This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Present Address: ]Simplex Holdings, Inc., 19F Toranomon Hills Mori Tower, 1-23-1 Toranomon, Minato-ku, Tokyo 105-6319, Japan; [email protected]

Theory of rigidity and numerical analysis of density of states of two-dimensional amorphous solids with dispersed frictional grains in the linear response regime

Daisuke Ishima1 [    Kuniyasu Saitoh2, Michio Otsuki3 and Hisao Hayakawa1 1Yukawa Institute for Theoretical Physics, Kyoto University,
Kitashirakawa-oiwake cho, Sakyo-ku, Kyoto 606-8502, Japan,
2Department of Physics, Faculty of Science, Kyoto Sangyo University,
Motoyama, Kamigamo, Kita-ku, Kyoto 603-8555, Japan,
3Graduate School of Engineering Science, Osaka University, Toyonaka, Osaka 560-8531, Japan
Abstract

Using the Jacobian matrix, we obtain theoretical expression of rigidity and the density of states of two-dimensional amorphous solids consisting of frictional grains in the linear response to an infinitesimal strain, in which we ignore the dynamical friction caused by the slip processes of contact points. The theoretical rigidity agrees with that obtained by molecular dynamics simulations. We confirm that the rigidity is smoothly connected to the value in the frictionless limit. For the density of states, we find that there are two modes in the density of states for sufficiently small kT/kNk_{T}/k_{N}, which is the ratio of the tangential to normal stiffness. Rotational modes exist at low frequencies or small eigenvalues, whereas translational modes exist at high frequencies or large eigenvalues. The location of the rotational band shifts to the high-frequency region with an increase in kT/kNk_{T}/k_{N} and becomes indistinguishable from the translational band for large kT/kNk_{T}/k_{N}.

I Introduction

Amorphous materials consisting of dispersed grains such as powders, colloids, bubbles, and emulsions are ubiquitous in nature Jaeger96 ; Durian95 ; Wyart05 ; Baule18 . These materials behave like liquids at low densities and exhibit solid-like mechanical responses above their jamming point Liu98 . In systems consisting of frictionless grains, the rigidity changes continuously, but the coordination number of grains changes discontinuously at the jamming point as a function of density Durian95 ; Wyart05 ; Hern03 . The critical behavior near the jamming point is of interest to physicists as a non-equilibrium phase transition Hern01 ; Trappe01 ; Zhang05 ; Majmudar07 ; Ciamarra11 . Dispersed grains above the jamming point are fragile and exhibit softening and yielding transition under certain loads Nagamanasa14 ; Knowlton14 ; Kawasaki16 ; Leishangthem17 ; Bohy17 ; Ozawa18 ; Clark18 ; Boschan19 ; Singh20 ; Otsuki21 .

For amorphous solids consisting of frictionless grains, it is useful to analyze the dynamical matrix or the Hessian matrix, which is defined as the second derivative of the potential of a collection of grains with respect to the displacements from their stable configuration Wyart05 ; WyarLettt05 ; Ellenbroek06 ; Lerner16 ; Gartner16 ; Bonfanti19 ; Baule18 . For instance, the rigidity can be determined by eigenvalues and eigenvectors DeGiuli14 ; Mizuno16 ; Maloney04 ; Maloney06 ; Lemaitre06 ; Zaccone11 . It has been reported that the minimum nonzero eigenvalue of the Hessian matrix decreases with increasing strain and eventually becomes negative, where an irreversible stress drop takes place Maloney04 ; Maloney06 ; Manning11 ; Dasgupta12 ; Ebrahem20 .

It is known that amorphous solids have characteristic properties at low temperatures (e.g., thermal conductivity and specific heat) that are quite different from those of crystalline solids since a long-time ago Zeller71 . These days, we have recognized that amorphous solids consisting of dispersed grains exhibit unique elastic-plastic behavior as a mechanical response to an applied strain Nicolas18 . Because these properties are related to the density of states (DOS), there have been lots of studies on the DOS Hern03 ; Schirmacher07 ; Lerner16 ; Mizuno17 ; Wang19 . The DOS for systems composed of anisotropic grains, such as ellipses, dimers, deformable grains, and grains with rough surfaces, have been studied with the aid of the Hessian matrix Zeravcic09 ; Mailman09 ; Schreck10 ; Yunker11 ; Schreck12 ; Shiraishi19 ; Papanikolaou13 ; Britoa18 ; Treado21 ; Ikeda20 ; Ikeda21 . Because of the rotation of such anisotropic grains, there exists a rotational band in the DOS that is distinguishable from the translational band Mailman09 ; Yunker11 ; Schreck12 ; Britoa18 ; Ikeda20 ; Ikeda21 .

Even for systems of spherical grains that cannot be free from inter-particle friction, similar results are expected as a result of grain rotations. However, few studies have reported the existence of rotational bands in the DOS. Because the frictional force between the grains depends on the contact history, it cannot be expressed as a conservative force. Therefore, stability analysis for frictional grains based on the Hessian cannot be used. Nevertheless, the Hessian analysis using an effective potential for frictional grains has been performed Somfai07 ; Henkes10 . Recently, Liu et al. suggested that the Hessian analysis with another effective potential can be used even if slip processes exist Liu21 . The previous studies Somfai07 ; Henkes10 reported that friction between grains causes a continuous change in the functional form of the DOS from that of frictionless systems. However, there are only a few reports on whether an isolated band in the DOS originating from friction between grains is visible at lower frequencies.

Recently, Chattoraj et al. discussed the stability of the grain configuration under strain using the Jacobian matrix of frictional grains Chattoraj19 . They performed eigenvalue analysis under athermal quasi-static shear processes and determined the existence of oscillatory instability originating from inter-particle friction at a certain strain Chattoraj19 ; Chattoraj19E ; Charan20 . However, they did not discuss the rigidity or the DOS.

The theoretical determination of the rigidity of amorphous solids consisting of frictional grains is important for controlling amorphous solids. However, we do not know how to determine the rigidity from the Jacobian for the frictional grains.

The purpose of this study is to clarify the role of mutual friction between grains in terms of the rigidity and DOS. We focus on the response to an infinitesimal strain from a stable configuration of grains without any strain to obtain tangible results. In this study, we assume that there is no slip between grains because of an infinitesimal strain, and we then deal with friction as static friction.

The remainder of this paper is organized as follows. In the next section, we introduce the numerical method. In Sec. III, we introduce the Jacobian. Section IV consists of Sec IV.1, which deals with the theoretical prediction of rigidity in the linear response regime, and Sec. IV.2, which deals with the DOS. In the final section, we summarize the results of our study and discuss future work. The appendix consists of seven sections. In Appendix A, we summarize the method for preparing a stable grain configuration before applying shear. In Appendix B, we explain the implementation of the numerical integration method in the proposed system. In Appendix C, we summarize some properties of the Jacobian. In Appendix D, we present the explicit expressions of the Jacobian. In Appendix E, we investigate the effects of rattlers. In Appendix F, we write down the explicit results of Jacobian. In Appendix G, we derive the theoretical prediction of rigidity using the Jacobian. In Appendix H, we introduce the DOS using the Hessian analysis. In Appendix I, we investigate the system size dependence of the DOS. In Appendix J, we study the density dependence of the DOS.

II Numerical model

Our system contains NN frictional spherical particles embedded in a monolayer configuration. We treat this system as a two-dimensional system (see Fig. 1).

Refer to caption
Figure 1: Schematics of our system.

To prevent the system from crystallizing Luding01 , we prepare an equal number of particles with diameters dd and d/1.4d/1.4. We assume that the mass of particle ii is proportional to di2d_{i}^{2}, where did_{i} is the diameter of ii-th particle. We introduce mm as the mass of a particle with diameter dd. In this study, xi,yix_{i},y_{i}, and θi\theta_{i} denote x,yx,y coordinates, and the rotational angle of the ii-th particle, respectively. We introduce the generalized coordinates of the ii-th particle 𝒒i:=(𝒓iT,i)T\bm{q}_{i}:=(\bm{r}_{i}^{\textrm{T}},\ell_{i})^{\textrm{T}} with 𝒒i:=(xi,yi)T\bm{q}_{i}:=(x_{i},y_{i})^{\textrm{T}} and i:=diθi/2\ell_{i}:=d_{i}\theta_{i}/2, where the superscript T denotes the transposition.

Let the force, and zz-component of the torque acting on the ii-th particle be 𝑭i:=(Fix,Fiy)T\bm{F}_{i}:=(F_{i}^{x},F_{i}^{y})^{\textrm{T}} and TiT_{i}, respectively. Then, the equations of motion of ii-th particle are expressed as

mid2𝒓idt2\displaystyle m_{i}\frac{d^{2}\bm{r}_{i}}{dt^{2}} =𝑭i,\displaystyle=\bm{F}_{i}, (1)
Iid2θidt2\displaystyle I_{i}\frac{d^{2}\theta_{i}}{dt^{2}} =Ti\displaystyle=T_{i} (2)

with the mass mim_{i} and the momentum of inertia Ii:=midi2/8I_{i}:=m_{i}d_{i}^{2}/8 of ii-th particle. In a system without volume forces such as gravity, we can write

𝑭i\displaystyle\bm{F}_{i} =ji𝒇ij,\displaystyle=\sum_{j\neq i}\bm{f}_{ij}, (3)
Ti\displaystyle T_{i} =jiTij,\displaystyle=\sum_{j\neq i}T_{ij}, (4)

where 𝒇ij\bm{f}_{ij} and TijT_{ij} are the force and zz-component of the torque acting on the ii-th particle from the jj-th particle, respectively. Here, TijT_{ij} is given by

Tij=di2(nijxfijynijyfijx),\displaystyle T_{ij}=-\frac{d_{i}}{2}({n}_{ij}^{x}{f}_{ij}^{y}-{n}_{ij}^{y}{f}_{ij}^{x}), (5)

where we have introduced the normal unit vector between ii and jj particles as 𝒏ij:=𝒓ij/|𝒓ij|:=(𝒓i𝒓j)/|𝒓i𝒓j|\bm{n}_{ij}:=\bm{r}_{ij}/|\bm{r}_{ij}|:=(\bm{r}_{i}-\bm{r}_{j})/|\bm{r}_{i}-\bm{r}_{j}|. Here, nijζn_{ij}^{\zeta} and fijζf_{ij}^{\zeta} refer to ζ\zeta-components of 𝒏ij\bm{n}_{ij} and 𝒇ij\bm{f}_{ij}, respectively. Note that ζ\zeta expresses ζ=x\zeta=x or yy throughout this study. The force 𝒇ij\bm{f}_{ij} can be divided into normal 𝒇N,ij\bm{f}_{N,ij} and tangential 𝒇T,ij\bm{f}_{T,ij} parts as

𝒇ij=(𝒇N,ij+𝒇T,ij)Θ(dij/2|𝒓ij|),\displaystyle\bm{f}_{ij}=(\bm{f}_{N,ij}+\bm{f}_{T,ij})\Theta(d_{ij}/2-|\bm{r}_{ij}|), (6)

where dij:=di+djd_{ij}:=d_{i}+d_{j} and Θ(x)\Theta(x) is Heaviside’s step function, taking Θ(x)=1\Theta(x)=1 for x>0x>0 and Θ(x)=0\Theta(x)=0 otherwise. We model the repulsive force between the contacted particles ii and jj as the Hertzian force in addition to the dissipative force proportional to the relative velocity with a damping constant ηD\eta_{D} Liu21 as follows:

𝒇N,ij:\displaystyle\bm{f}_{N,ij}: =kNξN,ij3/2𝒏ijηD𝒗N,ij,\displaystyle=k_{N}\xi_{N,ij}^{3/2}\bm{n}_{ij}-\eta_{D}\bm{v}_{N,ij}, (7)
𝒇T,ij:\displaystyle\bm{f}_{T,ij}: =kTξN,ij1/2ξT,ij𝒕ijηD𝒗T,ij,\displaystyle=k_{T}\xi_{N,ij}^{1/2}\xi_{T,ij}\bm{t}_{ij}-\eta_{D}\bm{v}_{T,ij}, (8)

where kNk_{N} and kTk_{T} are the stiffness parameters of normal and tangential contacts, respectively. For the normal compression length and its velocity are denoted as ξN,ij:=dij/2|𝒓ij|\xi_{N,ij}:=d_{ij}/2-|\bm{r}_{ij}| and 𝒗N,ij:=(𝒓˙ij𝒏ij)𝒏ij,\bm{v}_{N,ij}:=(\dot{\bm{r}}_{ij}\cdot\bm{n}_{ij})\bm{n}_{ij}, respectively. For the tangential deformation, with the aid of 𝒖ij:=(nijy,nijx)T\bm{u}_{ij}:=(n_{ij}^{y},-n_{ij}^{x})^{\textrm{T}}, the tangential velocity 𝒗T,ij\bm{v}_{T,ij} is defined as 𝒗T,ij:=𝒓˙ij𝒗N,ij+𝒖ij(diωi+djωj)/2\bm{v}_{T,ij}:=\dot{\bm{r}}_{ij}-\bm{v}_{N,ij}+\bm{u}_{ij}(d_{i}\omega_{i}+d_{j}\omega_{j})/2, where we have introduced

𝝃T,ij:=Cij𝑑t𝒗T,ij[(Cij𝑑t𝒗T,ij)𝒏ij]𝒏ij,\displaystyle\bm{\xi}_{T,ij}:=\int_{C_{ij}}dt\bm{v}_{T,ij}-\left[\left(\int_{C_{ij}}dt\bm{v}_{T,ij}\right)\cdot\bm{n}_{ij}\right]\bm{n}_{ij}, (9)

with ξT,ij:=|𝝃T,ij|\xi_{T,ij}:=|\bm{\xi}_{T,ij}| and 𝒕ij:=𝝃T,ij/ξT,ij\bm{t}_{ij}:=-\bm{\xi}_{T,ij}/\xi_{T,ij}. Here, 𝑨˙:=d𝑨/dt\dot{\bm{A}}:=d\bm{A}/dt and Cij𝑑t\int_{C_{ij}}dt is the integration over the duration time of contact between ii and jj particles. Although the dissipative force between grains interacting with the Hertzian force is proportional to the product of the relative velocity and ξN,ij1/2\xi_{N,ij}^{1/2} Kuwabara87 ; Brilliantov96 ; Morgado97 , we adopt simple dissipative forces as in Eqs. (7) and (8) because we are not interested in the relaxation dynamics. We note that Eqs. (7) and (8) assume the Hertzian contact force for the static repulsion of contacting spheres, but all calculations in this study are those for two-dimensional systems. Here, we do not consider the effects of slips in the tangential equation of motion. This treatment can be justified if we restrict our interest in the linear response regime to a stable configuration of particles without any strain. This situation corresponds to frictional grains with infinitely large dynamical friction constant, in which the friction is only characterized by static friction. Therefore, our analysis does not apply to systems with finite strain Otsuki17 , where the effect of slip is important.

To generate a stable configuration of frictional particles, we prepare a stable configuration of frictionless particles in a square box of linear size LL using a fast inertial relaxation engine (FIRE) Bitzek06 . Subsequently, we turn on the tangential force using Eqs. (1) and (2) to achieve a stable configuration in the force-balanced (FB) state for frictional particle111 For simplicity, we prepare the configuration before applying shear for frictionless particles at first, and then considered the friction between particles. If we prepare a configuration before applying shear by compressing frictional particles, we confirm that the configuration had an oscillatory instability that resulted from the appearance of a pair of imaginary eigenvalues of the Jacobian divided by mass matrix: λ=λr±iλi\lambda^{\prime}=\lambda^{\prime}_{r}\pm i\lambda^{\prime}_{i} Chattoraj19 ; Chattoraj19E ; Charan20 , where λ,λr\lambda^{\prime},\lambda^{\prime}_{r}, and λi\lambda^{\prime}_{i} are the complex, real, and imaginary eigenvalues of M1𝒥M^{-1}\mathcal{J}, respectively. Here, 𝒥\mathcal{J} is the Jacobian defined in Eq. (18), and MM is the mass matrix whose explicit form is given by M=[M1MN]M=\left[\begin{matrix}M_{1}&\\ &\ddots\\ &&M_{N}\end{matrix}\right], where Mi:=[mimi4Ii/di2]M_{i}:=\left[\begin{matrix}m_{i}&&\\ &m_{i}&\\ &&4I_{i}/d_{i}^{2}\end{matrix}\right]. Because the linearized equation of motion is expressed as d2qiα/dt2=k,κj,β(Mikακ)1𝒥kjκβqjβd^{2}q_{i}^{\alpha}/dt^{2}=-\sum_{k,\kappa}\sum_{j,\beta}(M_{ik}^{\alpha\kappa})^{-1}\mathcal{J}_{kj}^{\kappa\beta}q_{j}^{\beta}, there are four fundamental solutions 𝒒eiωnt\bm{q}\propto e^{i\omega_{n}^{\prime}t}, where iωni\omega_{n}^{\prime} consists of iω±1=ωi±iωri\omega^{\prime}_{\pm 1}=\omega^{\prime}_{i}\pm i\omega^{\prime}_{r} and iω±2=ωi±iωri\omega^{\prime}_{\pm 2}=-\omega^{\prime}_{i}\pm i\omega^{\prime}_{r}. Here, ωr\omega^{\prime}_{r} and ωi\omega^{\prime}_{i} satisfy the relation ωr±iωi=λr±iλi\omega^{\prime}_{r}\pm i\omega^{\prime}_{i}=\sqrt{\lambda^{\prime}_{r}\pm i\lambda^{\prime}_{i}}. Thus, to avoid the oscillatory instability of the configuration before applying shear, we adopt the protocol of creating the configuration with frictionless particles, and then let the system relax by adding static friction between particles. (see Appendix A for details). Here, the FB state satisfies the FB conditions |Fiζ|=0|F_{i}^{\zeta}|=0 and |Ti|=0|T_{i}|=0 for arbitrary particles. Note that we set θi=0\theta_{i}=0 when the tangential force is turned on.

We impose the Lees-Edwards boundary condition Lees72 ; Evans08 , where the direction parallel to the shear strain is xx-direction. After applying a step strain Δγ\Delta\gamma to all particles, xx-coordinate of the position of the ii-th particle is shifted by an affine displacement Δxi(Δγ):=ΔγyiFB(0)\Delta x_{i}(\Delta\gamma):=\Delta\gamma y_{i}^{\textrm{FB}}(0), where the superscript FB denotes the FB state. The system is then relaxed to the FB state by the contact forces between the particles expressed in Eqs. (7) and (8). Here, ζ\zeta-components of translational Δr̊iζ(Δγ)\Delta\mathring{r}_{i}^{\zeta}(\Delta\gamma) and rotational Δ̊i(Δγ)\Delta\mathring{\ell}_{i}(\Delta\gamma) nonaffine displacements of the ii-th particle after the relaxation process are, respectively, defined as

Δr̊iζ(Δγ):\displaystyle\Delta\mathring{r}_{i}^{\zeta}(\Delta\gamma): =riFB,ζ(Δγ)riFB,ζ(0)δζxΔγyiFB(0),\displaystyle=r_{i}^{\textrm{FB},\zeta}(\Delta\gamma)-r_{i}^{\textrm{FB},\zeta}(0)-\delta_{\zeta x}\Delta\gamma y_{i}^{\textrm{FB}}(0), (10)
Δ̊i:\displaystyle\Delta\mathring{\ell}_{i}: =iFB(Δγ)iFB(0).\displaystyle=\ell_{i}^{\textrm{FB}}(\Delta\gamma)-\ell_{i}^{\textrm{FB}}(0). (11)

Using Eqs. (10) and (11), we introduce the rate of nonaffine displacements as:

dr̊iζdγ:\displaystyle\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}: =limΔγ0Δr̊iζ(Δγ)Δγ\displaystyle=\lim_{\Delta\gamma\to 0}\frac{\Delta\mathring{r}_{i}^{\zeta}(\Delta\gamma)}{\Delta\gamma}
=limΔγ0riFB,ζ(Δγ)riFB,ζ(0)ΔγδζxyiFB(0),\displaystyle=\lim_{\Delta\gamma\to 0}\frac{r_{i}^{\textrm{FB},\zeta}(\Delta\gamma)-r_{i}^{\textrm{FB},\zeta}(0)}{\Delta\gamma}-\delta_{\zeta x}y_{i}^{\textrm{FB}}(0), (12)
d̊idγ:\displaystyle\frac{d\mathring{\ell}_{i}}{d\gamma}: =limΔγ0Δ̊i(Δγ)Δγ\displaystyle=\lim_{\Delta\gamma\to 0}\frac{\Delta\mathring{\ell}_{i}(\Delta\gamma)}{\Delta\gamma}
=limΔγ0iFB(Δγ)iFB(0)Δγ.\displaystyle=\lim_{\Delta\gamma\to 0}\frac{\ell_{i}^{\textrm{FB}}(\Delta\gamma)-\ell_{i}^{\textrm{FB}}(0)}{\Delta\gamma}. (13)

Our system is characterized by the generalized coordinate 𝒒(γ):=(𝒒1T(γ),𝒒2T(γ),,𝒒NT(γ))T\bm{q}(\gamma):=(\bm{q}_{1}^{\textrm{T}}(\gamma),\bm{q}_{2}^{\textrm{T}}(\gamma),\cdots,\bm{q}_{N}^{\textrm{T}}(\gamma))^{\textrm{T}}. The configuration in the FB state at strain γ\gamma is denoted as 𝒒FB(γ):=((𝒒1FB(γ))T,(𝒒2FB(γ))T,,(𝒒NFB(γ))T)T\bm{q}^{\textrm{FB}}(\gamma):=((\bm{q}_{1}^{\textrm{FB}}(\gamma))^{\textrm{T}},(\bm{q}_{2}^{\textrm{FB}}(\gamma))^{\textrm{T}},\cdots,(\bm{q}_{N}^{\textrm{FB}}(\gamma))^{\textrm{T}})^{\textrm{T}}. The shear stress σxy(γ)\sigma_{xy}(\gamma) at 𝒒FB(γ)\bm{q}^{\textrm{FB}}(\gamma) for one sample is given by:

σxy(𝒒FB(γ))=1L2ij>ifijx(𝒒FB(γ))rijy(𝒒FB(γ)).\displaystyle\sigma_{xy}(\bm{q}^{\textrm{FB}}(\gamma))=-\frac{1}{L^{2}}\sum_{i}\sum_{j>i}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(\gamma))r_{ij}^{y}(\bm{q}^{\textrm{FB}}(\gamma)). (14)

The rigidity gg in the linear response regime for one sample is defined as

g\displaystyle g :=dσxy(𝒒(γ))dγ|𝒒(γ)=𝒒FB(0),\displaystyle:=\left.\frac{d\sigma_{xy}(\bm{q}(\gamma))}{d\gamma}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}, (15)

where the differentiation on the right-hand side (RHS) of Eq. (15) is defined as follows:

dσxy(𝒒(γ))dγ|𝒒(γ)=𝒒FB(0)\displaystyle\left.\frac{d\sigma_{xy}(\bm{q}(\gamma))}{d\gamma}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}
:=limΔγ0σxy(𝒒FB(Δγ))σxy(𝒒FB(0))Δγ.\displaystyle\quad\quad\quad:=\lim_{\Delta\gamma\to 0}\frac{\sigma_{xy}(\bm{q}^{\textrm{FB}}(\Delta\gamma))-\sigma_{xy}(\bm{q}^{\textrm{FB}}(0))}{\Delta\gamma}. (16)

In the numerical calculation, we use a non-zero but sufficiently small Δγ\Delta\gamma for the evaluation of gg. Then, the rigidity GG in the linear response regime is defined as

G\displaystyle G :=g,\displaystyle:=\Braket{g}, (17)

where \langle\cdot\rangle is the ensemble average.

For the numerical FB condition, we use the condition |F~iα|<FTh|\tilde{F}_{i}^{\alpha}|<F_{\textrm{Th}} for arbitrary ii, where FThF_{\textrm{Th}} is the threshold force for the simulation and 𝑭~i\tilde{\bm{F}}_{i} is the generalized force, defined as 𝑭~i:=(F~ix,F~iy,F~i)T:=(Fix,Fiy,2Ti/di)T\tilde{\bm{F}}_{i}:=(\tilde{F}_{i}^{x},\tilde{F}_{i}^{y},\tilde{F}_{i}^{\ell})^{\textrm{T}}:=(F_{i}^{x},F_{i}^{y},2T_{i}/d_{i})^{\textrm{T}}.

In our simulation, we adopt ηD=(mkN)1/2d1/4\eta_{D}=(mk_{N})^{1/2}d^{1/4} and FTh=1.0×1014kNd3/2F_{\textrm{Th}}=1.0\times 10^{-14}k_{N}d^{3/2}. The control parameters are the ratio of the tangential to normal stiffness kT/kNk_{T}/k_{N} and projected area fraction to two-dimensional space ϕ\phi. The operating ranges of kT/kNk_{T}/k_{N} and ϕ\phi are 0.00.0 to 10.010.0 and 0.830.83 to 0.900.90, respectively. In this study, we mainly present the results for N=4096N=4096 and Δγ=1.0×106\Delta\gamma=1.0\times 10^{-6} with the ensemble averages of 1010 samples for each kT/kNk_{T}/k_{N} and ϕ\phi. Some results are obtained with N=1024N=1024, Δγ=1.0×106\Delta\gamma=1.0\times 10^{-6}, and five samples for each kT/kNk_{T}/k_{N} and ϕ\phi. We verify that the results are independent of the choice of Δγ\Delta\gamma for 1.0×106Δγ1.0×1041.0\times 10^{-6}\leq\Delta\gamma\leq 1.0\times 10^{-4}. We ignore the effect of dissipation between particles because the velocity of each particle is sufficiently small for infinitesimal agitation from the FB state. The time step of the simulation, Δt\Delta t, was set to Δt=1.0×102t0\Delta t=1.0\times 10^{-2}t_{0}, and numerical integration was performed using the velocity Verlet method (see Appendix B), where t0:=(m/kN)1/2d1/4t_{0}:=(m/k_{N})^{1/2}d^{1/4}. To obtain eigenvalues and eigenvectors of the Jacobian matrix, which will be introduced in detail in the next section, we have used the LAPACK, which provides a template library for linear algebra.

Figure 2 (a) shows an example of the affine displacements of particles, where the displacements exist only in the shear direction, and Fig. 2 (b) shows the nonaffine displacements.

Refer to caption
Figure 2: Plots of (a) affine displacements and (b) nonaffine displacements of particles with Δγ=1.0×106\Delta\gamma=1.0\times 10^{-6}. Here, the magnitudes of the vectors are multiplied by (a) 0.1Δγ10.1\Delta\gamma^{-1} and (b) 1.31.3 for the visualization. These figures are based on numerical results for N=128N=128.

III Theoretical Analysis

In this section, we introduce the Jacobian, the DOS, and theoretical expressions of the linear rigidity. Here, we omit the effects of 𝒒˙\dot{\bm{q}} because the dissipative term proportional to 𝒒˙\dot{\bm{q}} vanishes under quasistatic shear.

III.1 Jacobian and the DOS for frictional particles

In frictional systems, the stability of the system and DOS at 𝒒FB(γ)\bm{q}^{\textrm{FB}}(\gamma) are analyzed using the Jacobian (𝒥\mathcal{J}) defined as Chattoraj19 :

𝒥ijαβ:=F~iα(𝒒(γ))qjβ|𝒒(γ)=𝒒FB(0),\displaystyle\mathcal{J}_{ij}^{\alpha\beta}:=-\left.\frac{\partial\tilde{F}_{i}^{\alpha}(\bm{q}(\gamma))}{\partial q_{j}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}, (18)

where α\alpha and β\beta are any of x,yx,y and \ell, while ii and jj express particle indices. Therefore, the Jacobian matrix, which is a 3N×3N3N\times 3N matrix, can be written as

𝒥=[𝒥11𝒥1i𝒥1j𝒥1N𝒥i1𝒥ii𝒥ij𝒥iN𝒥j1𝒥ji𝒥jj𝒥jN𝒥N1𝒥Ni𝒥Nj𝒥NN],\displaystyle\mathcal{J}=\left[\begin{matrix}\mathcal{J}_{11}&\cdots&\mathcal{J}_{1i}&\cdots&\mathcal{J}_{1j}&\cdots&\mathcal{J}_{1N}\\ \vdots&\ddots&\vdots&&\vdots&&\vdots\\ \mathcal{J}_{i1}&\cdots&\mathcal{J}_{ii}&\cdots&\mathcal{J}_{ij}&\cdots&\mathcal{J}_{iN}\\ \vdots&&\vdots&\ddots&\vdots&&\vdots\\ \mathcal{J}_{j1}&\cdots&\mathcal{J}_{ji}&\cdots&\mathcal{J}_{jj}&\cdots&\mathcal{J}_{jN}\\ \vdots&&\vdots&&\vdots&\ddots&\vdots\\ \mathcal{J}_{N1}&\cdots&\mathcal{J}_{Ni}&\cdots&\mathcal{J}_{Nj}&\cdots&\mathcal{J}_{NN}\end{matrix}\right], (19)

where 𝒥ij\mathcal{J}_{ij} is a 3×33\times 3 submatrix of the Jacobian 𝒥\mathcal{J} for a pair of particles ii and jj:

𝒥ij=[𝒥ijxx𝒥ijxy𝒥ijx𝒥ijyx𝒥ijyy𝒥ijy𝒥ijx𝒥ijy𝒥ij].\displaystyle\mathcal{J}_{ij}=\left[\begin{matrix}\mathcal{J}_{ij}^{xx}&\mathcal{J}_{ij}^{xy}&\mathcal{J}_{ij}^{x\ell}\\ \mathcal{J}_{ij}^{yx}&\mathcal{J}_{ij}^{yy}&\mathcal{J}_{ij}^{y\ell}\\ \mathcal{J}_{ij}^{\ell x}&\mathcal{J}_{ij}^{\ell y}&\mathcal{J}_{ij}^{\ell\ell}\end{matrix}\right]. (20)

See Appendices C and D for detailed properties of the Jacobian. The right and left eigenvalue equations of 𝒥\mathcal{J} are, respectively, given by

𝒥|Rn\displaystyle\mathcal{J}\ket{R_{n}} =λn|Rn,\displaystyle=\lambda_{n}\ket{R_{n}}, (21)
Ln|𝒥\displaystyle\bra{L_{n}}\mathcal{J} =λnLn|,\displaystyle=\lambda_{n}\bra{L_{n}}, (22)

where |Rn\ket{R_{n}} and Ln|\bra{L_{n}} are the right and left eigenvectors corresponding to λn\lambda_{n}, respectively. Here, λn\lambda_{n} is the nn-th eigenvalue of 𝒥\mathcal{J}. Note that |Rn\ket{R_{n}} and Ln|\bra{L_{n}} satisfy the orthonormal relation Lm|Rn=δmn\braket{L_{m}}{R_{n}}=\delta_{mn} with normalization Rn|Rn=Ln|Ln=1\braket{R_{n}}{R_{n}}=\braket{L_{n}}{L_{n}}=1, if all eigenstates are non-degenerate. Here, the inner products for the right and left eigenvectors are defined as Rn|Rn=i=1Nα=x,y,|Rn,iα|2\braket{R_{n}}{R_{n}}=\sum_{i=1}^{N}\sum_{\alpha=x,y,\ell}|R_{n,i}^{\alpha}|^{2} and Ln|Ln=i=1Nα=x,y,|Ln,iα|2\braket{L_{n}}{L_{n}}=\sum_{i=1}^{N}\sum_{\alpha=x,y,\ell}|L_{n,i}^{\alpha}|^{2}, respectively. In the presence of friction, the eigenvalue λn\lambda_{n} is generally a complex number, but if we restrict our interest to infinitesimal distortions from stable configurations without shear strain, λn\lambda_{n} becomes real and can be expressed as λn=ωn2\lambda_{n}=\omega_{n}^{2}. The DOS is the distribution function of the eigenvalues, defined as:

D(ω):=13Nnδ(ωωn),\displaystyle D(\omega):=\frac{1}{3N}\sum_{n}\nolimits^{\prime}\langle\delta(\omega-\omega_{n})\rangle, (23)

where n\sum_{n}\nolimits^{\prime} on RHS of Eq. (23) expresses that the summation excludes the contribution of rattlers (see Appendix E for the details of rattlers). Using the force decomposition, the Jacobian can also be divided into

𝒥ijαβ=𝒥N,ijαβ+𝒥T,ijαβ,\displaystyle\mathcal{J}_{ij}^{\alpha\beta}=\mathcal{J}_{N,ij}^{\alpha\beta}+\mathcal{J}_{T,ij}^{\alpha\beta}, (24)

where

𝒥N,ijαβ:\displaystyle\mathcal{J}_{N,ij}^{\alpha\beta}: =f~N,ijα(𝒒(γ))qjβ|𝒒(γ)=𝒒FB(γ),\displaystyle=\left.\frac{\partial\tilde{f}_{N,ij}^{\alpha}(\bm{q}(\gamma))}{\partial q_{j}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)}, (25)
𝒥T,ijαβ:\displaystyle\mathcal{J}_{T,ij}^{\alpha\beta}: =f~T,ijα(𝒒(γ))qjβ|𝒒(γ)=𝒒FB(γ)\displaystyle=\left.\frac{\partial\tilde{f}_{T,ij}^{\alpha}(\bm{q}(\gamma))}{\partial q_{j}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)} (26)

for iji\neq j and

𝒥N,ijαβ:\displaystyle\mathcal{J}_{N,ij}^{\alpha\beta}: =(i,k)f~N,ikα(𝒒(γ))qiβ|𝒒(γ)=𝒒FB(γ),\displaystyle=\sum_{(i,k)}\left.\frac{\partial\tilde{f}_{N,ik}^{\alpha}(\bm{q}(\gamma))}{\partial q_{i}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)}, (27)
𝒥T,ijαβ:\displaystyle\mathcal{J}_{T,ij}^{\alpha\beta}: =(i,k)f~T,ikα(𝒒(γ))qiβ|𝒒(γ)=𝒒FB(γ)\displaystyle=\sum_{(i,k)}\left.\frac{\partial\tilde{f}_{T,ik}^{\alpha}(\bm{q}(\gamma))}{\partial q_{i}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)} (28)

for i=ji=j. Here, we have introduced 𝒇~N,ij:=(f~N,ijx,f~N,ijy,f~N,ij)T:=(fN,ijx,fN,ijy,0)T\tilde{\bm{f}}_{N,ij}:=(\tilde{f}_{N,ij}^{x},\tilde{f}_{N,ij}^{y},\tilde{f}_{N,ij}^{\ell})^{\textrm{T}}:=(f_{N,ij}^{x},f_{N,ij}^{y},0)^{\textrm{T}} and 𝒇~T,ij:=(f~T,ijx,f~T,ijy,f~T,ij)T:=(fT,ijx,fT,ijy,2Tij/di)T\tilde{\bm{f}}_{T,ij}:=(\tilde{f}_{T,ij}^{x},\tilde{f}_{T,ij}^{y},\tilde{f}_{T,ij}^{\ell})^{\textrm{T}}:=(f_{T,ij}^{x},f_{T,ij}^{y},2T_{ij}/d_{i})^{\textrm{T}}, where fN,ijζf_{N,ij}^{\zeta} and fT,ijζf_{T,ij}^{\zeta} are ζ\zeta-component of 𝒇N,ij\bm{f}_{N,ij} and 𝒇T,ij\bm{f}_{T,ij}, respectively. Note that (i,j)\sum_{(i,j)} denotes the summation for contacted particles of the ii-th particle. The explicit expressions of 𝒥N,ijαβ\mathcal{J}_{N,ij}^{\alpha\beta} and 𝒥T,ijαβ\mathcal{J}_{T,ij}^{\alpha\beta} are presented in Appendix F

III.2 Expressions of the linear rigidity via eigenmodes

Let us introduce |F~(𝒒(γ))\ket{\tilde{F}(\bm{q}(\gamma))} as

|F~(𝒒(γ)):=[𝑭~1T(𝒒(γ)),𝑭~2T(𝒒(γ)),,𝑭~NT(𝒒(γ))]T.\displaystyle\ket{\tilde{F}(\bm{q}(\gamma))}:=[\tilde{\bm{F}}_{1}^{\textrm{T}}(\bm{q}(\gamma)),\tilde{\bm{F}}_{2}^{\textrm{T}}(\bm{q}(\gamma)),\cdots,\tilde{\bm{F}}_{N}^{\textrm{T}}(\bm{q}(\gamma))]^{\textrm{T}}. (29)

Because the forces acting on the particles are balanced in the FB state, |F~(𝒒(γ))|𝒒(γ)=𝒒FB(γ)\ket{\tilde{F}(\bm{q}(\gamma))}|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)} satisfies

|F~(𝒒(γ))|𝒒(γ)=𝒒FB(γ)=|0,\displaystyle\left.\ket{\tilde{F}(\bm{q}(\gamma))}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(\gamma)}=\ket{0}, (30)

where |0\ket{0} is the ket vector containing 0 for all components. The stable configuration in the FB state satisfies

d|F~(𝒒(γ))dγ|𝒒(γ)=𝒒FB(0)=|0,\displaystyle\left.\frac{d\ket{\tilde{F}(\bm{q}(\gamma))}}{d\gamma}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}=\ket{0}, (31)

where

d|F~(𝒒(γ))dγ|𝒒(γ)=𝒒FB(0)\displaystyle\left.\frac{d\ket{\tilde{F}(\bm{q}(\gamma))}}{d\gamma}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}
:=limΔγ0|F~(𝒒FB(Δγ))|F~(𝒒FB(0))Δγ.\displaystyle\quad\quad\quad:=\lim_{\Delta\gamma\to 0}\frac{\Ket{\tilde{F}(\bm{q}^{\textrm{FB}}(\Delta\gamma))}-\Ket{\tilde{F}(\bm{q}^{\textrm{FB}}(0))}}{\Delta\gamma}. (32)

Introducing

|dq̊dγ:=[dr̊1xdγ,dr̊1ydγ,d̊1dγ,,dr̊Nxdγ,dr̊Nydγ,d̊Ndγ]T,\displaystyle\Ket{\frac{d\mathring{q}}{d\gamma}}:=\left[\frac{d\mathring{r}_{1}^{x}}{d\gamma},\frac{d\mathring{r}_{1}^{y}}{d\gamma},\frac{d\mathring{\ell}_{1}}{d\gamma},\cdots,\frac{d\mathring{r}_{N}^{x}}{d\gamma},\frac{d\mathring{r}_{N}^{y}}{d\gamma},\frac{d\mathring{\ell}_{N}}{d\gamma}\right]^{\textrm{T}}, (33)

the left-hand side (LHS) of Eq.(31) can be rewritten as:

d|F~(𝒒(γ))dγ|𝒒(γ)=𝒒FB(0)=|Ξ+𝒥~|dq̊dγ,\displaystyle\left.\frac{d\ket{\tilde{F}(\bm{q}(\gamma))}}{d\gamma}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}=-\ket{\Xi}+\tilde{\mathcal{J}}\Ket{\frac{d\mathring{q}}{d\gamma}}, (34)

where we have used Eqs. (12) and (13) (see Appendix G.1). The first and second terms on RHS of Eq. (34) represent the strain derivatives of the forces for the contributions from the affine and nonaffine displacements, respectively. The explicit form of |Ξ\ket{\Xi} is given by:

|Ξ:=[j1𝒥N,j1xxr1jyj1𝒥N,j1xyr1jyj1𝒥N,j1xr1jyjN𝒥N,jNxxrNjyjN𝒥N,jNxyrNjyjN𝒥N,jNxrNjy].\displaystyle\ket{\Xi}:=\left[\begin{matrix}\sum_{j\neq 1}\mathcal{J}_{N,j1}^{xx}r_{1j}^{y}\\ \sum_{j\neq 1}\mathcal{J}_{N,j1}^{xy}r_{1j}^{y}\\ \sum_{j\neq 1}\mathcal{J}_{N,j1}^{x\ell}r_{1j}^{y}\\ \vdots\\ \sum_{j\neq N}\mathcal{J}_{N,jN}^{xx}r_{Nj}^{y}\\ \sum_{j\neq N}\mathcal{J}_{N,jN}^{xy}r_{Nj}^{y}\\ \sum_{j\neq N}\mathcal{J}_{N,jN}^{x\ell}r_{Nj}^{y}\end{matrix}\right]. (35)

Note that the tangential displacements do not contribute to |Ξ\ket{\Xi}. This is because the affine displacements are applied to our system instantaneously as a step strain; thus, the integral interval of the tangential displacement during the affine deformation is zero. We have used 𝒥~\tilde{\mathcal{J}} in Eq. (34) defined as

𝒥~iiαβ\displaystyle\tilde{\mathcal{J}}_{ii}^{\alpha\beta} :={𝒥iix(α=,β=x)𝒥iiy(α=,β=y)𝒥iiαβ(otherwise)\displaystyle:=\left\{\begin{matrix}-\mathcal{J}_{ii}^{\ell x}&(\alpha=\ell,\ \beta=x)\\ -\mathcal{J}_{ii}^{\ell y}&(\alpha=\ell,\ \beta=y)\\ \mathcal{J}_{ii}^{\alpha\beta}&(\textrm{otherwise})\end{matrix}\right. (36)

and

𝒥~ijαβ\displaystyle\tilde{\mathcal{J}}_{ij}^{\alpha\beta} :={𝒥ijx(α=x,β=)𝒥ijy(α=y,β=)𝒥ijαβ(otherwise)\displaystyle:=\left\{\begin{matrix}-\mathcal{J}_{ij}^{x\ell}&(\alpha=x,\ \beta=\ell)\\ -\mathcal{J}_{ij}^{y\ell}&(\alpha=y,\ \beta=\ell)\\ \mathcal{J}_{ij}^{\alpha\beta}&(\textrm{otherwise})\end{matrix}\right. (37)

for iji\neq j.

Expanding the nonaffine displacements by the eigenfunctions of 𝒥~\tilde{\mathcal{J}} and using the fact that the LHS in Eq. (34) is zero, we obtain

|dq̊dγ=nL~n|Ξλ~n|R~n,\displaystyle\Ket{\frac{d\mathring{q}}{d\gamma}}=\sum_{n}\nolimits^{\prime}\frac{\braket{\tilde{L}_{n}}{\Xi}}{\tilde{\lambda}_{n}}\ket{\tilde{R}_{n}}, (38)

where λ~n,L~n|\tilde{\lambda}_{n},\bra{\tilde{L}_{n}}, and |R~n\ket{\tilde{R}_{n}} are the nn-th eigenvalue of 𝒥~\tilde{\mathcal{J}}, and the left and right eigenvectors corresponding to λ~n\tilde{\lambda}_{n}, respectively. Note that |R~n\ket{\tilde{R}_{n}} and L~n|\bra{\tilde{L}_{n}} satisfy the orthonormal relation L~m|R~n=δmn\braket{\tilde{L}_{m}}{\tilde{R}_{n}}=\delta_{mn}, if all eigenstates are non-degenerate. See Appendix G.1 for the derivation of Eq. (38).

The rigidity in the linear response regime under infinitesimal strain Δγ\Delta\gamma is decomposed into two parts:

G:=GA+GNA,\displaystyle G:=G_{\textrm{A}}+G_{\textrm{NA}}, (39)

where GAG_{\textrm{A}} and GNAG_{\textrm{NA}} are the rigidities corresponding to the affine and nonaffine displacements, respectively. With the aid of Eqs. (14), (17), and (37) the expressions of GAG_{\textrm{A}} and GNAG_{\textrm{NA}} are, respectively, given by (see Appendix G.2)

GA:\displaystyle G_{\textrm{A}}: =12L2i,j(ij)(rijy)2𝒥N,jixx,\displaystyle=\frac{1}{2L^{2}}\Braket{\sum_{i,j(i\neq j)}(r_{ij}^{y})^{2}\mathcal{J}_{N,ji}^{xx}}, (40)
GNA:\displaystyle G_{\textrm{NA}}: =12L2i,j(ij)[ζ=x,yrijy𝒥~ijxζdr̊ijζdγrijy𝒥~ijxd̊ijdγ],\displaystyle=\frac{1}{2L^{2}}\Braket{\sum_{i,j(i\neq j)}\Biggl{[}\sum_{\zeta=x,y}r_{ij}^{y}\tilde{\mathcal{J}}_{ij}^{x\zeta}\frac{d\mathring{r}_{ij}^{\zeta}}{d\gamma}-r_{ij}^{y}\tilde{\mathcal{J}}_{ij}^{x\ell}\frac{d\mathring{\ell}_{ij}}{d\gamma}\Biggl{]}}, (41)

where we have introduced

dr̊ijζdγ:\displaystyle\frac{d\mathring{r}_{ij}^{\zeta}}{d\gamma}: =dr̊iζdγdr̊jζdγ,\displaystyle=\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}, (42)
d̊ijdγ:\displaystyle\frac{d\mathring{\ell}_{ij}}{d\gamma}: =d̊idγ+d̊jdγ.\displaystyle=\frac{d\mathring{\ell}_{i}}{d\gamma}+\frac{d\mathring{\ell}_{j}}{d\gamma}. (43)

Substituting Eq. (38) into Eq. (41), GNAG_{\textrm{NA}} can be rewritten as follows:

GNA\displaystyle G_{\textrm{NA}} =1L2nL~n|ΞΘ|R~nλ~n,\displaystyle=-\frac{1}{L^{2}}\left\langle\sum_{n}\nolimits^{\prime}\frac{\braket{\tilde{L}_{n}}{\Xi}\braket{\Theta}{\tilde{R}_{n}}}{\tilde{\lambda}_{n}}\right\rangle, (44)

where we have introduced

Θ|:=[j1r1jy𝒥~j1xx,j1r1jy𝒥~j1xy,j1r1jy𝒥~j1x,,jNrNjy𝒥~jNxx,jNrNjy𝒥~jNxy,jNrNjy𝒥~jNx].\displaystyle\bra{\Theta}:=\left[\begin{matrix}\sum_{j\neq 1}r_{1j}^{y}\tilde{\mathcal{J}}_{j1}^{xx},\sum_{j\neq 1}r_{1j}^{y}\tilde{\mathcal{J}}_{j1}^{xy},\sum_{j\neq 1}r_{1j}^{y}\tilde{\mathcal{J}}_{j1}^{x\ell},\cdots,\sum_{j\neq N}r_{Nj}^{y}\tilde{\mathcal{J}}_{jN}^{xx},\sum_{j\neq N}r_{Nj}^{y}\tilde{\mathcal{J}}_{jN}^{xy},\sum_{j\neq N}r_{Nj}^{y}\tilde{\mathcal{J}}_{jN}^{x\ell}\end{matrix}\right]. (45)

The affine rigidity can be also expressed as

GA=1L2Y|Ξ,\displaystyle G_{\textrm{A}}=\frac{1}{L^{2}}\left\langle\braket{Y}{\Xi}\right\rangle, (46)

where

Y|:=[r1jy,0,0,r2jy,0,0,,rNjy,0,0].\displaystyle\bra{Y}:=\left[r_{1j}^{y},0,0,r_{2j}^{y},0,0,\cdots,r_{Nj}^{y},0,0\right]. (47)

IV Results

In this section, we present the results of eigenvalue analysis and rigidity based on the formulation explained in the previous section. In Sec. IV.1, rigidity is determined using the eigenmodes of the Jacobian. Section IV.2 clarifies the effects of translational and rotational motions on the DOS.

IV.1 Theoretical evaluation of GG

In this subsection, the validity of the theoretical rigidity presented in the previous section is demonstrated. For this purpose, at first, we examine the validity of Eq. (38), obtained by the eigenfunction expansion of the nonaffine displacements for RHS and by the simulation for LHS. Figures 3 (a) and (b) illustrate the nonaffine displacements on LHS and RHS of Eq. (38), respectively. In Figs. 3 (a) and (b), (x,y)(x,y) and \ell-components of d𝒒̊i/dγd\mathring{\bm{q}}_{i}/d\gamma at 𝒓i\bm{r}_{i} are represented by vectors and colors, respectively. Figure 3 (c) shows the RHS and LHS of Eq. (38) against the components of the vectors whose orders follow Eq. (33), that is, the local order of the component follows x,y,x,y, and \ell by fixing the particle number, and we align the components from the first particle to the NN-th particle without omitting modes with extremely small and zero eigenvalues. Figure 3 shows that the expression in Eq. (38) correctly reproduces the simulation results.

Refer to caption
Refer to caption
Refer to caption
Figure 3: Plots of nonaffine displacements on (a) RHS of Eq. (38) obtained by the eigenvalue analysis and (b) LHS of Eq. (38). The vector and color of each particle correspond to x,y,x,y, and \ell-components of the eigenvector of the particle, respectively. Here, the magnitude of the vectors is magnified by 1.31.3 times for visualization. (c) Plots of the RHS (open symbols) and LHS (filled symbols) obtained by the simulation of Eq. (38) for each component whose order follows Eq. (33). These figures are based on numerical results for N=1024N=1024.

The dimensionless rigidity obtained from Eqs. (39),(40) and (44) with the aid of G:=kNd1/2G^{*}:=k_{N}d^{1/2} is shown in Fig. 4. This indicates the quantitative agreement between the theoretical and numerical values.

Refer to caption
Figure 4: Plots of theoretical (Eq. (39): open symbols) and numerical (Eq. (16): filled symbols) GG against ϕ\phi for various kT/kNk_{T}/k_{N}. The figure obtained by the numerical results for N=1024N=1024.

Therefore, the rigidity in the linear response regime can be determined completely using the Jacobian analysis. On the contrary to the previous studies Hern03 ; Olsson11 ; Otsuki11 , we should note that GG is not proportional to ϕϕc\phi-\phi_{c} for a large kT/kNk_{T}/k_{N}, where ϕc\phi_{c} is the critical fraction of jamming transition for frictional grains.

Refer to caption
Figure 5: Plots of numerical GG against kT/kNk_{T}/k_{N} for various ϕ\phi. The figure is obtained by the numerical results for N=4096N=4096.

As is expected, the rigidity GG depends on kT/kNk_{T}/k_{N} only little for kT/kN1.0×104k_{T}/k_{N}\leq 1.0\times 10^{-4} (see Fig. 5), while GG depends on kT/kNk_{T}/k_{N} for kT/kN>104k_{T}/k_{N}>10^{-4}. We have confirmed that GG smoothly approaches the frictionless value in the limit kT0k_{T}\to 0 in contrary to Refs. Otsuki17 ; Ishima20 . Here, GG cannot be expressed as a factorization for large kT/kNk_{T}/k_{N} 222 We have confirmed that the factorization G(ϕ,kT/kN)=G0(ϕ)𝒢(kT/kN)G(\phi,k_{T}/k_{N})=G_{0}(\phi)\mathcal{G}(k_{T}/k_{N}) is not held, where G0(ϕ)G_{0}(\phi) is the rigidity of frictionless system. . When we consider the effect of the dynamical friction, that is, slips between particles, the rigidity is discontinuously changed in the frictionless limit Otsuki17 ; Ishima20 . However, the rigidity continuously changes with kT/kNk_{T}/k_{N} in our system and is smoothly connected to that of frictionless systems (kT=0k_{T}=0). Because our system can be regarded as having infinitely large static and dynamical frictional constants, there is no slip between the grains. Therefore, it might be natural for GG to continuously change the limit for kT/kN0k_{T}/k_{N}\to 0 in our system. In future work, we will consider the effects of slips, which are important for real frictional grains.

To clarify the contributions of nonaffine deformations to the rigidity, we plot GNAG_{\textrm{NA}} defined in Eq. (41) against ϕ\phi in Fig. 6, in which GNAG_{\textrm{NA}} becomes large as ϕ\phi increases. Remarkably, GNAG_{\textrm{NA}} is positive for kT/kN>2.0k_{T}/k_{N}>2.0, GNA0G_{\textrm{NA}}\approx 0 at kT/kN=2.0k_{T}/k_{N}=2.0, and GNAG_{\textrm{NA}} is negative for kT/kN<2.0k_{T}/k_{N}<2.0. The positive GNAG_{\textrm{NA}} for a large kT/kNk_{T}/k_{N} is counterintuitive, in which GG increases from GAG_{\textrm{A}} even when the system is relaxed to the FB state. In the future, we must clarify the origin of this counterintuitive GNAG_{\textrm{NA}}. We note that the negative GNAG_{\textrm{NA}} for a small kT/kNk_{T}/k_{N} can be understood by the relaxation process to look for a FB configuration after applying affine deformations to the system.

Refer to caption
Figure 6: Plots of GNAG_{\textrm{NA}} against ϕ\phi for various kT/kNk_{T}/k_{N}. The figure is obtained by the numerical results for N=1024N=1024.

IV.2 Analysis of eigenvalues and eigenvectors

In Fig. 7 we present some typical right eigenvectors |Rn\ket{R_{n}}, which was introduced in Eq. (21), and can be expressed as:

|Rn=[𝑹n,1𝑹n,2𝑹n,N],\displaystyle\ket{R_{n}}=\left[\begin{matrix}\bm{R}_{n,1}\\ \bm{R}_{n,2}\\ \vdots\\ \bm{R}_{n,N}\end{matrix}\right], (48)

where 𝑹n,i:=(Rn,ix,Rn,iy,Rn,i)T\bm{R}_{n,i}:=(R_{n,i}^{x},R_{n,i}^{y},R_{n,i}^{\ell})^{\textrm{T}}. Figure 7 illustrates vectors (Rn,ix,Rn,iy)T(R_{n,i}^{x},R_{n,i}^{y})^{\textrm{T}} and colors to characterize the rotation Rn,iR_{n,i}^{\ell} of particle ii for some characteristic ωn\omega_{n} with kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8} (Figs. 7 (a1)-(a3)) and kT/kN=1.0×104k_{T}/k_{N}=1.0\times 10^{-4} (Figs. 7 (b1)-(b3)). Figures 7 (a1) and (b1) show the eigenvectors at ωnt0=1.0×102\omega_{n}t_{0}=1.0\times 10^{-2} and ωnt0=1.0×104\omega_{n}t_{0}=1.0\times 10^{-4}, respectively, which are dominated by the rotational modes. In Fig. 7 (a2), we confirm that the eigenvector at ωnt0=1.3×102\omega_{n}t_{0}=1.3\times 10^{-2} is expressed only by translational modes, whereas the eigenvector at ωnt0=1.3×102\omega_{n}t_{0}=1.3\times 10^{-2} shown in Fig. 7 (b2) is expressed as a coupling mode of the rotational and translational modes. In Figs. 7 (a3) and (b3), we show the eigenvectors at ωnt0=1.0\omega_{n}t_{0}=1.0 which are dominated by the translational modes.

Refer to caption
Figure 7: Plots of eigenvectors for ϕ=0.90\phi=0.90 with (a1) ωnt0=1.0×104\omega_{n}t_{0}=1.0\times 10^{-4}, (a2) 1.3×1021.3\times 10^{-2}, (a3) 1.01.0, (b1) 1.0×1021.0\times 10^{-2}, (b2) 1.3×1021.3\times 10^{-2}, and (b3) 1.01.0, where (a1)-(a3), and (b1)-(b3) are the results for kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8} and 1.0×1041.0\times 10^{-4}, respectively. Here, (Rn,ix,Rn,iy)T(R_{n,i}^{x},R_{n,i}^{y})^{\textrm{T}} and Rn,iR_{n,i}^{\ell} are represented by vectors and colors for the ii-th particle, respectively. Note that the magnitudes of the vectors are magnified by 5050 times for visualization. These figures are based on numerical results for N=4096N=4096.

To clarify the translational and rotational contributions at each eigenvalue, we compute the translational and rotational participation fractions Yunker11 ; Shiraishi19 defined as

ψnT\displaystyle\psi_{n}^{T} :=i=1N[|Rn,ix|2+|Rn,iy|2],\displaystyle:=\sum_{i=1}^{N}\left[|R_{n,i}^{x}|^{2}+|R_{n,i}^{y}|^{2}\right], (49)
ψnR\displaystyle\psi_{n}^{R} :=i=1N|Rn,i|2=1ψnT,\displaystyle:=\sum_{i=1}^{N}|R_{n,i}^{\ell}|^{2}=1-\psi_{n}^{T}, (50)

respectively, where we have investigated the localization of the system with the participation ratio in Appendix E. Note that translational mode is dominant when ψnT\psi_{n}^{T} is close to 11 and rotational mode is dominant when ψnR\psi_{n}^{R} is close to 11. ψT(ω)\psi^{T}(\omega) and ψR(ω)\psi^{R}(\omega) are plotted for various kT/kNk_{T}/k_{N} in Fig. 8, where

ψT(ω):\displaystyle\psi^{T}(\omega): =nψnTδ(ωωn)nδ(ωωn),\displaystyle=\frac{\sum_{n}\nolimits^{\prime}\langle\psi_{n}^{T}\delta(\omega-\omega_{n})\rangle}{\sum_{n}\nolimits^{\prime}\langle\delta(\omega-\omega_{n})\rangle}, (51)
ψR(ω):\displaystyle\psi^{R}(\omega): =nψnRδ(ωωn)nδ(ωωn).\displaystyle=\frac{\sum_{n}\nolimits^{\prime}\langle\psi_{n}^{R}\delta(\omega-\omega_{n})\rangle}{\sum_{n}\nolimits^{\prime}\langle\delta(\omega-\omega_{n})\rangle}. (52)

Here, ψT\psi^{T} and ψR\psi^{R} are set to zero if there is no right eigenvalue for ω(s)<ωt0<ω(s+1)\omega^{(s)}<\omega t_{0}<\omega^{(s+1)} with the ss-th data point ω(s)\omega^{(s)}. Here, we have used the following steps to determine each data point. First, we divided the data interval into 5050 parts on a logarithmic scale. Then we linearly re-divided the data interval from the highest frequency to the 10th highest frequency region. Finally, we also linearly re-divided the data interval of the log scale corresponding to 0.1kT/kN<ωt0<2kT/kN0.1\sqrt{k_{T}/k_{N}}<\omega t_{0}<2\sqrt{k_{T}/k_{N}}. Note that for the linear re-division of the data, the regions were divided into 500500 or 100100 equally spaced inter-regional intervals for high frequency or 0.1kT/kN<ωt0<2kT/kN0.1\sqrt{k_{T}/k_{N}}<\omega t_{0}<2\sqrt{k_{T}/k_{N}}, respectively.

Refer to caption
Figure 8: Semi-logarithmic plots of ψT\psi^{T} (red filled squares) and ψR\psi^{R} (blue filled circles) against ωt0\omega t_{0} for ϕ=0.90\phi=0.90 at (a) kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8}, (b) 1.0×1041.0\times 10^{-4}, (c) 1.01.0, and (d) 10.010.0. The eigenvalues of (a1), (a2), (a3), (b1), (b2), and (b3) in these figures correspond to Figs. 7 (a1), (a2), (a3), (b1), (b2), and (b3), respectively. These figures are based on numerical results for N=4096N=4096.

As shown in Figs. 8 (a) and (b), we find the region of ψR1\psi^{R}\simeq 1 for low ω\omega and kT/kN<1.0×104k_{T}/k_{N}<1.0\times 10^{-4}. This region is referred to as Region I. We also find a region that satisfies ψT1\psi^{T}\simeq 1 for high ω\omega and kT/kN<1.0k_{T}/k_{N}<1.0, in which the translational modes are dominant. This region is referred to as Region II. Here, three characteristic behaviors depend on kT/kNk_{T}/k_{N} at ϕ=0.90\phi=0.90. First, the translational modes are separable from the rotational modes for kT/kN1.0×108k_{T}/k_{N}\leq 1.0\times 10^{-8} because we need a small amount of energy to excite the rotational mode in nearly frictionless situations, as shown in Figs. 8 (a) and (b). Second, the translational and rotational contributions are not separated for 1.0×106kT/kN1.0×1021.0\times 10^{-6}\leq k_{T}/k_{N}\leq 1.0\times 10^{-2}. Third, the translational and rotational contributions are indistinguishable for kT/kN1.0k_{T}/k_{N}\geq 1.0.

The DOS obtained from the Jacobian eigenvalues is shown in Fig. 9. Based on the results of ψT(ω)\psi^{T}(\omega) and ψR(ω)\psi^{R}(\omega), the DOS is also separated into two regions for kT/kN1.0×108k_{T}/k_{N}\leq 1.0\times 10^{-8}. The rotational band for low ω\omega shifts to the high ω\omega region as kT/kNk_{T}/k_{N} increases (see Figs. 9 (a) and (b)). In Region II with a high ω\omega (see Figs. 9 (a) and (b)), the DOS is almost independent of kT/kNk_{T}/k_{N} in which the translational modes are dominant for kT/kN1.0×102k_{T}/k_{N}\leq 1.0\times 10^{-2}. The distinctions between the two regions for the DOS are visible with a distinct gap between the adjacent regions for 1.0×1010kT/kN1.0×1081.0\times 10^{-10}\leq k_{T}/k_{N}\leq 1.0\times 10^{-8}. For 1.0×106kT/kN1.0×1021.0\times 10^{-6}\leq k_{T}/k_{N}\leq 1.0\times 10^{-2}; however, the high ω\omega region of the DOS in Region I partially overlaps with the low ω\omega region of Region II. Furthermore, Regions I and II are completely merged for kT/kN1.0k_{T}/k_{N}\geq 1.0 (see Figs. 9 (c) and (d)). Isolated DOS bands for low ω\omega have been observed in systems containing anisotropic grains, such as elliptical grains and dimers Yunker11 ; Schreck12 ; Shiraishi19 . However, to the best of our knowledge, there is no paper pointing out the existence of isolated bands of DOS in systems of frictional grains.

Refer to caption
Figure 9: Double logarithmic plots of D(ω)D(\omega) against ωt0\omega t_{0} for ϕ=0.90\phi=0.90 at (a) kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8}, (b) 1.0×1041.0\times 10^{-4}, (c) 1.01.0, and (d) 10.010.0. The eigenvalues of (a1), (a2), (a3), (b1), (b2) and (b3) in these figures correspond to Figs. 7 (a1), (a2), (a3), (b1), (b2) and (b3), respectively. These figures are based on numerical results for N=4096N=4096.

Because we have confirmed the existence of a peak of D(ω)D(\omega) around ωt0(kT/kN)1/2\omega t_{0}\simeq(k_{T}/k_{N})^{1/2}, Fig. 10 shows the scaling of the DOS in Region I by plotting ωRD(ω/ωR)\omega_{R}D(\omega/\omega_{R}), where ωR:=kT/kNt01\omega_{R}:=\sqrt{k_{T}/k_{N}}t_{0}^{-1} 333 The reason why D(ω)D(\omega) is multiplied by ωR\omega_{R} in Fig. 10 is as follows. The integral value of the DOS within Region I I𝑑ωD(ω)\int_{\textrm{I}}d\omega D(\omega) is almost independent of kT/kNk_{T}/k_{N}. Then, LHS can be rewritten as a variable I𝑑ωD(ω)=I𝑑ω^D(ω^)\int_{\textrm{I}}d\omega D(\omega)=\int_{\textrm{I}}d\hat{\omega}D^{*}(\hat{\omega}), where ω^:=ω/ωR\hat{\omega}:=\omega/\omega_{R}, D(ω^):=ωRD(ω/ωR)D^{*}(\hat{\omega}):=\omega_{R}D(\omega/\omega_{R}) and I\int_{\textrm{I}} represents the integral in Region I. . From Fig. 10 we have confirmed that ωRD(ω)\omega_{R}D(\omega) can be expressed as a universal scaling function of ω/ωR\omega/\omega_{R} for 0.1<ω/ωR<10.1<\omega/\omega_{R}<1 and kT/kN0.01k_{T}/k_{N}\leq 0.01.

Refer to caption
Figure 10: Scaling plots of ωRD(ω)\omega_{R}D(\omega) versus ω/ωR\omega/\omega_{R} for ϕ=0.90\phi=0.90 and various kT/kNk_{T}/k_{N} in 0.1<ω/ωR<10.00.1<\omega/\omega_{R}<10.0. The figure is based on numerical results for N=4096N=4096.

To clarify the behavior of the DOS in the frictionless limit, we compare the DOS for kT/kN=108k_{T}/k_{N}=10^{-8} with that for frictionless particles by plotting both cases, where we adopt the Hessian matrix to calculate the DOS for frictionless systems (see Fig. 11 and Appendix H). As expected, there is no singularity of the DOS for the translational mode, while the isolated rotational band in low ω\omega is absent in frictionless particles, as shown in Fig. 11.

Refer to caption
Figure 11: Double logarithmic plots of D(ω)/t0D(\omega)/t_{0} versus ωt0\omega t_{0} with ϕ=0.90\phi=0.90, where red circles are the DOS for frictional grains, while blue triangles are the DOS for frictionless grains. The figure is based on numerical results for N=4096N=4096.

At the end of this subsection, we examine the usefulness of the effective Hessian \mathcal{H} introduced in Refs. Somfai07 ; Henkes10 ; Liu21 by comparing D(ω)D(\omega) with DH(ω)D_{H}(\omega), where DH(ω)D_{H}(\omega) is the DOS obtained from \mathcal{H} (see Appendix H). As shown in Figs. 12 (a) and (b), D(ω)D(\omega) and DH(ω)D_{H}(\omega) for various kT/kNk_{T}/k_{N} are almost identical. Here, the peak of the DOS near ω=0\omega=0 is caused by the rotational motion of the grains. This agreement between the Jacobian and Hessian analyses is natural because the configuration before the application of shear was prepared with frictionless particles, and the tangential displacement ξT,ij\xi_{T,ij} is sufficiently small.

Refer to caption
Figure 12: Plots of D(ω)D(\omega) (filled symbols) and DH(ω)D_{H}(\omega) (open symbols) for ϕ=0.90\phi=0.90 at (a) kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8}, (b) 1.0×1041.0\times 10^{-4}, (c) 1.01.0, and (d) 10.0. These figures are based on numerical results for N=4096N=4096.

V Concluding Remarks

We analyzed the eigenmodes of the Jacobian and obtained an expression for the rigidity of amorphous solids of frictional particles under an infinitesimal strain. We reproduced the rigidity in the linear response regime using the eigenvalues and eigenfunctions of the Jacobian with modifications in the rotational part.

Further, we confirmed that the DOS can be divided into two regions for small kT/kNk_{T}/k_{N}. In the low-frequency region (Region I), the rotation of the particles plays a dominant role. These modes are characterized by the frequency (kT/kN)1/2/t0(k_{T}/k_{N})^{1/2}/t_{0}. Region I merges into the high-frequency region (Region II) for large kT/kNk_{T}/k_{N}, where Region II is dominated by translational modes.

It should be noted that our results are almost independent of system size, as shown in Appendix I. Moreover, we have briefly analyzed the density dependence of the DOS in Appendix J, where the rotational band shifts to a lower frequency region and the plateau of the translational band become longer as the density approaches the jamming point.

We have also confirmed that the results of our Jacobian analysis are almost equivalent to those of the Hessian matrix. This is because our preparation of the initial configuration of grains is made by frictionless grains. Nevertheless, we expect that the Hessian analysis might be sufficient for stable configurations of grains.

However, the applicability of this theory is limited. The method used in this study cannot be used for finite strains because it is obvious that the eigenvectors are not orthogonal in the sheared state. Moreover, there are plastic deformations of the grains under large strains, which were not considered in this study. Therefore, we cannot predict the correct value of the theoretical rigidity at the stress drop point. More importantly, the effect of the history dependence of the frictional force is significant even in the linear response regime, although we have ignored such contacts because of the difficulty in constructing a proper theory. This issue should be addressed in future studies Ishima22 .

VI ACKNOWLEDGMENTS

The authors thank N. Oyama and T. Kawasaki for fruitful discussions and useful comments. This work was partially supported by a Grant-in-Aid from the MEXT for Scientific Research (Grant No. JP22K03459, JP21H01006, and JP19K03670) and by the Grant-in-Aid from the Japan Society for Promotion of Science JSPS Research Fellow (Grant No. JP20J20292).

Appendix A Method of preparing configuration before applying the shear

In this Appendix, we summarize the method of preparing a stable configuration of grains before applying the shear. For this purpose, as the first step, we perform the relaxation for frictionless particles with the FIRE. As the second step, the system is relaxed taking into account the static friction between particles. In the first subsection, we summarize how to prepare the configurations of frictionless particles by the FIRE Bitzek06 . In the second subsection, we describe the details of the numerical method including the force with static friction.

A.1 Method of preparing configuration before applying shear by FIRE

At first, we place particles at random without any overlaps of particles with the initial fraction ϕini=0.6\phi_{\textrm{ini}}=0.6. We increase the projected area fraction of the system by the increment of the fraction Φ:=ϕNewϕOld\Phi:=\phi^{\textrm{New}}-\phi^{\textrm{Old}} up to the target fraction ϕ\phi, where ϕOld\phi^{\textrm{Old}} and ϕNew\phi^{\textrm{New}} are the projected area fraction of the system before and after each step of the increment, respectively. After each step of the increment, the system is relaxed by the FIRE Bitzek06 .

To implement the process of increasing the area fraction, we scale the system as

LNew\displaystyle L^{\textrm{New}} =LOldϕOldϕNew,\displaystyle=L^{\textrm{Old}}\sqrt{\frac{\phi^{\textrm{Old}}}{\phi^{\textrm{New}}}}, (1)
𝒓iNew\displaystyle\bm{r}_{i}^{\textrm{New}} =𝒓iOldϕOldϕNew,\displaystyle=\bm{r}_{i}^{\textrm{Old}}\sqrt{\frac{\phi^{\textrm{Old}}}{\phi^{\textrm{New}}}}, (2)

where LOld/LNewL^{\textrm{Old}}/L^{\textrm{New}} and 𝒓iOld/𝒓iNew\bm{r}^{\textrm{Old}}_{i}/\bm{r}^{\textrm{New}}_{i}are the linear system size and the position of the ii-th particle before/after rescaling, respectively. We adopt Φ=104\Phi=10^{-4}. When there are overlaps between particles at ϕNew\phi^{\textrm{New}}, the system relaxes to a stable configuration with the aid of the FIRE.

The FIRE is a fast relaxation method of minimizing potentials U(𝒓)U(\bm{r}) depending on the configuration of the particles 𝒓:=(𝒓1T,𝒓2T,,𝒓NT)T\bm{r}:=(\bm{r}_{1}^{\textrm{T}},\bm{r}_{2}^{\textrm{T}},\cdots,\bm{r}_{N}^{\textrm{T}})^{\textrm{T}} with 𝒓i:=(rix,riy)T:=(xi,yi)T\bm{r}_{i}:=(r_{i}^{x},r_{i}^{y})^{\textrm{T}}:=(x_{i},y_{i})^{\textrm{T}} Bitzek06 . Here, we use the Hertzian potential for U(𝒓)U(\bm{r}) which is defined as

U(𝒓):=25kNjiξN,ij5/2Θ(dij/2|𝒓ij|).\displaystyle U(\bm{r}):=\frac{2}{5}k_{N}\sum_{j\neq i}\xi_{N,ij}^{5/2}\Theta(d_{ij}/2-|\bm{r}_{ij}|). (3)

Let us introduce ζ\zeta-component of the force FF,iζF_{F,i}^{\zeta} acting on the ii-th particle as

FF,iζ:\displaystyle F_{F,i}^{\zeta}: =Uriζ=jikNξN,ij3/2nijζΘ(dij/2|𝒓ij|),\displaystyle=-\frac{\partial U}{\partial r_{i}^{\zeta}}=\sum_{j\neq i}k_{N}\xi_{N,ij}^{3/2}n_{ij}^{\zeta}\Theta(d_{ij}/2-|\bm{r}_{ij}|), (4)

where ζ=x\zeta=x or yy. Note that FF,iζF_{F,i}^{\zeta} only consists of the normal repulsive force. In the FIRE the position 𝒓\bm{r} and velocity 𝒗F:=(𝒗F,1,𝒗F,2,,𝒗F,N)T\bm{v}_{F}:=(\bm{v}_{F,1},\bm{v}_{F,2},\cdots,\bm{v}_{F,N})^{\textrm{T}} with 𝒗F,i:=(vF,ix,vF,iy)T\bm{v}_{F,i}:=(v_{F,i}^{x},v_{F,i}^{y})^{\textrm{T}} are updated by the following rules from (i) to (iv) with the variable time increment ΔtF\Delta t_{F}. (i) The numerical integration via the velocity Verlet method is performed on 𝒓\bm{r} and 𝒗F\bm{v}_{F}:

riζ\displaystyle r_{i}^{\zeta} riζ+ΔtFviζ+ΔtF2FF,iζ(𝒓)2mi,\displaystyle\to r_{i}^{\zeta}+\Delta t_{F}v_{i}^{\zeta}+\Delta t_{F}^{2}\frac{F_{F,i}^{\zeta}(\bm{r})}{2m_{i}}, (5)
vF,iζ\displaystyle v_{F,i}^{\zeta} vF,iζ+ΔtFFF,iζ(𝒓~)+FF,iζ(𝒓)2mi,\displaystyle\to v_{F,i}^{\zeta}+\Delta t_{F}\frac{F_{F,i}^{\zeta}(\tilde{\bm{r}})+F_{F,i}^{\zeta}(\bm{r})}{2m_{i}}, (6)

where 𝒓~\tilde{\bm{r}} is the updated configuration in Eq. (5). (ii) We calculate P:=𝑭F𝒗FP:=\bm{F}_{F}\cdot\bm{v}_{F}, where 𝑭F:=(𝑭F,1T,𝑭F,2T,,𝑭F,NT)T\bm{F}_{F}:=(\bm{F}_{F,1}^{\textrm{T}},\bm{F}_{F,2}^{\textrm{T}},\cdots,\bm{F}_{F,N}^{\textrm{T}})^{\textrm{T}} with 𝑭F,i:=(FF,ix,FF,iy)T\bm{F}_{F,i}:=(F_{F,i}^{x},F_{F,i}^{y})^{\textrm{T}}. (iii) The velocity 𝒗F\bm{v}_{F} is updated as

𝒗F,i𝒗F,i+χ(𝒗^F,i𝑭^F,i)|𝒗F,i|,\displaystyle\bm{v}_{F,i}\to\bm{v}_{F,i}+\chi(\hat{\bm{v}}_{F,i}-\hat{\bm{F}}_{F,i})|\bm{v}_{F,i}|, (7)

where χ\chi is the relaxation parameter and 𝒂^:=𝒂/|𝒂|\hat{\bm{a}}:=\bm{a}/|\bm{a}| for an arbitrary vector 𝒂\bm{a}. (iv) We update χ\chi and ΔtF\Delta t_{F} in the FIRE according to the positive or negative value of PP. To speed up the relaxation when the motion is along a potential gradient, we increase ΔtF\Delta t_{F}. Note that this process is performed only when the number of numerical integrations along the potential gradient is larger than a certain number of times NminN_{\min} to stabilize the numerical calculation. To implement this update rule, if P>0P>0 and the number of numerical integrations of P>0P>0 is larger than NminN_{\min}, ΔtF\Delta t_{F} and χ\chi are updated as

ΔtF\displaystyle\Delta t_{F} min(ΔtFfinc,ΔtF,max),\displaystyle\to\min(\Delta t_{F}f_{\textrm{inc}},\Delta t_{F,\max}), (8)
χ\displaystyle\chi χfχ,\displaystyle\to\chi f_{\chi}, (9)

where min(a,b)\min(a,b) is a selecting function of smaller one from aa and bb, the parameter fincf_{\textrm{inc}} is introduced to speed up the relaxation, and fχf_{\chi}, and ΔtF,max\Delta t_{F,\max} are parameters to stabilize numerical calculations. Here, we adopt finc=1.1,fχ=0.99,Nmin=5,ΔtF,max=10ΔtF,inif_{\textrm{inc}}=1.1,f_{\chi}=0.99,N_{\min}=5,\Delta t_{F,\max}=10\Delta t_{F,\textrm{ini}}, and ΔtF,ini=1.0×102t0\Delta t_{F,\textrm{ini}}=1.0\times 10^{-2}t_{0} Bitzek06 ; Saitoh19 . Note that NminN_{\min} is necessary for the stability of the algorithm. In the case of P0P\leq 0, we set

𝒗F\displaystyle\bm{v}_{F} 𝟎,\displaystyle\to\bm{0}, (10)
χ\displaystyle\chi χstart,\displaystyle\to\chi_{\textrm{start}}, (11)
ΔtF\displaystyle\Delta t_{F} ΔtFfdec,\displaystyle\to\Delta t_{F}f_{\textrm{dec}}, (12)

where we adopt fdec=0.5f_{\textrm{dec}}=0.5 and χstart=0.1\chi_{\textrm{start}}=0.1 Bitzek06 ; Saitoh19 .

We repeat the operations (i) through (iv) until |FF,iζ|<FTh|F_{F,i}^{\zeta}|<F_{\textrm{Th}} for arbitrary ii and ζ\zeta. Note that we have used the initial values for ΔtF=ΔtF,ini\Delta t_{F}=\Delta t_{F,\textrm{ini}} and χ=χstart\chi=\chi_{\textrm{start}} at the starting point of the FIRE. Here, 𝒓\bm{r} is given and we set 𝒗F=𝟎\bm{v}_{F}=\bm{0} at the starting point of the FIRE.

A.2 Numerical method for relaxation of configuration of frictional particles

After we obtain a stable configuration of frictionless particles at a target fraction in terms of the FIRE, we consider the effect of static friction in the relaxation process of frictional particles. The time evolution of the system is given by Eqs. (1)-(8) until F~iα<FTh\tilde{F}_{i}^{\alpha}<F_{\textrm{Th}} for arbitrary ii and α\alpha. For the time integration, we adopt the velocity Verlet method with the time increment Δt=1.0×102t0\Delta t=1.0\times 10^{-2}t_{0}.

Appendix B Velocity Verlet method in our system

In this Appendix, we first verify the accuracy of the velocity Verlet method. Next, we summarize the implementation of the velocity Verlet method. To simplify the notation, we introduce the generalized force 𝒇~:=(𝒇~1T,𝒇~2T,,𝒇~NT)T\tilde{\bm{f}}:=(\tilde{\bm{f}}_{1}^{\textrm{T}},\tilde{\bm{f}}_{2}^{\textrm{T}},\cdots,\tilde{\bm{f}}_{N}^{\textrm{T}})^{\textrm{T}} with 𝒇~i:=(f~ix,f~iy,f~i)T:=(F~ix/mix,F~iy/miy,F~i/mi)T\tilde{\bm{f}}_{i}:=(\tilde{f}_{i}^{x},\tilde{f}_{i}^{y},\tilde{f}_{i}^{\ell})^{T}:=(\tilde{F}_{i}^{x}/m_{i}^{x},\tilde{F}_{i}^{y}/m_{i}^{y},\tilde{F}_{i}^{\ell}/m_{i}^{\ell})^{\textrm{T}} in this Appendix, where miαm_{i}^{\alpha} is mim_{i} for α=x,y\alpha=x,y and 4Ii/di24I_{i}/d_{i}^{2} for α=\alpha=\ell. Note that F~iα\tilde{F}_{i}^{\alpha} is the generalized force which depends on 𝒒\bm{q} and 𝒒˙\dot{\bm{q}} as in Eqs. (3)-(8), where 𝒂˙:=d𝒂/dt\dot{\bm{a}}:=d\bm{a}/dt for an arbitrary vector 𝒂\bm{a}.

B.1 Accuracy of the velocity Verlet method for the force depending on the velocity

In this subsection, we check the accuracy of the velocity Verlet method for the force depending on the velocity with the aid of discretization based on the Taylor expansion. The velocity Verlte method is given by a set of equations

qiα(t+Δt)\displaystyle q_{i}^{\alpha}(t+\Delta t) =qiα(t)+Δtq˙iα(t)+12Δt2f~iα(t),\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\tilde{f}_{i}^{\alpha}(t), (13)
q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) =q˙iα(t)+Δtf~iα(t)+f~iα(t+Δt)2.\displaystyle=\dot{q}_{i}^{\alpha}\left(t\right)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(t)+\tilde{f}_{i}^{\alpha}(t+\Delta t)}{2}. (14)

The first equation is called the velocity Velret equation for qiα(t)q_{i}^{\alpha}(t) and the second one is the equation for q˙iα(t)\dot{q}_{i}^{\alpha}(t). It is known that the velocity Verlet algorithm has the accuracy of O(Δt2)O(\Delta t^{2}) in Hamiltonian systems Swope82 , but the accuracy of this method for dissipative dynamics is little known. Therefore, we clarify the accuracy of this method in this Appendix.

Here, we show from the Taylor expansion that the velocity Verlet method has a second-order and first-order accuracies of Δt\Delta t for 𝒒\bm{q} and 𝒒˙\dot{\bm{q}}, respectively. Based on the Taylor expansion of qiα(t+Δt)q_{i}^{\alpha}(t+\Delta t), we obtain

qiα(t+Δt)\displaystyle q_{i}^{\alpha}(t+\Delta t) =qiα(t)+Δtq˙iα(t)+12Δt2q¨iα(t)+O(Δt3)\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\ddot{q}_{i}^{\alpha}(t)+O(\Delta t^{3})
=qiα(t)+Δtq˙iα(t)+12Δt2f~iα(t)+O(Δt3),\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\tilde{f}_{i}^{\alpha}(t)+O(\Delta t^{3}), (15)

where A˙:=dA/dt\dot{A}:=dA/dt for an arbitrary function AA. We can obtain the quadratic precision of Δt\Delta t for qiα(t+Δt)q_{i}^{\alpha}(t+\Delta t), because the RHS of Eq. (15) is a function of current time tt.

On the other hand, based on the Taylor expansion of q˙iα(t+Δt)\dot{q}_{i}^{\alpha}(t+\Delta t), we obtain

q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) =q˙iα(t)+Δtq¨iα(t)+12Δt2q˙˙˙iα(t)+O(Δt3)\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\ddot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\dddot{q}_{i}^{\alpha}(t)+O(\Delta t^{3})
=q˙iα(t)+Δtf~iα(t)+12Δt2f~˙iα(t)+O(Δt3).\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\tilde{f}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\dot{\tilde{f}}_{i}^{\alpha}(t)+O(\Delta t^{3}). (16)

By using f~iα(t+Δt)=f~iα(t)+Δtf~˙iα(t)+O(Δt2)\tilde{f}_{i}^{\alpha}(t+\Delta t)=\tilde{f}_{i}^{\alpha}(t)+\Delta t\dot{\tilde{f}}_{i}^{\alpha}(t)+O(\Delta t^{2}), we evaluate f~˙iα(t)\dot{\tilde{f}}_{i}^{\alpha}(t) as

f~˙iα(t)=f~iα(t+Δt)f~iα(t)Δt+O(Δt).\displaystyle\dot{\tilde{f}}_{i}^{\alpha}(t)=\frac{\tilde{f}_{i}^{\alpha}(t+\Delta t)-\tilde{f}_{i}^{\alpha}(t)}{\Delta t}+O(\Delta t). (17)

Substituting Eq. (17) into Eq. (16), we obtain Swope82

q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) =q˙iα(t)+Δtf~iα(t+Δt)+f~iα(t)2+O(Δt3).\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(t+\Delta t)+\tilde{f}_{i}^{\alpha}(t)}{2}+O(\Delta t^{3}). (18)

If the force f~iα(t+Δt)\tilde{f}_{i}^{\alpha}(t+\Delta t) is independent of 𝒒˙\dot{\bm{q}}, we can obtain f~iα(𝒒(t+Δt))\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t)) from 𝒒(t+Δt)\bm{q}(t+\Delta t) with the aid of Eq. (15) Swope82 . However, if f~iα(t+Δt)\tilde{f}_{i}^{\alpha}(t+\Delta t) depends on 𝒒˙\dot{\bm{q}}, we have to evaluate f~iα(𝒒(t+Δt),𝒒˙(t+Δt))\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{q}}(t+\Delta t)), because f~iα(𝒒(t+Δt),𝒒˙(t+Δt))\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{q}}(t+\Delta t)) requires LHS of Eq. (18). Thus, we adopt the following replacements:

f~iα(t+Δt):\displaystyle\tilde{f}_{i}^{\alpha}(t+\Delta t): =f~iα(𝒒(t+Δt),𝒒˙(t+Δt))f~iα(𝒒(t+Δt),𝑸˙(t)),\displaystyle=\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{q}}(t+\Delta t))\to\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{Q}}(t)), (19)
f~iα(t):\displaystyle\tilde{f}_{i}^{\alpha}(t): =f~iα(𝒒(t),𝒒˙(t))f~iα(𝒒(t),𝑸˙(tΔt)),\displaystyle=\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t))\to\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t)), (20)

where we have introduced

Q˙iα(t):\displaystyle\dot{Q}_{i}^{\alpha}(t): =q˙iα(t)+Δtf~iα(𝒒(t),𝑸˙(tΔt))2,\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t))}{2}, (21)
Q˙iα(tΔt):\displaystyle\dot{Q}_{i}^{\alpha}(t-\Delta t): =q˙iα(tΔt)+Δtf~iα(𝒒(tΔt),𝑸˙(t2Δt))2.\displaystyle=\dot{q}_{i}^{\alpha}(t-\Delta t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(\bm{q}(t-\Delta t),\dot{\bm{Q}}(t-2\Delta t))}{2}. (22)

Here, the difference between f~iα(𝒒(t+Δt),𝒒˙(t+Δt))\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{q}}(t+\Delta t)) and f~iα(𝒒(t+Δt),𝑸˙(t))\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{Q}}(t)) caused by Eq. (19) is given by

Δf~iα(t+Δt)\displaystyle\Delta\tilde{f}_{i}^{\alpha}(t+\Delta t) :=f~iα(𝒒(t+Δt),𝒒˙(t+Δt))f~iα(𝒒(t+Δt),𝑸˙(t))\displaystyle:=\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{q}}(t+\Delta t))-\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{Q}}(t))
=f~iα(𝒒(t+Δt),𝒒˙(t)+Δt𝒇~(t)+O(Δt2))f~iα(𝒒(t+Δt),𝒒˙(t)+Δt𝒇~(𝒒(t),𝒒˙(t))2+O(Δt2))\displaystyle=\tilde{f}_{i}^{\alpha}\left(\bm{q}(t+\Delta t),\dot{\bm{q}}(t)+\Delta t\tilde{\bm{f}}(t)+O(\Delta t^{2})\right)-\tilde{f}_{i}^{\alpha}\left(\bm{q}(t+\Delta t),\dot{\bm{q}}(t)+\Delta t\frac{\tilde{\bm{f}}(\bm{q}(t),\dot{\bm{q}}(t))}{2}+O(\Delta t^{2})\right)
=Δtj=1Nβ=x,y,f~jβ(𝒒(t),𝒒˙(t))2f~iα(𝒒(t),𝒒˙(t))q˙jβ+O(Δt2).\displaystyle=\Delta t\sum_{j=1}^{N}\sum_{\beta=x,y,\ell}\frac{\tilde{f}_{j}^{\beta}(\bm{q}(t),\dot{\bm{q}}(t))}{2}\frac{\partial\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t))}{\partial\dot{q}_{j}^{\beta}}+O(\Delta t^{2}). (23)

Similarly, the difference between f~iα(t):=f~iα(𝒒(t),𝒒˙(t))\tilde{f}_{i}^{\alpha}(t):=\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t)) and f~iα(𝒒(t),𝑸˙(tΔt))\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t)) in Eq. (20) can be evaluated as

Δf~iα(t)\displaystyle\Delta\tilde{f}_{i}^{\alpha}(t) :=f~iα(𝒒(t),𝒒˙(t))f~iα(𝒒(t),𝑸˙(tΔt))\displaystyle:=\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t))-\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t))
=Δtj=1Nβ=x,y,f~jβ(𝒒(t),𝒒˙(t))2f~iα(𝒒(t),𝒒˙(t))q˙jβ+O(Δt2).\displaystyle=\Delta t\sum_{j=1}^{N}\sum_{\beta=x,y,\ell}\frac{\tilde{f}_{j}^{\beta}(\bm{q}(t),\dot{\bm{q}}(t))}{2}\frac{\partial\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t))}{\partial\dot{q}_{j}^{\beta}}+O(\Delta t^{2}). (24)

Thus, the replacement of Eqs. (19) and (20) in Eq. (18) with Eqs. (23) and (24) leads to

q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) =q˙iα(t)+Δtf~iα(𝒒(t+Δt),𝑸˙(t))+𝒇~(𝒒(t),𝑸˙(tΔt))2\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{Q}}(t))+\tilde{\bm{f}}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t))}{2}
Δt2j=1Nβ=x,y,f~jβ(𝒒(t),𝒒˙(t))2f~iα(𝒒(t),𝒒˙(t))q˙jβ+O(Δt3).\displaystyle\quad-\Delta t^{2}\sum_{j=1}^{N}\sum_{\beta=x,y,\ell}\frac{\tilde{f}_{j}^{\beta}(\bm{q}(t),\dot{\bm{q}}(t))}{2}\frac{\partial\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{q}}(t))}{\partial\dot{q}_{j}^{\beta}}+O(\Delta t^{3}). (25)

Omitting the term including O(Δt2)O(\Delta t^{2}) in Eq.(25), we obtain the following numerical integration methods for 𝒒˙\dot{\bm{q}}:

q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) q˙iα(t)+Δtf~iα(𝒒(t+Δt),𝑸˙(t))+𝒇~(𝒒(t),𝑸˙(tΔt))2.\displaystyle\to\dot{q}_{i}^{\alpha}(t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}(\bm{q}(t+\Delta t),\dot{\bm{Q}}(t))+\tilde{\bm{f}}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t))}{2}. (26)

Note that Eq. (26) is the precise expression of the velocity Verlet scheme for 𝒒˙\dot{\bm{q}} presented in Eq. (14). From the comparison between Eqs. (25) and (26), we have confirmed that the velocity Verlet scheme has the first-order precision of Δt\Delta t for 𝒒˙\dot{\bm{q}}. We also note that if f~iα\tilde{f}_{i}^{\alpha} is independent of 𝒒˙\dot{\bm{q}} as in the case of Hamiltonian systems, the term proportional to Δt2\Delta t^{2} is zero and thus, the second-order accuracy of Δt\Delta t for q˙iα\dot{q}_{i}^{\alpha} is guaranteed.

Let us go back to Eq. (15) with the replacement of Eq. (20) for f~iα(t)\tilde{f}_{i}^{\alpha}(t):

qiα(t+Δt)\displaystyle q_{i}^{\alpha}(t+\Delta t) =qiα(t)+Δtq˙iα(t)+12Δt2f~iα(t)+O(Δt3)\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\tilde{f}_{i}^{\alpha}(t)+O(\Delta t^{3})
=qiα(t)+Δtq˙iα(t)+12Δt2f~iα(𝒒(t),𝑸˙(tΔt))+O(Δt3).\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t))+O(\Delta t^{3}). (27)

Omitting the term including O(Δt3)O(\Delta t^{3}) in Eq.(27), we obtain the following numerical integration methods for 𝒒\bm{q}:

qiα(t+Δt)\displaystyle q_{i}^{\alpha}(t+\Delta t) qiα(t)+Δtq˙iα(t)+12Δt2f~iα(𝒒(t),𝑸˙(tΔt)).\displaystyle\to q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\frac{1}{2}\Delta t^{2}\tilde{f}_{i}^{\alpha}(\bm{q}(t),\dot{\bm{Q}}(t-\Delta t)). (28)

Here, Eq. (28) is the precise expression of the velocity Verlet scheme for 𝒒\bm{q} presented in Eq. (13). From Eqs. (27) and (28) we have confirmed that the velocity Verlet scheme has the second-order precision of Δt\Delta t for 𝒒\bm{q}.

B.2 Implementation of the velocity Verlet method for the force depending on the velocity

In this subsection, we explain how to adopt the velocity Verlet method in our system. In the main text, we adopt the following equations:

qiα(t+Δt)\displaystyle q_{i}^{\alpha}(t+\Delta t) =qiα(t)+Δtq˙iα(t)+Δt2f~iα(𝒒(t),𝑸˙(tΔt))2,\displaystyle=q_{i}^{\alpha}(t)+\Delta t\dot{q}_{i}^{\alpha}(t)+\Delta t^{2}\frac{\tilde{f}_{i}^{\alpha}\left(\bm{q}(t),\dot{\bm{Q}}\left(t-\Delta t\right)\right)}{2}, (29)
Q˙iα(t)\displaystyle\dot{Q}_{i}^{\alpha}\left(t\right) =q˙iα(t)+Δtf~iα(𝒒(t),𝑸˙(tΔt))2.\displaystyle=\dot{q}_{i}^{\alpha}(t)+\Delta t\frac{\tilde{f}_{i}^{\alpha}\left(\bm{q}(t),\dot{\bm{Q}}\left(t-\Delta t\right)\right)}{2}. (30)

Here, the updated configuration qiα(t+Δt)q_{i}^{\alpha}(t+\Delta t) and modified velocity Q˙iα(t)\dot{Q}_{i}^{\alpha}\left(t\right) are used to obtain the force f~iα(𝒒(t+Δt),𝑸˙(t))\tilde{f}_{i}^{\alpha}\left(\bm{q}(t+\Delta t),\dot{\bm{Q}}\left(t\right)\right). Then, we update the velocity as follows

q˙iα(t+Δt)\displaystyle\dot{q}_{i}^{\alpha}(t+\Delta t) =Q˙iα(t)+Δtf~iα(𝒒(t+Δt),𝑸˙(t))2.\displaystyle=\dot{Q}_{i}^{\alpha}\left(t\right)+\Delta t\frac{\tilde{f}_{i}^{\alpha}\left(\bm{q}(t+\Delta t),\dot{\bm{Q}}\left(t\right)\right)}{2}. (31)

Appendix C Jacobian Properties

In this Appendix, we summarize the properties of the Jacobian introduced in Eq. (18).

C.1 Jacobian block elements

Let us write 3×33\times 3 sub-matrix 𝒥ij\mathcal{J}_{ij}, which is (ij)(ij) block element of the Jacobian obtained from Eq. (18):

[𝒥ij]αβ\displaystyle[\mathcal{J}_{ij}]^{\alpha\beta} :=F~iαqjβ\displaystyle:=-\frac{\partial\tilde{F}_{i}^{\alpha}}{\partial q_{j}^{\beta}}
=[qjxFixqjyFixqjFixqjxFiyqjyFiyqjFiyqjxT~iqjyT~iqjT~i]\displaystyle=\left[\begin{matrix}-\partial_{q_{j}^{x}}F_{i}^{x}&-\partial_{q_{j}^{y}}F_{i}^{x}&-\partial_{q_{j}^{\ell}}F_{i}^{x}\\ -\partial_{q_{j}^{x}}F_{i}^{y}&-\partial_{q_{j}^{y}}F_{i}^{y}&-\partial_{q_{j}^{\ell}}F_{i}^{y}\\ -\partial_{q_{j}^{x}}\tilde{T}_{i}&-\partial_{q_{j}^{y}}\tilde{T}_{i}&-\partial_{q_{j}^{\ell}}\tilde{T}_{i}\end{matrix}\right]
=[k=1;kjNqjxfikxk=1;kjNqjyfikxk=1;kjNqjfikxk=1;kjNqjxfikyk=1;kjNqjyfikyk=1;kjNqjfikyk=1;kjNqjxT~ikk=1;kjNqjyT~ikk=1;kjNqjT~ik]\displaystyle=\left[\begin{matrix}-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{x}}f_{ik}^{x}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{y}}f_{ik}^{x}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{\ell}}f_{ik}^{x}\\ -\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{x}}f_{ik}^{y}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{y}}f_{ik}^{y}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{\ell}}f_{ik}^{y}\\ -\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{x}}\tilde{T}_{ik}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{y}}\tilde{T}_{ik}&-\sum_{k=1;k\neq j}^{N}\partial_{q_{j}^{\ell}}\tilde{T}_{ik}\end{matrix}\right]
={[qjxfijxqjyfijxqjfijxqjxfijyqjyfijyqjfijyqjxT~ijqjyT~ijqjT~ij](ij)[k=1;kiNqixfikxk=1;kiNqiyfikxk=1;kiNqifikxk=1;kiNqixfikyk=1;kiNqiyfikyk=1;kiNqifikyk=1;kiNqixT~ikk=1;kiNqiyT~ikk=1;kiNqiT~ik](i=j),\displaystyle=\begin{cases}\left[\begin{matrix}-\partial_{q_{j}^{x}}f_{ij}^{x}&-\partial_{q_{j}^{y}}f_{ij}^{x}&-\partial_{q_{j}^{\ell}}f_{ij}^{x}\\ -\partial_{q_{j}^{x}}f_{ij}^{y}&-\partial_{q_{j}^{y}}f_{ij}^{y}&-\partial_{q_{j}^{\ell}}f_{ij}^{y}\\ -\partial_{q_{j}^{x}}\tilde{T}_{ij}&-\partial_{q_{j}^{y}}\tilde{T}_{ij}&-\partial_{q_{j}^{\ell}}\tilde{T}_{ij}\end{matrix}\right](i\neq j)\\ \left[\begin{matrix}-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{x}}f_{ik}^{x}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{y}}f_{ik}^{x}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{\ell}}f_{ik}^{x}\\ -\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{x}}f_{ik}^{y}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{y}}f_{ik}^{y}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{\ell}}f_{ik}^{y}\\ -\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{x}}\tilde{T}_{ik}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{y}}\tilde{T}_{ik}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{i}^{\ell}}\tilde{T}_{ik}\end{matrix}\right](i=j)\end{cases}, (32)

where the superscripts α\alpha and β\beta correspond to x,y,x,y,\ell-components, and ii and jj are the particle numbers (see Appendix F for each component of 𝒥\mathcal{J}). Here, fijζ,T~ijf_{ij}^{\zeta},\tilde{T}_{ij} are ζ\zeta-component of 𝒇ij\bm{f}_{ij} and scaled torque that the ii-th particle receives from the jj-th particle, respectively. The sub-matrix for i=ji=j is given by

[𝒥ii]αβ=[k=1;kiNqkxfikxk=1;kiNqkyfikxk=1;kiNqkfikxk=1;kiNqkxfikyk=1;kiNqkyfikyk=1;kiNqkfikyk=1;kiNqkxT~ikk=1;kiNqkyT~ikk=1;kiNqkT~ik],\displaystyle[\mathcal{J}_{ii}]^{\alpha\beta}=\left[\begin{matrix}\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{x}}f_{ik}^{x}&\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{y}}f_{ik}^{x}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{\ell}}f_{ik}^{x}\\ \sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{x}}f_{ik}^{y}&\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{y}}f_{ik}^{y}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{\ell}}f_{ik}^{y}\\ \sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{x}}\tilde{T}_{ik}&\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{y}}\tilde{T}_{ik}&-\sum_{k=1;k\neq i}^{N}\partial_{q_{k}^{\ell}}\tilde{T}_{ik}\end{matrix}\right], (33)

where we have used qiκfikζ=qkκfikζ,qiκT~ik=qiκT~ik,qifikζ=qkfikζ\partial_{q_{i}^{\kappa}}f_{ik}^{\zeta}=-\partial_{q_{k}^{\kappa}}f_{ik}^{\zeta},\partial_{q_{i}^{\kappa}}\tilde{T}_{ik}=-\partial_{q_{i}^{\kappa}}\tilde{T}_{ik},\partial_{q_{i}^{\ell}}f_{ik}^{\zeta}=\partial_{q_{k}^{\ell}}f_{ik}^{\zeta}, and qiT~ik=qiT~ik\partial_{q_{i}^{\ell}}\tilde{T}_{ik}=\partial_{q_{i}^{\ell}}\tilde{T}_{ik}. Here, the superscripts ζ\zeta and κ\kappa correspond to x,yx,y components.

From Eqs. (32) and (33) 𝒥ijζβ\mathcal{J}_{ij}^{\zeta\beta} satisfies

𝒥iiζβ\displaystyle\mathcal{J}_{ii}^{\zeta\beta} =ji𝒥ijζβ\displaystyle=-\sum_{j\neq i}\mathcal{J}_{ij}^{\zeta\beta} (34)

Thus, introducing Jnm(n,m=1,2,,3N)J_{nm}\ (n,m=1,2,\cdots,3N) which is a rewriting of 𝒥ijαβ\mathcal{J}_{ij}^{\alpha\beta} in Eq. (19) by the index from ii and α\alpha to nn, we obtain

n=1,4,,3N2Jnm\displaystyle\sum_{n=1,4,\cdots,3N-2}J_{nm} =0,\displaystyle=0, (35)
n=2,5,,3N1Jnm\displaystyle\sum_{n=2,5,\cdots,3N-1}J_{nm} =0.\displaystyle=0. (36)

where n=1,4,,3N2\sum_{n=1,4,\cdots,3N-2} and n=2,5,,3N1\sum_{n=2,5,\cdots,3N-1} express the summations of modulus 11 and modulus 22 with the intervals 33, respectively. Here, we write 3N3N-dimensional vector translating in the xx direction 𝒆x\bm{e}_{x} as

𝒆x\displaystyle\bm{e}_{x} =[𝒆x,1𝒆x,2𝒆x,N].\displaystyle=\left[\begin{matrix}\bm{e}_{x,1}\\ \bm{e}_{x,2}\\ \vdots\\ \bm{e}_{x,N}\end{matrix}\right]. (37)

where 𝒆x,i:=(1,0,0)T\bm{e}_{x,i}:=(1,0,0)^{\textrm{T}} for i=1,2,,Ni=1,2,\cdots,N. Here, the nn-th component of the action of 𝒥\mathcal{J} on 𝒆x\bm{e}_{x} satisfies

{𝒥𝒆x}n\displaystyle\{\mathcal{J}\bm{e}_{x}\}_{n} =mJnmex,m\displaystyle=\sum_{m}J_{nm}e_{x,m}
=m=1,4,,3N2Jnm\displaystyle=\sum_{m=1,4,\cdots,3N-2}J_{nm}
=0,\displaystyle=0, (38)

where we have used Eq. (35) for the last equality. Thus, we obtain 𝒥𝒆x=𝟎\mathcal{J}\bm{e}_{x}=\bm{0}, where 𝟎\bm{0} is zero vector. Similarly, using

𝒆y\displaystyle\bm{e}_{y} =[𝒆y,1𝒆y,2𝒆y,N],\displaystyle=\left[\begin{matrix}\bm{e}_{y,1}\\ \bm{e}_{y,2}\\ \vdots\\ \bm{e}_{y,N}\end{matrix}\right], (39)

with 𝒆y,i:=(0,1,0)T\bm{e}_{y,i}:=(0,1,0)^{\textrm{T}} we also obtain 𝒥𝒆y=𝟎\mathcal{J}\bm{e}_{y}=\bm{0}. Therefore, 𝒆x\bm{e}_{x} and 𝒆y\bm{e}_{y} are the zero modes for 𝒥\mathcal{J}.

Appendix D Explicit Jacobian expressions

In this Appendix, we present the explicit expressions of the Jacobian based on Eqs. (6)-(8). Then, we clarify the difference between the present results and the case where the tangential force is approximated by the conservative force used in the previous studies Somfai07 ; Henkes10 .

D.1 Calculation of Jacobian

Let us consider only the normal and tangential elastic contact forces

𝒇N,ij\displaystyle\bm{f}_{N,ij} =kNξN,ij3/2𝒏ij,\displaystyle=k_{N}\xi_{N,ij}^{3/2}\bm{n}_{ij}, (40)
𝒇T,ij\displaystyle\bm{f}_{T,ij} =kTξN,ij1/2𝝃T,ij,\displaystyle=k_{T}\xi_{N,ij}^{1/2}\bm{\xi}_{T,ij}, (41)

where the integration of d𝝃T,ijd\bm{\xi}_{T,ij}

𝝃T,ij:=Cij𝑑𝝃T,ij\displaystyle\bm{\xi}_{T,ij}:=\int_{C_{ij}}d\bm{\xi}_{T,ij} (42)

is performed during the contact between ii and jj particles. Since Eq. (42) does not contain the second term on RHS of Eq. (9), 𝝃T,ij\bm{\xi}_{T,ij} may not be perpendicular to 𝝃N,ij\bm{\xi}_{N,ij}. Neverthelss, we adopt Eq. (42) for simplicity. Here, d𝝃T,ijd{\bm{\xi}}_{T,ij} is defined as

d𝝃T,ij=d𝒓ij(d𝒓ij𝒏ij)𝒏ijdij×𝒏ij,\displaystyle d{\bm{\xi}}_{T,ij}=d{\bm{r}}_{ij}-(d{\bm{r}}_{ij}\cdot\bm{n}_{ij})\bm{n}_{ij}-d{\bm{\ell}}_{ij}\times\bm{n}_{ij}, (43)

where ij\bm{\ell}_{ij} is defined as

ij:=[00i+j].\displaystyle\bm{\ell}_{ij}:=\left[\begin{matrix}0\\ 0\\ \ell_{i}+\ell_{j}\end{matrix}\right]. (44)

Each component of Eq. (43) is written as

dξT,ijx\displaystyle d{\xi}_{T,ij}^{x} =drijx(d𝒓ij𝒏ij)nijx+dijnijy,\displaystyle=d{{r}}_{ij}^{x}-(d{\bm{r}}_{ij}\cdot\bm{n}_{ij}){n}_{ij}^{x}+d{{\ell}}_{ij}{n}_{ij}^{y}, (45)
dξT,ijy\displaystyle d{\xi}_{T,ij}^{y} =drijy(d𝒓ij𝒏ij)nijydijnijx.\displaystyle=d{{r}}_{ij}^{y}-(d{\bm{r}}_{ij}\cdot\bm{n}_{ij}){n}_{ij}^{y}-d{{\ell}}_{ij}{n}_{ij}^{x}. (46)

The derivative of the normal force is given by

riζfN,ijκ\displaystyle\partial_{r_{i}^{\zeta}}f_{N,ij}^{\kappa} =kN[δζκξN,ij3/2rij(32+ξN,ijrij)ξN,ij1/2nijζnijκ],\displaystyle=k_{N}\left[\delta_{\zeta\kappa}\frac{\xi_{N,ij}^{3/2}}{r_{ij}}-\left(\frac{3}{2}+\frac{\xi_{N,ij}}{r_{ij}}\right)\xi_{N,ij}^{1/2}n_{ij}^{\zeta}n_{ij}^{\kappa}\right], (47)
ifN,ijκ\displaystyle\partial_{\ell_{i}}f_{N,ij}^{\kappa} =0,\displaystyle=0, (48)

where Kronecker’s delta δζκ\delta_{\zeta\kappa} satisfies δζκ=1\delta_{\zeta\kappa}=1 for ζ=κ\zeta=\kappa and δζκ=0\delta_{\zeta\kappa}=0 otherwise. We have used

nijζriκ\displaystyle\frac{\partial n_{ij}^{\zeta}}{\partial r_{i}^{\kappa}} =1rij(δζκnijζnijκ),\displaystyle=\frac{1}{r_{ij}}\left(\delta_{\zeta\kappa}-n_{ij}^{\zeta}n_{ij}^{\kappa}\right), (49)
rijriζ\displaystyle\frac{\partial r_{ij}}{\partial r_{i}^{\zeta}} =nijζ\displaystyle=n_{ij}^{\zeta} (50)

to obtain Eq. (47).

The derivative of the tangential force is written as

riζfT,ijκ\displaystyle\partial_{r_{i}^{\zeta}}f_{T,ij}^{\kappa} =12kTξN,ij1/2nijζξT,ijκkTξN,ij1/2(δζκnijζnijκ),\displaystyle=\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}n_{ij}^{\zeta}\xi_{T,ij}^{\kappa}-k_{T}\xi_{N,ij}^{1/2}\left(\delta_{\zeta\kappa}-n_{ij}^{\zeta}n_{ij}^{\kappa}\right), (51)
ifT,ijκ\displaystyle\partial_{\ell_{i}}f_{T,ij}^{\kappa} =εκkTξN,ij1/2nijνκ,\displaystyle=-\varepsilon_{\kappa}k_{T}\xi_{N,ij}^{1/2}n_{ij}^{\nu_{\kappa}}, (52)

where εζ\varepsilon_{\zeta} and νζ\nu_{\zeta} are, respectively, defined as

εζ:\displaystyle\varepsilon_{\zeta}: ={1(ζ=x)1(ζ=y),\displaystyle=\left\{\begin{matrix}1\quad(\zeta=x)\\ -1\quad(\zeta=y),\end{matrix}\right. (53)
νζ:\displaystyle\nu_{\zeta}: ={y(ζ=x)x(ζ=y).\displaystyle=\left\{\begin{matrix}y\quad(\zeta=x)\\ x\quad(\zeta=y).\end{matrix}\right. (54)

Here, riζξT,ijκ\partial_{r_{i}^{\zeta}}{\xi}_{T,ij}^{\kappa} and iξT,ijκ\partial_{\ell_{i}}{\xi}_{T,ij}^{\kappa} in Eqs. (51) and (52) satisfy

ξT,ijκriζ\displaystyle\frac{\partial{\xi}_{T,ij}^{\kappa}}{\partial r_{i}^{\zeta}} =δζκnijζnijκ,\displaystyle=\delta_{\zeta\kappa}-n_{ij}^{\zeta}n_{ij}^{\kappa}, (55)
ξT,ijκi\displaystyle\frac{\partial{\xi}_{T,ij}^{\kappa}}{\partial\ell_{i}} =εκnijνκ.\displaystyle=\varepsilon_{\kappa}n_{ij}^{\nu_{\kappa}}. (56)

The derivation of Eqs. (55) and (56) are as follows Chattoraj19 . From Eq. (43) dξT,ijζd{\xi}_{T,ij}^{\zeta} can be written as

dξT,ijζ=drijζ(d𝒓ij𝒏ij)nijζ+(1)ζ(di+dj)nijνζ.\displaystyle d{\xi}_{T,ij}^{\zeta}=dr_{ij}^{\zeta}-(d\bm{r}_{ij}\cdot\bm{n}_{ij})n_{ij}^{\zeta}+(-1)^{\zeta}(d\ell_{i}+d\ell_{j})n_{ij}^{\nu_{\zeta}}. (57)

Then, dξT,ijxd\xi_{T,ij}^{x} satisfies

dξT,ijx\displaystyle d{\xi}_{T,ij}^{x} =drijxκ=x,ydrijκnijκnijx+nijy(di+dj)\displaystyle=dr_{ij}^{x}-\sum_{\kappa=x,y}dr_{ij}^{\kappa}n_{ij}^{\kappa}n_{ij}^{x}+n_{ij}^{y}(d\ell_{i}+d\ell_{j})
=(1(nijx)2)drijxnijxnijydrijy+nijy(di+dj)\displaystyle=(1-(n_{ij}^{x})^{2})dr_{ij}^{x}-n_{ij}^{x}n_{ij}^{y}dr_{ij}^{y}+n_{ij}^{y}(d\ell_{i}+d\ell_{j})
=(nijy)2drijxnijxnijydrijy+nijy(di+dj)\displaystyle=(n_{ij}^{y})^{2}dr_{ij}^{x}-n_{ij}^{x}n_{ij}^{y}dr_{ij}^{y}+n_{ij}^{y}(d\ell_{i}+d\ell_{j})
=(nijy)2(dxidxj)nijxnijy(dyidyj)+nijy(di+dj).\displaystyle=(n_{ij}^{y})^{2}(dx_{i}-dx_{j})-n_{ij}^{x}n_{ij}^{y}(dy_{i}-dy_{j})+n_{ij}^{y}(d\ell_{i}+d\ell_{j}). (58)

Similarly, dξT,ijyd\xi_{T,ij}^{y} also satisfies

dξT,ijy\displaystyle d\xi_{T,ij}^{y} =nijxnijy(dxidxj)+(nijy)2(dyidyj)nijx(di+dj).\displaystyle=-n_{ij}^{x}n_{ij}^{y}(dx_{i}-dx_{j})+(n_{ij}^{y})^{2}(dy_{i}-dy_{j})-n_{ij}^{x}(d\ell_{i}+d\ell_{j}). (59)

Here, dξT,ijζd{\xi}_{T,ij}^{\zeta}is the function of xi,yi,i,xj,yj,x_{i},y_{i},\ell_{i},x_{j},y_{j}, and j\ell_{j}. We obtain the differential form of dξT,ijζd{\xi}_{T,ij}^{\zeta}:

dξT,ijζ\displaystyle d{\xi}_{T,ij}^{\zeta} =(ξT,ijζxi)(yi,i,xj,yj,j)dxi+(ξT,ijζxj)(xi,yi,i,yj,j)dxj\displaystyle=\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial x_{i}}\right)_{(y_{i},\ell_{i},x_{j},y_{j},\ell_{j})}dx_{i}+\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial x_{j}}\right)_{(x_{i},y_{i},\ell_{i},y_{j},\ell_{j})}dx_{j}
+(ξT,ijζyi)(xi,i,xj,yj,j)dyi+(ξT,ijζyj)(xi,yi,i,xj,j)dyj\displaystyle\quad+\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial y_{i}}\right)_{(x_{i},\ell_{i},x_{j},y_{j},\ell_{j})}dy_{i}+\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial y_{j}}\right)_{(x_{i},y_{i},\ell_{i},x_{j},\ell_{j})}dy_{j}
+(ξT,ijζi)(xi,yi,xj,yj,j)di+(ξT,ijζj)(xi,yi,i,xj,yj)dj.\displaystyle\quad+\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial\ell_{i}}\right)_{(x_{i},y_{i},x_{j},y_{j},\ell_{j})}d\ell_{i}+\left(\frac{\partial{\xi}_{T,ij}^{\zeta}}{\partial\ell_{j}}\right)_{(x_{i},y_{i},\ell_{i},x_{j},y_{j})}d\ell_{j}. (60)

Then, we obtain Eqs. (55), (56), by comparing Eqs. (58) and (59) with Eq. (60).

Since the scaled torque T~ij\tilde{T}_{ij} satisfies

T~ij:=2Tijdi=nijxfT,ijy+nijyfT,ijx,\displaystyle\tilde{T}_{ij}:=\frac{2T_{ij}}{d_{i}}=-n_{ij}^{x}f_{T,ij}^{y}+n_{ij}^{y}f_{T,ij}^{x}, (61)

we obtain

riζT~ij\displaystyle\partial_{r_{i}^{\zeta}}\tilde{T}_{ij} =(riζnijx)fT,ijynijxriζfT,ijy+(riζnijy)fT,ijx+nijyriζfT,ijx\displaystyle=-\left(\partial_{r_{i}^{\zeta}}n_{ij}^{x}\right)f_{T,ij}^{y}-n_{ij}^{x}\partial_{r_{i}^{\zeta}}f_{T,ij}^{y}+\left(\partial_{r_{i}^{\zeta}}n_{ij}^{y}\right)f_{T,ij}^{x}+n_{ij}^{y}\partial_{r_{i}^{\zeta}}f_{T,ij}^{x}
=(δζxrijnijζnijxrij)fT,ijynijx[12kTξN,ij1/2ξT,ijnijζtijykTξN,ij1/2(δζynijζnijy)]\displaystyle=-\left(\frac{\delta_{\zeta x}}{r_{ij}}-\frac{n_{ij}^{\zeta}n_{ij}^{x}}{r_{ij}}\right)f_{T,ij}^{y}-n_{ij}^{x}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{y}-k_{T}\xi_{N,ij}^{1/2}\left(\delta_{\zeta y}-n_{ij}^{\zeta}n_{ij}^{y}\right)\right]
+(δζyrijnijζnijyrij)fT,ijx+nijy[12kTξN,ij1/2ξT,ijnijζtijxkTξN,ij1/2(δζxnijζnijx)]\displaystyle\quad+\left(\frac{\delta_{\zeta y}}{r_{ij}}-\frac{n_{ij}^{\zeta}n_{ij}^{y}}{r_{ij}}\right)f_{T,ij}^{x}+n_{ij}^{y}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{x}-k_{T}\xi_{N,ij}^{1/2}\left(\delta_{\zeta x}-n_{ij}^{\zeta}n_{ij}^{x}\right)\right]
=nijx[12kTξN,ij1/2ξT,ijnijζtijykTξN,ij1/2(δζynijζnijy)]\displaystyle=-n_{ij}^{x}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{y}-k_{T}\xi_{N,ij}^{1/2}\left(\delta_{\zeta y}-n_{ij}^{\zeta}n_{ij}^{y}\right)\right]
+nijy[12kTξN,ij1/2ξT,ijnijζtijxkTξN,ij1/2(δζxnijζnijx)]\displaystyle\quad+n_{ij}^{y}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{x}-k_{T}\xi_{N,ij}^{1/2}\left(\delta_{\zeta x}-n_{ij}^{\zeta}n_{ij}^{x}\right)\right]\quad
=nijx[12kTξN,ij1/2ξT,ijnijζtijykTξN,ij1/2δζy]+nijy[12kTξN,ij1/2ξT,ijnijζtijxkTξN,ij1/2δζx],\displaystyle=-n_{ij}^{x}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{y}-k_{T}\xi_{N,ij}^{1/2}\delta_{\zeta y}\right]+n_{ij}^{y}\left[\frac{1}{2}k_{T}\xi_{N,ij}^{-1/2}\xi_{T,ij}n_{ij}^{\zeta}t_{ij}^{x}-k_{T}\xi_{N,ij}^{1/2}\delta_{\zeta x}\right], (62)
iT~ij\displaystyle\partial_{\ell_{i}}\tilde{T}_{ij} =nijxifT,ijy+nijyifT,ijx\displaystyle=-n_{ij}^{x}\partial_{\ell_{i}}f_{T,ij}^{y}+n_{ij}^{y}\partial_{\ell_{i}}f_{T,ij}^{x}
=nijxkTξN,ij1/2nijxnijykTξN,ij1/2nijy\displaystyle=-n_{ij}^{x}k_{T}\xi_{N,ij}^{1/2}n_{ij}^{x}-n_{ij}^{y}k_{T}\xi_{N,ij}^{1/2}n_{ij}^{y}
=kTξN,ij1/2,\displaystyle=-k_{T}\xi_{N,ij}^{1/2}, (63)

where we have used ζfT,ijζnijζ=0\sum_{\zeta}f_{T,ij}^{\zeta}n_{ij}^{\zeta}=0.

The terms proportional to ξT,ij\xi_{T,ij} in the Jacobian include the history-dependent tangential displacements which are ignored in the effective potential (see Appendix H) Somfai07 ; Henkes10 ; Liu21 . The reason we use the Jacobian is to include the history-dependent tangential displacements in the dynamical matrix.

Appendix E Effects of rattlers

In this Appendix, we investigate the effects of rattlers. In the first subsection, we investigate the effects of rattlers for the DOS. In the second subsection, we clarify the contributions of rattlers by using the participation ratio.

E.1 Effects of rattlers on the DOS

In this subsection, we investigate the role of rattlers. We call particle ii a rattler, if its coordination number ZiZ_{i} is ZiZThZ_{i}\leq Z_{\textrm{Th}}. Since the coordination number of isostatic state is three, ZThZ_{\textrm{Th}} can be 11 or 22 for frictional grains. The rattlers are determined by the following method. Given a particle configuration, we measure the coordination number Zi(n=1)Z_{i}^{(n=1)} of each particle. Then, we regard N1N_{1} particles satisfying Zi(n=1)ZThZ_{i}^{(n=1)}\leq Z_{\textrm{Th}} as rattlers at the first trial. We measure the coordination number Zi(n=2)Z_{i}^{(n=2)} after we remove the rattler particles. In the second trial, we regard particles satisfying Zi(n=2)ZThZ_{i}^{(n=2)}\leq Z_{\textrm{Th}} as new rattlers. We repeat these processes until the number of rattlers is converged. As shown in Fig. 13, at which we adopt ZTh=2Z_{\textrm{Th}}=2, low-frequency modes in Region I and intermediate modes between Regions I and II are contributions from rattlers. Thus, we conclude that the contributions of rattlers on the DOS are not important.

Refer to caption
Figure 13: Double logarithmic plots of D(ω)D(\omega) with (red circles) and without (blue triangles by using ZTh=2Z_{\textrm{Th}}=2) rattlers against ωt0\omega t_{0} for ϕ=0.90\phi=0.90 at (a) kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8} and (b) kT/kN=1.0×104k_{T}/k_{N}=1.0\times 10^{-4}. These figures are based on numerical results for N=4096N=4096.

E.2 Participation ratio

In this subsection, to clarify whether the mode at ωn\omega_{n} is localized or spread to the whole system, we introduce a participation ratio pnp_{n} Yunker11 ; Shiraishi19

pn:=(i=1N|𝑹n,i|2)2Ni=1N|𝑹n,i|4.\displaystyle p_{n}:=\frac{\left(\sum_{i=1}^{N}|\bm{R}_{n,i}|^{2}\right)^{2}}{N\sum_{i=1}^{N}|\bm{R}_{n,i}|^{4}}. (64)

We plot p(ω)p(\omega) defined as

p(ω):=npnδ(ωωn)nδ(ωωn)\displaystyle p(\omega):=\frac{\sum_{n}\nolimits^{\prime}\langle p_{n}\delta(\omega-\omega_{n})\rangle}{\sum_{n}\nolimits^{\prime}\langle\delta(\omega-\omega_{n})\rangle} (65)

against ωt0\omega t_{0} for ϕ=0.90\phi=0.90 in Fig. 14. Note that p(ω)p(\omega) are set to be zero if there is no right eigenvalue in the region (ω(s)<ω<ω(s+1))(\omega^{(s)}<\omega<\omega^{(s+1)}).

Refer to caption
Figure 14: Double logarithmic plots of p(ω)p(\omega) for ϕ=0.90\phi=0.90 against ωt0\omega t_{0} at (a) kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8} and (b) kT/kN=1.0×104k_{T}/k_{N}=1.0\times 10^{-4}, where the guide line is 1/N1/N. These figures are based on numerical results for N=4096N=4096.

Figure 14 shows that the modes at ωt0105\omega t_{0}\approx 10^{-5} in Fig. 14 (a) and at ωt0103\omega t_{0}\approx 10^{-3} in Fig. 14 (b) are nearly equal to p1/Np\approx 1/N. Recalling that those modes consist of the rattler, we conclude that the contribution of the rattler is localized. In the middle range of ω\omega in Fig. 14, there is an isolated band which shifts to the large ω\omega as kT/kNk_{T}/k_{N} increases with keeping its shape which can be seen in the main text.

Appendix F Explicit results of 𝒥N\mathcal{J}_{N} and 𝒥T\mathcal{J}_{T}

In this Appendix, we have written down the explicit results of 𝒥N\mathcal{J}_{N} and 𝒥T\mathcal{J}_{T}. From the results for the derivative of F~iα\tilde{F}_{i}^{\alpha} in Appendix D , the non-diagonal block elements 𝒥N,ijαβ,𝒥T,ijαβ(ij)\mathcal{J}_{N,ij}^{\alpha\beta},\mathcal{J}_{T,ij}^{\alpha\beta}(i\neq j) are given by

𝒥N,ijxx\displaystyle\mathcal{J}_{N,ij}^{xx} =kNξij,N3/2rijkN[32+ξij,Nrij]ξij,N1/2(nijx)2,\displaystyle=k_{N}\frac{\xi_{ij,N}^{3/2}}{r_{ij}}-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}(n_{ij}^{x})^{2}, (66)
𝒥T,ijxx\displaystyle\mathcal{J}_{T,ij}^{xx} kTξij,N1/2(nijy)2+12kTξij,N1/2ξij,Tnijxtijx,\displaystyle-k_{T}\xi_{ij,N}^{1/2}(n_{ij}^{y})^{2}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}t_{ij}^{x}, (67)
𝒥N,ijxy\displaystyle\mathcal{J}_{N,ij}^{xy} =kN[32+ξij,Nrij]ξij,N1/2nijxnijy,\displaystyle=-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}, (68)
𝒥T,ijxy\displaystyle\mathcal{J}_{T,ij}^{xy} =kTξij,N1/2nijxnijy+12kTξij,N1/2ξij,Tnijxtijy,\displaystyle=k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}t_{ij}^{y}, (69)
𝒥N,ijx\displaystyle\mathcal{J}_{N,ij}^{x\ell} =0,\displaystyle=0, (70)
𝒥T,ijx\displaystyle\mathcal{J}_{T,ij}^{x\ell} =kTξij,N1/2nijy+12kTξij,N1/2ξij,Tnijx(nijxtijynijytijx),\displaystyle=k_{T}\xi_{ij,N}^{1/2}n_{ij}^{y}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}(n_{ij}^{x}t_{ij}^{y}-n_{ij}^{y}t_{ij}^{x}), (71)
𝒥N,ijyx\displaystyle\mathcal{J}_{N,ij}^{yx} =kN[32+ξij,Nrij]ξij,N1/2nijxnijy,\displaystyle=-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}, (72)
𝒥T,ijyx\displaystyle\mathcal{J}_{T,ij}^{yx} =kTξij,N1/2nijxnijy+12kTξij,N1/2ξij,Tnijytijx,\displaystyle=k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}t_{ij}^{x}, (73)
𝒥N,ijyy\displaystyle\mathcal{J}_{N,ij}^{yy} =kNξij,N3/2rijkN[32+ξij,Nrij]ξij,N1/2(nijy)2,\displaystyle=k_{N}\frac{\xi_{ij,N}^{3/2}}{r_{ij}}-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}(n_{ij}^{y})^{2}, (74)
𝒥T,ijyy\displaystyle\mathcal{J}_{T,ij}^{yy} =kTξij,N1/2(nijx)2+12kTξij,N1/2ξij,Tnijytijy,\displaystyle=-k_{T}\xi_{ij,N}^{1/2}(n_{ij}^{x})^{2}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}t_{ij}^{y}, (75)
𝒥N,ijy\displaystyle\mathcal{J}_{N,ij}^{y\ell} =0,\displaystyle=0, (76)
𝒥T,ijy\displaystyle\mathcal{J}_{T,ij}^{y\ell} =kTξij,N1/2nijx+12kTξij,N1/2ξij,Tnijy(nijxtijynijytijx),\displaystyle=-k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}(n_{ij}^{x}t_{ij}^{y}-n_{ij}^{y}t_{ij}^{x}), (77)
𝒥N,ijx\displaystyle\mathcal{J}_{N,ij}^{\ell x} =0,\displaystyle=0, (78)
𝒥T,ijx\displaystyle\mathcal{J}_{T,ij}^{\ell x} =kTξij,N1/2nijy,\displaystyle=-k_{T}\xi_{ij,N}^{1/2}n_{ij}^{y}, (79)
𝒥N,ijy\displaystyle\mathcal{J}_{N,ij}^{\ell y} =0,\displaystyle=0, (80)
𝒥T,ijy\displaystyle\mathcal{J}_{T,ij}^{\ell y} =kTξij,N1/2nijx,\displaystyle=k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}, (81)
𝒥N,ij\displaystyle\mathcal{J}_{N,ij}^{\ell\ell} =0,\displaystyle=0, (82)
𝒥T,ij\displaystyle\mathcal{J}_{T,ij}^{\ell\ell} =kTξij,N1/2.\displaystyle=k_{T}\xi_{ij,N}^{1/2}. (83)

Similarly, the diagonal block elements 𝒥N,ijαβ,𝒥T,ijαβ(i=j)\mathcal{J}_{N,ij}^{\alpha\beta},\mathcal{J}_{T,ij}^{\alpha\beta}(i=j) are given by

𝒥N,iixx\displaystyle\mathcal{J}_{N,ii}^{xx} =ji{kNξij,N3/2rijkN[32+ξij,Nrij]ξij,N1/2(nijx)2},\displaystyle=-\sum_{j\neq i}\left\{k_{N}\frac{\xi_{ij,N}^{3/2}}{r_{ij}}-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}(n_{ij}^{x})^{2}\right\}, (84)
𝒥T,iixx\displaystyle\mathcal{J}_{T,ii}^{xx} =ji{kTξij,N1/2(nijy)2+12kTξij,N1/2ξij,Tnijxtijx},\displaystyle=-\sum_{j\neq i}\left\{-k_{T}\xi_{ij,N}^{1/2}(n_{ij}^{y})^{2}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}t_{ij}^{x}\right\}, (85)
𝒥N,iixy\displaystyle\mathcal{J}_{N,ii}^{xy} =ji{kN[32+ξij,Nrij]ξij,N1/2nijxnijy},\displaystyle=-\sum_{j\neq i}\left\{-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}\right\}, (86)
𝒥T,iixy\displaystyle\mathcal{J}_{T,ii}^{xy} =ji{kTξij,N1/2nijxnijy+12kTξij,N1/2ξij,Tnijxtijy},\displaystyle=-\sum_{j\neq i}\left\{k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}t_{ij}^{y}\right\}, (87)
𝒥N,iix\displaystyle\mathcal{J}_{N,ii}^{x\ell} =0,\displaystyle=0, (88)
𝒥T,iix\displaystyle\mathcal{J}_{T,ii}^{x\ell} =ji{kTξij,N1/2nijy+12ξij,N1/2ξij,Tnijx(nijxtijynijytijx)},\displaystyle=\sum_{j\neq i}\left\{k_{T}\xi_{ij,N}^{1/2}n_{ij}^{y}+\frac{1}{2}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{x}(n_{ij}^{x}t_{ij}^{y}-n_{ij}^{y}t_{ij}^{x})\right\}, (89)
𝒥N,iiyx\displaystyle\mathcal{J}_{N,ii}^{yx} =ji{kN[32+ξij,Nrij]ξij,N1/2nijxnijy},\displaystyle=-\sum_{j\neq i}\left\{-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}\right\}, (90)
𝒥T,iiyx\displaystyle\mathcal{J}_{T,ii}^{yx} =ji{kTξij,N1/2nijxnijy+12kTξij,N1/2ξij,Tnijytijx},\displaystyle=-\sum_{j\neq i}\left\{k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}n_{ij}^{y}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}t_{ij}^{x}\right\}, (91)
𝒥N,iiyy\displaystyle\mathcal{J}_{N,ii}^{yy} =ji{kNξij,N3/2rijkN[32+ξij,Nrij]ξij,N1/2(nijy)2},\displaystyle=-\sum_{j\neq i}\left\{k_{N}\frac{\xi_{ij,N}^{3/2}}{r_{ij}}-k_{N}\left[\frac{3}{2}+\frac{\xi_{ij,N}}{r_{ij}}\right]\xi_{ij,N}^{1/2}(n_{ij}^{y})^{2}\right\}, (92)
𝒥T,iiyy\displaystyle\mathcal{J}_{T,ii}^{yy} =ji{kTξij,N1/2(nijx)2+12kTξij,N1/2ξij,Tnijytijy},\displaystyle=-\sum_{j\neq i}\left\{-k_{T}\xi_{ij,N}^{1/2}(n_{ij}^{x})^{2}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}t_{ij}^{y}\right\}, (93)
𝒥N,iiyy\displaystyle\mathcal{J}_{N,ii}^{yy} =0,\displaystyle=0, (94)
𝒥T,iiy\displaystyle\mathcal{J}_{T,ii}^{y\ell} =ji{kTξij,N1/2nijx+12kTξij,N1/2ξij,Tnijy(nijxtijynijytijx)},\displaystyle=-\sum_{j\neq i}\left\{k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}+\frac{1}{2}k_{T}\xi_{ij,N}^{-1/2}\xi_{ij,T}n_{ij}^{y}(n_{ij}^{x}t_{ij}^{y}-n_{ij}^{y}t_{ij}^{x})\right\}, (95)
𝒥N,iix\displaystyle\mathcal{J}_{N,ii}^{\ell x} =0,\displaystyle=0, (96)
𝒥T,iix\displaystyle\mathcal{J}_{T,ii}^{\ell x} =jikTξij,N1/2nijy,\displaystyle=\sum_{j\neq i}k_{T}\xi_{ij,N}^{1/2}n_{ij}^{y}, (97)
𝒥N,iiy\displaystyle\mathcal{J}_{N,ii}^{\ell y} =0,\displaystyle=0, (98)
𝒥T,iiy\displaystyle\mathcal{J}_{T,ii}^{\ell y} =jikTξij,N1/2nijx,\displaystyle=-\sum_{j\neq i}k_{T}\xi_{ij,N}^{1/2}n_{ij}^{x}, (99)
𝒥N,ii\displaystyle\mathcal{J}_{N,ii}^{\ell\ell} =0,\displaystyle=0, (100)
𝒥T,ii\displaystyle\mathcal{J}_{T,ii}^{\ell\ell} =jikTξij,N1/2.\displaystyle=\sum_{j\neq i}k_{T}\xi_{ij,N}^{1/2}. (101)

Note that the terms proportional to ξij,T\xi_{ij,T} in 𝒥T\mathcal{J}_{T} include the history-dependent tangential displacements which are ignored in the effective potential Somfai07 ; Henkes10 ; Liu21 .

Appendix G The detailed derivation of GG in the Jacobian analysis

In this Appendix, we derive Eq. (38) that gives the rigidity. First, nonaffine displacements are expanded in terms of eigenfunctions of the Jacobian. Next, we express the rigidity as the eigenvalues and eigenfunctions of the Jacobian. Note that we adopt the abbreviation dA(𝒒FB(0))/dγ:=dA(𝒒(γ))/dγ|𝒒(γ)=𝒒FB(0)dA(\bm{q}^{\textrm{FB}}(0))/d\gamma:=dA(\bm{q}(\gamma))/d\gamma|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)} in this Appendix.

G.1 Expansion for nonaffine displacements via eigenfunction of Jacobian

At FB state, F~iα/dγ\tilde{F}_{i}^{\alpha}/d\gamma is expressed as

dF~iαdγ\displaystyle\frac{d\tilde{F}_{i}^{\alpha}}{d\gamma} =limΔγ0F~iα(𝒒FB(Δγ))F~iα(𝒒FB(0))Δγ\displaystyle=\lim_{\Delta\gamma\to 0}\frac{\tilde{F}_{i}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma))-\tilde{F}_{i}^{\alpha}(\bm{q}^{\textrm{FB}}(0))}{\Delta\gamma}
=ji[fijαqixyij(𝒒FB(0))+ζ=x,yfijαriζdr̊ijζ(𝒒FB(0))dγ+fijαi(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)].\displaystyle=\sum_{j\neq i}\left[\frac{\partial f_{ij}^{\alpha}}{\partial q_{i}^{x}}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}\frac{\partial f_{ij}^{\alpha}}{\partial r_{i}^{\zeta}}\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{\partial f_{ij}^{\alpha}}{\partial\ell_{i}}\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right]. (102)

Using the Jacobian, we rewrite Eq. (102) as

dF~iαdγ\displaystyle\frac{d\tilde{F}_{i}^{\alpha}}{d\gamma} =ji[𝒥jiαxyij(𝒒FB(0))+ζ=x,y𝒥jiαζdr̊ijζ(𝒒FB(0))dγ+𝒥jiα(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)],\displaystyle=-\sum_{j\neq i}\left[\mathcal{J}_{ji}^{\alpha x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}\mathcal{J}_{ji}^{\alpha\zeta}\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\mathcal{J}_{ji}^{\alpha\ell}\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right], (103)

where the first and second terms on the RHS represent the contributions from the affine and nonaffine displacements, respectively. Since the affine displacements are applied to our system instantaneously as a step strain, the integral interval of tangential displacements during the affine deformation are zero. Thus, only the normal contributions in the first term on RHS of Eq. (103) survive in the affine displacements. Then, we rewrite 𝒥ijαβ\mathcal{J}_{ij}^{\alpha\beta} as 𝒥N,ijαβ\mathcal{J}_{N,ij}^{\alpha\beta} in Eq. (103):

dF~iαdγ\displaystyle\frac{d\tilde{F}_{i}^{\alpha}}{d\gamma} =ji[𝒥N,jiαxyij(𝒒FB(0))+ζ=x,y𝒥jiαζdr̊ijζ(𝒒FB(0))dγ+𝒥jiα(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)].\displaystyle=-\sum_{j\neq i}\left[\mathcal{J}_{N,ji}^{\alpha x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}\mathcal{J}_{ji}^{\alpha\zeta}\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\mathcal{J}_{ji}^{\alpha\ell}\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right]. (104)

Introducing

|Ξi:=[ji𝒥N,jixxyijji𝒥N,jixyyijji𝒥N,jixyij]\displaystyle\ket{\Xi_{i}}:=\left[\begin{matrix}&\sum_{j\neq i}\mathcal{J}_{N,ji}^{xx}y_{ij}\\ &\sum_{j\neq i}\mathcal{J}_{N,ji}^{xy}y_{ij}\\ &\sum_{j\neq i}\mathcal{J}_{N,ji}^{x\ell}y_{ij}\end{matrix}\right] (105)

and with the aid of dF~iα/dγ=0d\tilde{F}_{i}^{\alpha}/d\gamma=0 at the FB state in Eq. (104), we obtain

Ξiα\displaystyle\Xi_{i}^{\alpha} =ji[ζ=x,y𝒥jiαζdr̊ijζdγ+𝒥jiα(d̊idγ+d̊jdγ)].\displaystyle=-\sum_{j\neq i}\left[\sum_{\zeta=x,y}\mathcal{J}_{ji}^{\alpha\zeta}\frac{d\mathring{r}_{ij}^{\zeta}}{d\gamma}+\mathcal{J}_{ji}^{\alpha\ell}\left(\frac{d\mathring{\ell}_{i}}{d\gamma}+\frac{d\mathring{\ell}_{j}}{d\gamma}\right)\right]. (106)

Since 𝒥\mathcal{J} satisfies 𝒥iiκβ=ji𝒥jiκβ\mathcal{J}_{ii}^{\kappa\beta}=-\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\beta}, we obtain

Ξiκ\displaystyle\Xi_{i}^{\kappa} =ζ=x,y[(ji𝒥jiκζ)dr̊iζdγji𝒥jiκζdr̊jζdγ][(ji𝒥jiκ)d̊idγ+ji𝒥jiκd̊jdγ]\displaystyle=-\sum_{\zeta=x,y}\left[\left(\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\zeta}\right)\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]-\left[\left(\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\ell}\right)\frac{d\mathring{\ell}_{i}}{d\gamma}+\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}\right]
=ζ=x,y[𝒥iiκζdr̊iζdγji𝒥jiκζdr̊jζdγ][𝒥iiκd̊idγ+ji𝒥jiκd̊jdγ]\displaystyle=-\sum_{\zeta=x,y}\left[-\mathcal{J}_{ii}^{\kappa\zeta}\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]-\left[-\mathcal{J}_{ii}^{\kappa\ell}\frac{d\mathring{\ell}_{i}}{d\gamma}+\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}\right]
=ζ=x,yj=1N𝒥jiκζdr̊jζdγ+𝒥iiκd̊idγji𝒥jiκd̊jdγ.\displaystyle=\sum_{\zeta=x,y}\sum_{j=1}^{N}\mathcal{J}_{ji}^{\kappa\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}+\mathcal{J}_{ii}^{\kappa\ell}\frac{d\mathring{\ell}_{i}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\kappa\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}. (107)

Since 𝒥\mathcal{J} satisfies 𝒥iiβ=ji𝒥jiβ\mathcal{J}_{ii}^{\ell\beta}=\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\beta}, we obtain

Ξi\displaystyle\Xi_{i}^{\ell} =ζ=x,y[(ji𝒥jiζ)dr̊iζdγji𝒥jiζdr̊jζdγ][(ji𝒥ji)d̊idγ+ji𝒥jid̊jdγ]\displaystyle=-\sum_{\zeta=x,y}\left[\left(\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\zeta}\right)\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]-\left[\left(\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\ell}\right)\frac{d\mathring{\ell}_{i}}{d\gamma}+\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}\right]
=ζ=x,y[𝒥iiζdr̊iζdγji𝒥jiζdr̊jζdγ]+[𝒥iid̊idγ+ji𝒥jid̊jdγ]\displaystyle=-\sum_{\zeta=x,y}\left[\mathcal{J}_{ii}^{\ell\zeta}\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]+\left[\mathcal{J}_{ii}^{\ell\ell}\frac{d\mathring{\ell}_{i}}{d\gamma}+\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}\right]
=ζ=x,y[𝒥iiζdr̊iζdγji𝒥jiζdr̊jζdγ]+j=1N𝒥jid̊jdγ.\displaystyle=-\sum_{\zeta=x,y}\left[\mathcal{J}_{ii}^{\ell\zeta}\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}-\sum_{j\neq i}\mathcal{J}_{ji}^{\ell\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]+\sum_{j=1}^{N}\mathcal{J}_{ji}^{\ell\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}. (108)

Let us introduce 𝒥~iiαβ\tilde{\mathcal{J}}_{ii}^{\alpha\beta} as

𝒥~iiαβ\displaystyle\tilde{\mathcal{J}}_{ii}^{\alpha\beta} :={𝒥iix(α=,β=x)𝒥iiy(α=,β=y)𝒥iiαβ(otherwise)\displaystyle:=\left\{\begin{matrix}-\mathcal{J}_{ii}^{\ell x}&(\alpha=\ell,\ \beta=x)\\ -\mathcal{J}_{ii}^{\ell y}&(\alpha=\ell,\ \beta=y)\\ \mathcal{J}_{ii}^{\alpha\beta}&(\textrm{otherwise})\end{matrix}\right. (109)

and

𝒥~ijαβ:=𝒥ijαβ.\displaystyle\tilde{\mathcal{J}}_{ij}^{\alpha\beta}:=\mathcal{J}_{ij}^{\alpha\beta}. (110)

Here, 𝒥~ij\tilde{\mathcal{J}}_{ij} satisfies

𝒥~jiαβ\displaystyle\tilde{\mathcal{J}}_{ji}^{\alpha\beta} ={𝒥ijx(α=x,β=)𝒥ijy(α=y,β=)𝒥ijαβ(otherwise).\displaystyle=\left\{\begin{matrix}-\mathcal{J}_{ij}^{x\ell}&(\alpha=x,\ \beta=\ell)\\ -\mathcal{J}_{ij}^{y\ell}&(\alpha=y,\ \beta=\ell)\\ \mathcal{J}_{ij}^{\alpha\beta}&(\textrm{otherwise}).\end{matrix}\right. (111)

With the aid of 𝒥~\tilde{\mathcal{J}} Eqs. (107) and (108) are rewritten as

Ξiα\displaystyle\Xi_{i}^{\alpha} =ζ=x,yj=1N𝒥~ijαζdr̊jζdγ+𝒥~iiαd̊idγ+ji𝒥~ijαd̊jdγ\displaystyle=\sum_{\zeta=x,y}\sum_{j=1}^{N}\tilde{\mathcal{J}}_{ij}^{\alpha\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}+\tilde{\mathcal{J}}_{ii}^{\alpha\ell}\frac{d\mathring{\ell}_{i}}{d\gamma}+\sum_{j\neq i}\tilde{\mathcal{J}}_{ij}^{\alpha\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}
=ζ=x,yj=1N𝒥~ijαζdr̊jζdγ+j=1N𝒥~ijαd̊jdγ\displaystyle=\sum_{\zeta=x,y}\sum_{j=1}^{N}\tilde{\mathcal{J}}_{ij}^{\alpha\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}+\sum_{j=1}^{N}\tilde{\mathcal{J}}_{ij}^{\alpha\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}
=β=x,y,j=1N𝒥~ijαβdq̊jβdγ,\displaystyle=\sum_{\beta=x,y,\ell}\sum_{j=1}^{N}\tilde{\mathcal{J}}_{ij}^{\alpha\beta}\frac{d\mathring{q}_{j}^{\beta}}{d\gamma}, (112)
Ξi\displaystyle\Xi_{i}^{\ell} =ζ=x,y[𝒥iiζdr̊iζdγ+ji𝒥ijζdr̊jζdγ]+j=1N𝒥jid̊jdγ\displaystyle=\sum_{\zeta=x,y}\left[\mathcal{J}_{ii}^{\ell\zeta}\frac{d\mathring{r}_{i}^{\zeta}}{d\gamma}+\sum_{j\neq i}\mathcal{J}_{ij}^{\ell\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}\right]+\sum_{j=1}^{N}\mathcal{J}_{ji}^{\ell\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}
=ζ=x,yj=1N𝒥ijζdr̊jζdγ+j=1N𝒥jid̊jdγ\displaystyle=\sum_{\zeta=x,y}\sum_{j=1}^{N}\mathcal{J}_{ij}^{\ell\zeta}\frac{d\mathring{r}_{j}^{\zeta}}{d\gamma}+\sum_{j=1}^{N}\mathcal{J}_{ji}^{\ell\ell}\frac{d\mathring{\ell}_{j}}{d\gamma}
=β=x,y,j=1N𝒥ijβdq̊jβdγ.\displaystyle=\sum_{\beta=x,y,\ell}\sum_{j=1}^{N}\mathcal{J}_{ij}^{\ell\beta}\frac{d\mathring{q}_{j}^{\beta}}{d\gamma}. (113)

Equations (112) and (113) can be rewritten as

Ξiα\displaystyle\Xi_{i}^{\alpha} =j=1Nβ=x,y,𝒥~ijαβdq̊jβdγ.\displaystyle=\sum_{j=1}^{N}\sum_{\beta=x,y,\ell}\tilde{\mathcal{J}}_{ij}^{\alpha\beta}\frac{d\mathring{q}_{j}^{\beta}}{d\gamma}. (114)

Furthermore, Eq. (114) can be expressed as

|Ξ=𝒥~|dq̊dγ,\displaystyle\Ket{\Xi}=\tilde{\mathcal{J}}\Ket{\frac{d\mathring{q}}{d\gamma}}, (115)

which corresponds to Eq. (34) in the main text, where |dq̊/dγ\Ket{d\mathring{q}/d\gamma} is introduced in Eq. (33).

Let us expand |dq̊/dγ\ket{d\mathring{q}/d\gamma} by the right eigenfunction |R~n\ket{\tilde{R}_{n}} of 𝒥~\tilde{\mathcal{J}} as

|dq̊dγ\displaystyle\Ket{\frac{d\mathring{q}}{d\gamma}} =an|R~n.\displaystyle=a_{n}\ket{\tilde{R}_{n}}. (116)

Substituting Eq. (116) into Eq. (115), we obtain

|Ξ\displaystyle\Ket{\Xi} =λ~nan|R~n.\displaystyle=\tilde{\lambda}_{n}a_{n}\ket{\tilde{R}_{n}}. (117)

Multiplying L~m|\bra{\tilde{L}_{m}} to Eq. (117) with the aid of the orthonormal relation, we obtain

L~m|Ξ\displaystyle\Braket{\tilde{L}_{m}}{\Xi} =λ~nanL~m|R~n\displaystyle=\tilde{\lambda}_{n}a_{n}\Braket{\tilde{L}_{m}}{\tilde{R}_{n}}
=λ~mam.\displaystyle=\tilde{\lambda}_{m}a_{m}. (118)

Substituting this into Eq. (116), we obtain Eq. (38).

G.2 The expression of GG

Let us evaluate the rigidity GG defined as Eq. (17). Substituting Eqs. (14) and (15) into Eq. (17), we obtain

G\displaystyle G =limΔγ012ΔγL2i,j(ij)[fijx(𝒒FB(Δγ))rijy(𝒒FB(Δγ))fijx(𝒒FB(0))rijy(𝒒FB(0))],\displaystyle=-\Braket{\lim_{\Delta\gamma\to 0}\frac{1}{2\Delta\gamma L^{2}}\sum_{i,j(i\neq j)}\left[f_{ij}^{x}(\bm{q}^{\textrm{FB}}(\Delta\gamma))r_{ij}^{y}(\bm{q}^{\textrm{FB}}(\Delta\gamma))-f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))r_{ij}^{y}(\bm{q}^{\textrm{FB}}(0))\right]}, (119)

where we have adopted the symmetric expression for ii and jj in the summation in Eq. (119).

Expanding rijα(𝒒FB(Δγ))r_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma)) in Eq. (119) by Δγ\Delta\gamma from the zero strain state, we obtain

rijα(𝒒FB(Δγ))\displaystyle r_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma)) =riα(𝒒FB(Δγ))rjα(𝒒FB(Δγ))\displaystyle=r_{i}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma))-r_{j}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma))
rijα(𝒒FB(0))+Δγ{δαx(yi(𝒒FB(0))yj(𝒒FB(0)))+dr̊iα(𝒒FB(0))dγdr̊jα(𝒒FB(0))dγ}\displaystyle\simeq r_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))+\Delta\gamma\left\{\delta_{\alpha x}\left(y_{i}(\bm{q}^{\textrm{FB}}(0))-y_{j}(\bm{q}^{\textrm{FB}}(0))\right)+\frac{d\mathring{r}_{i}^{\alpha}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}-\frac{d\mathring{r}_{j}^{\alpha}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right\}
=rijα(𝒒FB(0))+Δγ{δαxyij(𝒒FB(0))+dr̊ijα(𝒒FB(0))dγ}.\displaystyle=r_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))+\Delta\gamma\left\{\delta_{\alpha x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right\}. (120)

Similarly, expanding fijα(Δγ)f_{ij}^{\alpha}(\Delta\gamma) in Eq. (119) from the zero strain state, we obtain

fijα(𝒒FB(Δγ))\displaystyle f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma)) fijα(𝒒FB(0))+k=1Nζ=x,yΔγfijαrkζdrkζdγ+k=1NΔγfijαkdkdγ\displaystyle\simeq f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))+\sum_{k=1}^{N}\sum_{\zeta=x,y}\Delta\gamma\frac{\partial f_{ij}^{\alpha}}{\partial r_{k}^{\zeta}}\frac{dr_{k}^{\zeta}}{d\gamma}+\sum_{k=1}^{N}\Delta\gamma\frac{\partial f_{ij}^{\alpha}}{\partial\ell_{k}}\frac{d\ell_{k}}{d\gamma}
=fijα(𝒒FB(0))+ζ=x,yΔγ[fijαriζ(δζxyi(𝒒FB(0))+dr̊iζ(𝒒FB(0))dγ)+fijαrjζ(δζxyj(𝒒FB(0))+dr̊jζ(𝒒FB(0))dγ)]\displaystyle=f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}\Delta\gamma\left[\frac{\partial f_{ij}^{\alpha}}{\partial r_{i}^{\zeta}}\Biggl{(}\delta_{\zeta x}y_{i}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{i}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)+\frac{\partial f_{ij}^{\alpha}}{\partial r_{j}^{\zeta}}\left(\delta_{\zeta x}y_{j}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{j}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\Biggl{)}\right]
+Δγ[fijαi(δxyi(𝒒FB(0))+d̊i(𝒒FB(0))dγ)+fijαj(δxyj(𝒒FB(0))+d̊j(𝒒FB(0))dγ)].\displaystyle\quad+\Delta\gamma\left[\frac{\partial f_{ij}^{\alpha}}{\partial\ell_{i}}\Biggl{(}\delta_{\ell x}y_{i}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)+\frac{\partial f_{ij}^{\alpha}}{\partial\ell_{j}}\left(\delta_{\ell x}y_{j}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\Biggl{)}\right]. (121)

Furthermore, using fijα/rjζ=fijα/riζ{\partial f_{ij}^{\alpha}}/{\partial r_{j}^{\zeta}}=-{\partial f_{ij}^{\alpha}}/{\partial r_{i}^{\zeta}} and fijα/j=fijα/i{\partial f_{ij}^{\alpha}}/{\partial\ell_{j}}={\partial f_{ij}^{\alpha}}/{\partial\ell_{i}}, fijαf_{ij}^{\alpha} can be written as

fijα(𝒒FB(Δγ))\displaystyle f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(\Delta\gamma)) =fijα(𝒒FB(0))+ζ=x,yΔγfijαriζ(δζxyij(𝒒FB(0))+dr̊ijζ(𝒒FB(0))dγ)+Δγfijαi(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ).\displaystyle=f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}\Delta\gamma\frac{\partial f_{ij}^{\alpha}}{\partial r_{i}^{\zeta}}\left(\delta_{\zeta x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)+\Delta\gamma\frac{\partial f_{ij}^{\alpha}}{\partial\ell_{i}}\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right). (122)

Substituting Eqs. (120) and (122) into Eq. (119), we obtain

G\displaystyle G =12L2i,j(ij)[fijx(𝒒FB(0))dq̊ijy(𝒒FB(0))dγ+ζ=x,yfijx(𝒒FB(0))riζrijy(𝒒FB(0))(δζxyij(𝒒FB(0))+dr̊ijζ(𝒒FB(0))dγ)\displaystyle=-\frac{1}{2L^{2}}\left\langle\sum_{i,j(i\neq j)}\Biggl{[}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))\frac{d\mathring{q}_{ij}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\sum_{\zeta=x,y}\frac{\partial f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))}{\partial r_{i}^{\zeta}}r_{ij}^{y}(\bm{q}^{\textrm{FB}}(0))\left(\delta_{\zeta x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right.
+fijx(𝒒FB(0))irijy(𝒒FB(0))(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)].\displaystyle\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad+\left.\frac{\partial f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))}{\partial\ell_{i}}r_{ij}^{y}(\bm{q}^{\textrm{FB}}(0))\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\Biggl{]}\right\rangle. (123)

Because i(ij)fijα(𝒒FB(0))=0\sum_{i(i\neq j)}f_{ij}^{\alpha}(\bm{q}^{\textrm{FB}}(0))=0 at the FB state, the first term on RHS of Eq. (123) can be written as

i,j(ij)fijx(𝒒FB(0))dq̊ijy(𝒒FB(0))dγ\displaystyle\sum_{i,j(i\neq j)}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))\frac{d\mathring{q}_{ij}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma} =i,j(ij)fijx(𝒒FB(0))(dq̊iy(𝒒FB(0))dγdq̊jy(𝒒FB(0))dγ)\displaystyle=\sum_{i,j(i\neq j)}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))\left(\frac{d\mathring{q}_{i}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}-\frac{d\mathring{q}_{j}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)
=j(j(ji)fijx(𝒒FB(0)))dq̊iy(𝒒FB(0))dγi(i(ij)fijx(𝒒FB(0)))dq̊jy(𝒒FB(0))dγ\displaystyle=\sum_{j}\left(\sum_{j(j\neq i)}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))\right)\frac{d\mathring{q}_{i}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}-\sum_{i}\left(\sum_{i(i\neq j)}f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))\right)\frac{d\mathring{q}_{j}^{y}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}
=0.\displaystyle=0. (124)

Thus, GG is expressed as

G\displaystyle G =12L2i,j(ij)[ζ=x,yfijx(𝒒FB(0))riζyij(𝒒FB(0))(δζxyij(𝒒FB(0))+dr̊ijζ(𝒒FB(0))dγ)\displaystyle=-\frac{1}{2L^{2}}\left\langle\sum_{i,j(i\neq j)}\left[\sum_{\zeta=x,y}\frac{\partial f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))}{\partial r_{i}^{\zeta}}y_{ij}(\bm{q}^{\textrm{FB}}(0))\left(\delta_{\zeta x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right.\right.
+fijx(𝒒FB(0))iyij(𝒒FB(0))(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)].\displaystyle\quad\quad\quad\quad\quad\quad\quad\quad\left.\left.+\frac{\partial f_{ij}^{x}(\bm{q}^{\textrm{FB}}(0))}{\partial\ell_{i}}y_{ij}(\bm{q}^{\textrm{FB}}(0))\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right]\right\rangle. (125)

With the aid of 𝒥ijαβ:=qjβfijα(ij)\mathcal{J}_{ij}^{\alpha\beta}:=-\partial_{q_{j}^{\beta}}f_{ij}^{\alpha}\ (i\neq j), we can express GG as

G\displaystyle G =12L2i,j(ij)[ζ=x,yyij(𝒒FB(0))𝒥jixζ(𝒒FB(0))(δζxyij(𝒒FB(0))+dr̊ijζ(𝒒FB(0))dγ)\displaystyle=\frac{1}{2L^{2}}\left\langle\sum_{i,j(i\neq j)}\left[\sum_{\zeta=x,y}y_{ij}(\bm{q}^{\textrm{FB}}(0))\mathcal{J}_{ji}^{x\zeta}(\bm{q}^{\textrm{FB}}(0))\left(\delta_{\zeta x}y_{ij}(\bm{q}^{\textrm{FB}}(0))+\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right.\right.
+yij(𝒒FB(0))𝒥jix(𝒒FB(0))(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)]\displaystyle\quad\quad\quad\quad\quad\quad\quad\left.\left.+y_{ij}(\bm{q}^{\textrm{FB}}(0))\mathcal{J}_{ji}^{x\ell}(\bm{q}^{\textrm{FB}}(0))\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right]\right\rangle
=12L2i,j(ij)[yij2(𝒒FB(0))𝒥jixx(𝒒FB(0))+ζ=x,yyij(𝒒FB(0))𝒥jixζ(𝒒FB(0))dr̊ijζ(𝒒FB(0))dγ\displaystyle=\frac{1}{2L^{2}}\left\langle\sum_{i,j(i\neq j)}\left[y_{ij}^{2}(\bm{q}^{\textrm{FB}}(0))\mathcal{J}_{ji}^{xx}(\bm{q}^{\textrm{FB}}(0))+\sum_{\zeta=x,y}y_{ij}(\bm{q}^{\textrm{FB}}(0))\mathcal{J}_{ji}^{x\zeta}(\bm{q}^{\textrm{FB}}(0))\frac{d\mathring{r}_{ij}^{\zeta}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right.\right.
+yij(𝒒FB(0))𝒥jix(𝒒FB(0))(d̊i(𝒒FB(0))dγ+d̊j(𝒒FB(0))dγ)].\displaystyle\quad\quad\quad\quad\quad\quad\quad\left.\left.+y_{ij}(\bm{q}^{\textrm{FB}}(0))\mathcal{J}_{ji}^{x\ell}(\bm{q}^{\textrm{FB}}(0))\left(\frac{d\mathring{\ell}_{i}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}+\frac{d\mathring{\ell}_{j}(\bm{q}^{\textrm{FB}}(0))}{d\gamma}\right)\right]\right\rangle. (126)

Thus, with Eqs (36) and (37), we obtain Eqs. (39)-(41).

Appendix H DOS in terms of the effective Hessian

In this Appendix, we introduce the DOS with the aid of the effective Hessian as in Refs. Somfai07 ; Henkes10 ; Liu21 . The effective Hessian \mathcal{H} at the FB state is defined as

ijαβ:=2Ueffqiαqjβ|𝒒(γ)=𝒒FB(0),\displaystyle\mathcal{H}_{ij}^{\alpha\beta}:=\left.\frac{\partial^{2}U_{\textrm{eff}}}{\partial q_{i}^{\alpha}\partial q_{j}^{\beta}}\right|_{\bm{q}(\gamma)=\bm{q}^{\textrm{FB}}(0)}, (127)

where UeffU_{\textrm{eff}} is the effective potential defined as

Ueff:=12ij[kN(δ𝒓ij𝒏ij)2|𝒇N,ij|rijFB(δ𝒓ij𝒕ij)2+kTδtij2]\displaystyle U_{\textrm{eff}}:=\frac{1}{2}\sum_{\braket{ij}}\left[k_{N}(\delta\bm{r}_{ij}\cdot\bm{n}_{ij})^{2}-\frac{|\bm{f}_{N,ij}|}{r_{ij}^{\textrm{FB}}}(\delta\bm{r}_{ij}\cdot\bm{t}_{ij})^{2}+k_{T}\delta t_{ij}^{2}\right] (128)

with δ𝒓ij:=δ𝒓iδ𝒓j,δ𝒓i:=𝒓i𝒓iFB,rijFB:=|𝒓iFB𝒓jFB|,δtij:=δ𝒓ij𝒕ij(δi+δj)\delta\bm{r}_{ij}:=\delta\bm{r}_{i}-\delta\bm{r}_{j},\delta\bm{r}_{i}:=\bm{r}_{i}-\bm{r}_{i}^{\textrm{FB}},r_{ij}^{\textrm{FB}}:=|\bm{r}^{\textrm{FB}}_{i}-\bm{r}_{j}^{\textrm{FB}}|,\delta t_{ij}:=\delta\bm{r}_{ij}\cdot\bm{t}_{ij}-(\delta\ell_{i}+\delta\ell_{j}), and δi:=iiFB\delta\ell_{i}:=\ell_{i}-\ell_{i}^{\textrm{FB}}. Here, 𝒓iFB\bm{r}_{i}^{\textrm{FB}} and iFB\ell_{i}^{\textrm{FB}} are the position of ii-th particle and 33rd component of 𝒒i\bm{q}_{i} at the FB state, respectively. Thus, \mathcal{H} is a 3N×3N3N\times 3N matrix corresponding to the Jacobian. We note that this Hessian matrix is a real symmetric matrix, and thus, it can be diagonalized by an orthogonal matrix, where the eigenvectors are orthogonal with each other, and the corresponding eigenvalues are real number.

The eigenvalue equation of \mathcal{H} is expressed as

|n=λH,n|n,\displaystyle\mathcal{H}\ket{n}=\lambda_{H,n}\ket{n}, (129)

where λH,n\lambda_{H,n} and |n\ket{n} are the nn-th eigenvalue and eigenvector of \mathcal{H}, respectively. Note that the left eigenvalue is also given by n|=λH,nn|\langle n|\mathcal{H}=\lambda_{H,n}\langle n|, where n|=|nT\langle n|=|n\rangle^{T}. Then, we introduce the DOS DHD_{H} in terms of \mathcal{H} as

DH(ω):=13Nnδ(ωωH,n),\displaystyle D_{H}(\omega):=\frac{1}{3N}\sum_{n}\nolimits^{\prime}\langle\delta(\omega-\omega_{H,n})\rangle, (130)

where ωH,n:=λH,n\omega_{H,n}:=\sqrt{\lambda_{H,n}}.

Appendix I System size dependence for the DOS

The system size dependence of the DOS is investigated in this Appendix. From Fig. 15 we have confirmed that both the rotational and translational bands show little dependence on system size. This means that the rotational band is not a virtual band that can only be observed in a small system, but an intrinsic band that can be observed in the thermodynamic limit. Thus, we expect that our observed results will remain unchanged even if we are interested in larger systems.

Refer to caption
Figure 15: Double logarithmic plots of D(ω)D(\omega) against ωt0\omega t_{0} for various NN at kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8}, ϕ=0.90\phi=0.90.

Appendix J Density dependence for DOS

In this Appendix, we investigate the density dependence for the DOS. As shown in Fig. 16 for kT/kN=108k_{T}/k_{N}=10^{-8}, the DOS depends on ϕ\phi, where the rotational band shifts to a lower frequency region and the plateau of the translational band becomes longer as the density approaches the jamming point. The latter result is well known from previous studies such as Ref. Wyart05 .

Refer to caption
Figure 16: Double logarithmic plots of D(ω)D(\omega) against ωt0\omega t_{0} for various ϕ\phi at kT/kN=1.0×108k_{T}/k_{N}=1.0\times 10^{-8} and N=4096N=4096.

References