 5, Er=10−6\mathrm{Er} = 10^{-6}, and w=10−2w = 10^{-2}–10010^0, here the smallest value of Λk\Lambda_k is found to range between when k=6k = 6 (for w=10−2w = 10^{-2}) and k=2k = 2 (for w=100w = 10^0), giving the estimate to dimensionless buckling Λ∗≈60.5\Lambda^* \approx60.5–296296. An improved estimation is found by fitting (55) to the coupled system, as presented in the main text.

Supplementary movies

Supplementary movies are available on Zenodo (https://doi.org/10.5281/zenodo.17917861).

  1. Movie S1: Non-motile E. coli proliferating in LB media without DSCG.

  1. Movie S2: Non-motile E. coli proliferating in LB media supplemented with 15 w/w% DSCG.

  1. Movie S3: Non-motile E. coli proliferating in LB media supplemented with 15 w/w% DSCG. At t=0t = 0 min we flush a suspension of chains with LC free liquid at a flow rate of 2.5 μ\muL/min, thus decreasing the LC concentration as time increases.

  1. Movie S4: Non-motile E. coli proliferating in LB media supplemented with 15 w/w% DSCG. At t=0t = 0 min we flush a suspension of chains with 15 w/w% DSCG solution at a flow rate of 2.5 μ\muL/min keeping the LC concentration constant as time increases.

E. coli doubling times are similar across different LC conditions

Figure S3. E. coli doubling times are similar across different LC conditions. (A) and (B) show the growth curves and doubling times for E. coli cells in different LC solutions. The growth curves were obtained in triplicate where the solid curve represents the mean among all samples while the shaded regions are the corresponding standard deviations. The doubling times were measured by fitting the first 500 min of each sample’s growth curve to an exponential equation: Afrac=A0eαtA_{\mathrm{frac}} = A_0 e^{\alpha t} where tt is time, A0A_0 is the area fraction at t=0t = 0, α\alpha is the maximal growth rate, and AfracA_{\mathrm{frac}} is the cell collective’s total area fraction. We convert the bacteria’s maximal growth rate, α\alpha, to a doubling time via td=log⁡(2)/αt_d = \log(2)/\alpha. We observe that, throughout all LC concentrations tested, E. coli’s doubling time remains between 52–68 min and does not depend appreciably on LC concentration. We obtained the growth curves under different LC concentrations via segmentation of images in MATLAB to then plot the area fraction that cells occupy as a function of time.

References

References

  1. [1]N. Figueroa-Morales, L. Dominguez-Rubio, T. L. Ott, and I. S. Aranson, Scientific reports 9, 9713 (2019).
  2. [2]C. Viney, A. E. Huber, and P. Verdugo, Macromolecules 26, 852 (1993).
  3. [3]J. M. Davies and C. Viney, Thermochimica acta 315, 39 (1998).
  4. [4]A. Repula, E. Abraham, V. Cherpak, and I. I. Smalyukh, Proceedings of the National Academy of Sciences 119, e2200930119 (2022).
  5. [5]P. R. Secor, L. K. Jennings, L. A. Michaels, J. M. Sweere, P. K. Singh, W. C. Parks, and P. L. Bollyky, Microbial Cell 3, 49 (2015).
  6. [6]P. R. Secor, J. M. Sweere, L. A. Michaels, A. V. Malkovskiy, D. Lazzareschi, E. Katznelson, J. Rajadas, M. E. Birnbaum, A. Arrigoni, K. R. Braun, et al., Cell host & microbe 18, 549 (2015).
  7. [7]A. K. Tarafder, A. von Kugelgen, A. J. Mellul, U. Schulze, D. G. A. L. Aarts, and T. A. M. Bharat, Proceedings of the National Academy of Sciences 117, 4774 (2020).
  8. [8]P. C. Mushenheim, R. R. Trivedi, H. H. Tuson, D. B. Weibel, and N. L. Abbott, Soft Matter 10, 88 (2014).
  9. [9]S. Zhou, A. Sokolov, O. D. Lavrentovich, and I. S. Aranson, Biophysical Journal 106, 420a (2014).
  10. [10]M. S. Krieger, S. E. Spagnolie, and T. Powers, Soft Matter 11, 9115 (2015).
  11. [11]M. M. Genkin, A. Sokolov, O. D. Lavrentovich, and I. S. Aranson, Physical Review X 7, 011029 (2017).
  12. [12]Z. Lin, S. Chen, and T. Gao, Journal of Fluid Mechanics 921, A25 (2021).
  13. [13]M. Goral, E. Clement, T. Darnige, T. Lopez-Leon, and A. Lindner, Interface Focus 12, 20220039 (2022).
  14. [14]A. G. Prabhune, A. S. Garcia-Gordillo, I. S. Aranson, T. R. Powers, and N. Figueroa-Morales, PRX Life 2, 033004 (2024).
  15. [15]Z. Mou, Y. Li, Z. You, and R. Zhang, Physical Review E 111, 065410 (2025).
  16. [16]M. M. Genkin, A. Sokolov, O. D. Lavrentovich, and I. S. Aranson, Phys. Rev. X 7, 011029 (2017).
  17. [17]T. Turiv, R. Koizumi, K. Thijssen, M. M. Genkin, H. Yu, C. Peng, Q.-H. Wei, J. M. Yeomans, I. S. Aranson, A. Doostmohammadi, et al., Nature Physics 16, 481 (2020).
  18. [18]H. Chi, A. Gavrikov, L. Berlyand, and I. S. Aranson, Commun. Phys. 5, 274 (2022).
  19. [19]N. Figueroa-Morales, M. M. Genkin, A. Sokolov, and I. S. Aranson, Commun. Phys. 5, 301 (2022).
  20. [20]J. Li, L. Ohm, and S. E. Spagnolie, Phys. Rev. Fluids 10, 083301 (2025).
  21. [21]H. Baza, F. Chen, T. Turiv, S. Shiyanovskii, and O. D. Lavrentovich, Soft Matter (2025).
  22. [22]J. Wu, M. Liu, Z. Mou, Y. Li, R. Wang, Z. Asilehan, Q. Shi, Y. Cheng, Z. You, R. Zhang, et al., Nature Communications 16, 10605 (2025).
  23. [23]Y.-c. Fung, Biomechanics: mechanical properties of living tissues (Springer Science & Business Media, 2013).
  24. [24]A. L. Koch, Bacterial growth and form (Springer Science & Business Media, 2013).
  25. [25]R. Farhadifar, J.-C. Roper, B. Aigouy, S. Eaton, and F. Julicher, Curr. Biol. 17, 2095 (2007).
  26. [26]Z. You, D. J. Pearce, A. Sengupta, and L. Giomi, Phys. Rev. X 8, 031065 (2018).
  27. [27]O. Hallatschek, S. S. Datta, K. Drescher, J. Dunkel, J. Elgeti, B. Waclaw, and N. S. Wingreen, Nature Rev. Phys. 5, 407 (2023).
  28. [28]S. Weady, B. Palmer, A. Lamson, T. Kim, R. Farhadifar, and M. J. Shelley, Phys. Rev. Lett. 133, 158402 (2024).
  29. [29]J. A. Shapiro, Scientific American 258, 62 (1988).
  30. [30]O. Hallatschek, S. S. Datta, K. Drescher, J. Dunkel, J. Elgeti, B. Waclaw, and N. S. Wingreen, Nature Reviews Physics 5, 407 (2023).
  31. [31]S. E. Spagnolie and P. T. Underhill, Annual Review of Condensed Matter Physics 14, 381 (2023).
  32. [32]A. Folkesson, L. Jelsbak, L. Yang, H. K. Johansen, O. Ciofu, N. Hoiby, and S. Molin, Nature Reviews Microbiology 10, 841 (2012).
  33. [33]M. H. Rau, S. K. Hansen, H. K. Johansen, L. E. Thomsen, C. T. Workman, K. F. Nielsen, L. Jelsbak, N. Hoiby, L. Yang, and S. Molin, Environmental microbiology 12, 1643 (2010).
  34. [34]L. Cullen and S. McClean, Pathogens 4, 66 (2015).
  35. [35]K. M. Blair, L. Turner, J. T. Winkelman, H. C. Berg, and D. B. Kearns, science 320, 1636 (2008).
  36. [36]K. Sauer, A. K. Camper, G. D. Ehrlich, J. W. Costerton, and D. G. Davies, Pseudomonas aeruginosa displays multiple phenotypes during development as a biofilm (2002).DOI
  37. [37]C. Fei, S. Mao, J. Yan, R. Alert, H. A. Stone, B. L. Bassler, N. S. Wingreen, and A. Kosmrlj, Proceedings of the National Academy of Sciences 117, 7622 (2020).
  38. [38]M. Trejo, C. Douarche, V. Bailleux, C. Poulard, S. Mariot, C. Regeard, and E. Raspaud, Proceedings of the National Academy of Sciences 110, 2011 (2013).
  39. [39]A. Martinez-Calvo, T. Bhattacharjee, R. K. Bay, H. N. Luu, A. M. Hancock, N. S. Wingreen, and S. S. Datta, Proceedings of the National Academy of Sciences 119, e2208019119 (2022).
  40. [40]H. Fujikawa and M. Matsushita, Journal of the Physical Society of Japan 58, 3875 (1989).
  41. [41]F. D. Farrell, O. Hallatschek, D. Marenduzzo, and B. Waclaw, Physical Review Letters 111 (2013).
  42. [42]B. Bottura, L. M. Rooney, P. A. Hoskisson, and G. McConnell, Biofilm 4, 100084 (2022).
  43. [43]O. Hallatschek, P. Hersen, S. Ramanathan, and D. R. Nelson, Proceedings of the National Academy of Sciences 104, 19926 (2007).
  44. [44]M. J. I. Muller, B. I. Neugeboren, D. R. Nelson, and A. W. Murray, Proceedings of the National Academy of Sciences 111, 1037 (2014).
  45. [45]F. D. Farrell, M. Gralka, O. Hallatschek, and B. Waclaw, Journal of The Royal Society Interface 14, 20170073 (2017).
  46. [46]P. R. Secor, L. A. Michaels, A. Ratjen, L. K. Jennings, and P. K. Singh, Proceedings of the National Academy of Sciences of the United States of America 115, 10.1073/pnas.1806005115 (2018).
  47. [47]M. Whitley, M. G. Bangera, R. E. Bumgarner, M. R. Parsek, G. M. Teitzel, S. Lory, and E. P. Greenberg, Nature 413, 860 (2001).
  48. [48]T.-F. Mah, B. Pitts, B. Pellock, G. C. Walker, P. S. Stewart, and G. A. O’Toole, Nature 426, 306 (2003).
  49. [49]K. D. Young, Microbiology and molecular biology reviews 70, 660 (2006).
  50. [50]K. D. Young, Current opinion in microbiology 10, 596 (2007).
  51. [51]S. Shikano, L. S. Luckinbill, and Y. Kurihara, Microbial Ecology 20, 75 (1990).
  52. [52]M. W. Hahn and M. G. Hofle, Applied and Environmental Microbiology 65, 4863 (1999).
  53. [53]T. Posch, K. Simek, J. Vrba, J. Pernthaler, B. Sattler, B. Sonntag, R. Psenner, et al., Aquatic Microbial Ecology 18, 235 (1999).
  54. [54]G. Corno and K. Jurgens, Applied and environmental microbiology 72, 78 (2006).
  55. [55]Y. Cheli, R. J. Stjevick, E. Kornobis, V. Briolat, J.-M. Ghigo, and C. Beloin, Microbiology Spectrum 11, e00690 (2023).
  56. [56]E.-s. Q. Nwoko and I. N. Okeke, Biochemical Society Transactions 49, 1147 (2021).
  57. [57]L.-L. Cheng, Y.-Y. Luk, C. J. Murphy, B. A. Israel, and N. L. Abbott, Biomaterials 26, 7173 (2005).
  58. [58]S. Helfinstine, O. Laurentovich, and C. Woolverton, Letters in applied microbiology 43, 27 (2006).
  59. [59]N. Zimmermann, G. Junnemann-Held, P. J. Collings, and H.-S. Kitzerover, Soft Matter 11, 1547 (2015).
  60. [60]S. Zhou, in Lyotropic Chromonic Liquid Crystals: From Viscoelastic Properties to Living Liquid Crystals (Springer, 2017) pp. 51–75.DOI
  61. [61]A. Suh and D. K. Yoon, Scientific Reports 8, 9460 (2018).
  62. [62]S. M. Rubinstein, I. Kolodkin-Gal, A. Mcloon, L. Chai, R. Kolter, R. Losick, and D. A. Weitz, Molecular microbiology 86, 426 (2012).
  63. [63]R. Buda, Y. Liu, J. Yang, S. Hegde, K. Stevenson, F. Bai, and T. Pilizota, Proceedings of the National Academy of Sciences 113, E5838 (2016).
  64. [64]D. S. Cayley, H. J. Guttman, and M. T. Record, Biophysical Journal 78 (2000).
  65. [65]Y. Deng, M. Sun, and J. W. Shaevitz, Physical Review Letters 107 (2011).
  66. [66]O. D. Lavrentovich, Liquid crystals 24, 117 (1998).
  67. [67]P. Poulin and D. Weitz, Physical Review E 57, 626 (1998).
  68. [68]I. I. Smalyukh, J. Butler, J. D. Shrout, M. R. Parsek, and G. C. L. Wong, Physical Review E 78, 030701 (2008).
  69. [70]P. C. Mushenheim, R. R. Trivedi, H. H. Tuson, D. B. Weibel, and N. L. Abbott, Soft Matter 10, 88 (2014).
  70. [71]I. I. Smalyukh, J. Butler, J. D. Shrout, M. R. Parsek, and G. C. Wong, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics 78, 030701 (2008).
  71. [72]M. Tasinkevych, F. Mondiot, O. Mondain-Monval, and J.-C. Loudet, Soft Matter 10, 2047 (2014).
  72. [73]M. D. Graham, Microhydrodynamics, Brownian motion, and complex fluids, Vol. 58 (Cambridge University Press, 2018).DOI
  73. [74]D. Vella, Nat. Rev. Phys. 1, 425 (2019).
  74. [75]L. Li, H. Manikantan, D. Saintillan, and S. E. Spagnolie, J. Fluid Mech. 735, 705 (2013).
  75. [76]H. Manikantan and D. Saintillan, Physical Review E 92, 041002 (2015).
  76. [77]O. Du Roure, A. Lindner, E. N. Nazockdast, and M. J. Shelley, Annual Review of Fluid Mechanics 51, 539 (2019).
  77. [78]B. Chakrabarti, Y. Liu, J. LaGrone, R. Cortez, L. Fauci, O. Du Roure, D. Saintillan, and A. Lindner, Nature Physics 16, 689 (2020).
  78. [79]P.-G. De Gennes and J. Prost, The physics of liquid crystals, 83 (Oxford university press, 1993).
  79. [80]B. J. Walker, K. Ishimoto, H. Gadelha, and E. A. Gaffney, J. Fluid Mech. 879, 808 (2019).
  80. [81]I. Duchesne, S. Rainville, and T. Galstian, Biophysical journal 109, 2137 (2015).
  81. [82]M. Gomez-Gonzalez and J. C. Del Alamo, Soft matter 12, 5758 (2016).
  82. [83]S. Zhou, O. Tovkach, D. Golovaty, A. Sokolov, I. S. Aranson, and O. D. Lavrentovich, New Journal of Physics 19, 055006 (2017).
  83. [84]W. Chi, M. Potomkin, L. Zhang, L. Berlyand, and I. S. Aranson, Communications Physics 3, 162 (2020).
  84. [85]D. P. Holmes and A. J. Crosby, Physical review letters 105, 038303 (2010).
  85. [86]C. P. Brangwynne, F. C. MacKintosh, S. Kumar, N. A. Geisse, J. Talbot, L. Mahadevan, K. K. Parker, D. E. Ingber, and D. A. Weitz, The journal of cell biology 173, 733 (2006).
  86. [87]K. Moor, M. Diard, M. E. Sellin, B. Felmy, S. Y. Wotzka, A. Toska, E. Bakkeren, M. Arnoldini, F. Bansetp, A. D. Co, T. Voller, A. Minola, B. Fernandez-Rodriguez, G. Agatic, S. Barbieri, L. Piccoli, C. Casiraghi, D. Corti, A. Lanzavecchia, R. R. Regoes, C. Loverdo, R. Stocker, D. R. Brumley, W.-D. Hardt, and E. Slack, Nature 544, 498 (2017).
  87. [88]R. Mishra, M. Hannebelle, V. P. Patil, A. Dubois, C. Garcia-Mouton, G. M. Kirsch, M. Jan, K. Sharma, N. Guex, J. Sordet-Dessimoz, J. Perez-Gil, M. Prakash, G. W. Knott, N. Dhar, J. D. McKinney, and V. V. Thacker, Cell , S0092867423010371 (2023).
  88. [89]S. Azimi, M. Thomas, S. E. Cleland, J. E. Curtis, J. B. Goldberg, and S. P. Diggle, MBio 12, 10 (2021).
  89. [90]D. Puri, X. Fang, and K. R. Allison, iScience 26, 105795 (2023).
  90. [91]S. Gonzalez La Corte, C. A. Stevens, G. Carcamo-Oyarce, K. Ribbeck, N. S. Wingreen, and S. S. Datta, Science Advances 11, eadg7797 (2025).
  91. [92]R. M. Losick, Current Biology 30, R1146 (2020).
  92. [93]R. Chawla, R. Gupta, T. P. Lele, and P. P. Lele, Journal of molecular biology 432, 523 (2020).
  93. [94]F. Gomand, W. H. Mitchell, J. Burgain, J. Petit, F. Borges, S. E. Spagnolie, and C. Gaiani, Soft Matter 16, 9273 (2020).
  94. [95]N. Edwards, S. Beeton, A. T. Bull, and J. C. Merchuk, Applied microbiology and biotechnology 30, 190 (1989).
  95. [96]R. Mondiot, R. Botet, P. Snabre, O. Mondain-Monval, and J.-C. Loudet, Proceedings of the National Academy of Sciences 111, 5831 (2014).
  96. [97]D. G. Sudha, J. Ochoa, and L. S. Hirst, Soft Matter 17, 7532 (2021).
  97. [98]R. P. Trivedi, M. Tasinkevych, and I. I. Smalyukh, Physical Review E 94, 062703 (2016).
  98. [99]M. Sezgin and B. I. Sankur, Journal of Electronic Imaging 13, 146 (2004).
  99. [100]P. van der Schoot, The Journal of Chemical Physics 112, 9132 (2000).
  100. [101]H. N. W. Lekkerkerker, Colloids and the depletion interaction / Henk N. W. Lekkerkerker, Remco Tuinier (Springer, Dordrecht ; 2011). series Title: Lecture Notes in physics, v. 833 Publication Title: Colloids and the depletion interaction.
  101. [102]T. Vissers, A. T. Brown, N. Koumakis, A. Dawson, M. Hermes, J. Schwarz-Linek, A. B. Schofield, J. M. French, V. Kout-sos, J. Arlt, et al., Science Advances 4, eaao1170 (2018).
  102. [103]A. Stepulane, A. K. Rajasekharan, and M. Andersson, ACS Applied Bio Materials 5, 5289 (2022).
  103. [104]T. G. J. Chandler and S. E. Spagnolie, Phys. Rev. Fluids 9, 110511 (2024).
  104. [105]A. Rapini and M. Papoular, J. Phys. Colloq. 30, C4 (1969).
  105. [106]A. Habibi, C. Blanc, N. B. Mbarek, and T. Soltani, Journal of Molecular Liquids 288, 111027 (2019).
  106. [107]S. Zhou, K. Neupane, Y. A. Nastishin, A. R. Baldwin, S. V. Shiyanovskii, O. D. Lavrentovich, and S. Sprunt, Soft Matter 10, 6571 (2014).

Paper details

vlnc:W_vzbk452s
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents