betaFeedback

WorksPhysics and Astronomy

New approach to approximate analytical solutions of a harmonic oscillator with weak to moderate nonlinear damping: Part I

Abstract

We introduce a new approach to deriving approximate analytical solutions of a harmonic oscillator damped by purely nonlinear, or combinations of linear and nonlinear damping forces. Our approach is based on choosing a suitable trial solution, i.e. an ansatz, which is the product of the time-dependent amplitude and the oscillatory (trigonometric) function that has the same frequency but different initial phase, compared to the undamped case. We derive the equation for the amplitude decay using the connection of the energy dissipation rate with the power of the total damping force and the approximation that the amplitude changes slowly over time compared to the oscillating part of the ansatz. By matching our ansatz to the initial conditions, we obtain the equations for the corresponding initial amplitude and initial phase. Here we demonstrate the validity of our approach in the case of damping quadratic in velocity, Coulomb damping, and a combination of the two, i.e. in this paper we consider purely nonlinear damping, while the dynamics with combinations of damping linear in velocity and nonlinear damping will be analyzed in a follow-up paper. In the case of damping quadratic in velocity, by comparing our approximate analytical solutions with the corresponding numerical solutions, we find that our solutions excellently describe the dynamics of the oscillator in the regime of weak to moderately strong quadratic damping. In the case of Coulomb damping, as well as in the case of a combined Coulomb and quadratic damping, our approximate analytical solutions agree well with the corresponding numerical solutions until the last few half-periods of the motion. Therefore, for these two cases, we introduce improved variants of our approximate solutions which describe the dynamics well until the very end.

Introduction

In physics and engineering, the decrease in the amplitude of free harmonic oscillations is usually attributed to one of the three different types of damping: damping linear in velocity, damping quadratic in velocity, and sliding friction. In the case of a harmonic oscillator damped by a force linear in velocity (viscous damping), the corresponding equation of motion can be solved exactly and the solutions can be easily analyzed 1. In case of a harmonic oscillator damped with sliding friction (Coulomb damping), the corresponding equation of motion is nonlinear because the friction force reverses direction with velocity, which is modeled via the signum function, but it can still be solved exactly by splitting the motion into left and right moving segments 2, i.e. the motion needs to be analyzed over half-periods and the solution thus obtained is piecewise continuous. In the case of a harmonic oscillator with damping quadratic in velocity, the corresponding equation of motion is a nonlinear differential equation that cannot be solved analytically and approximate or numerical methods must be used [^7, 9–11]. Therefore, even for a harmonic oscillator with only one type of damping present, the analysis of the dynamics can be challenging if the damping is nonlinear, i.e. Coulomb damping or damping quadratic in velocity.

Although the damped harmonic oscillator model, with a single type of damping, can successfully describe some simple real-world systems, such as physical pendulums 3, block-spring systems 4, RLC circuits 5, etc., for a more complete and precise description, it is necessary to take into account that oscillations are damped simultaneously by two or all three types of damping. For example, in 6 a simple physical pendulum was studied, both experimentally and theoretically, and it was shown that the contribution of all three types of damping must be taken into account

[^7, 9–11]: [7, 9, 10, 11] for an adequate general description of pendulum damping, while in some specific cases two types of damping were sufficient.

In a recent paper [14], a harmonic oscillator damped simultaneously by all three types of damping was studied and approximate analytical solutions were found that agree excellently with numerical solutions, but only in the weak damping regime. The approach used in [14] is based on ansatz of the form x~(t)=A0f~(t)cos⁡(ω0t+φ0)\tilde{x}(t)=A_0\tilde{f}(t)\cos(\omega_0t+\varphi_0), where ω0\omega_0, A0A_0, and φ0\varphi_0 are the angular frequency, amplitude, and initial phase of the corresponding undamped solution, while f~(t)\tilde{f}(t) is the unknown function that describes the decay of the amplitude over time. The connection of the energy dissipation rate and the power of the total damping force was used to determine the unknown function f~(t)\tilde{f}(t) [14]. More precisely, in the case of weak damping, one can take that f~(t)\tilde{f}(t) remains approximately constant over half cycles (half periods) and average the oscillatory parts of the energy dissipation rate to obtain the first-order differential equation for f~(t)\tilde{f}(t). The same approach was previously used to determine the amplitude decay in the case of weak damping by each of the three types of damping individually [11], by combination of viscous damping and damping quadratic in velocity [15], and by combination of Coulomb damping and viscous damping [16]. The advantage of this approach is mathematical simplicity, but, of course, with the increase in the strength of the damping the values of angular frequency, initial phase, and (initial) amplitude can change significantly compared to the undamped case.

The analysis of the influence of the damping strength on the initial phase and frequency carried out in [17] for viscous damping and carried out in [14] for damping quadratic in velocity clearly shows that better approximate analytical solutions can be obtained already if we simply take that the initial phase and the initial amplitude of the approximate analytical solution differ from φ0\varphi_0 and A0A_0, but still keep ω0\omega_0 as its angular frequency. Thus, here we use an ansatz of the form x(t)=Af(t)cos⁡(ω0t+φ)x(t)=Af(t)\cos(\omega_0t+\varphi) as the basis of our approach. In addition to the unknown function f(t)f(t), we now have to determine the constants AA and φ\varphi, i.e. we have a somewhat more complex mathematical problem compared to the approach used in [14], but, as we will show, we obtain new approximate analytical solutions that describe the dynamics significantly better and are valid for a wider range of values of the corresponding damping constants than the solutions obtained in [14].

This paper is organized into five sections. In § 2, we consider the case of harmonic oscillator damped simultaneously by all three types of damping, introduce the theory and approximations we use and derive the first-order differential equation for the amplitude decay, the equation for the initial amplitude, and the equation for the initial phase. In § 3, we obtain new approximate analytical solutions for damping quadratic in velocity. In § 4, we obtain new approximate analytical solutions in the case of Coulomb damping. In § 6, we obtain new approximate analytical solutions for combined Coulomb and quadratic damping. In each of § 3, § 4, and § 6 we compare new solutions to the corresponding solutions obtained by the approach used in [14] and numerical solutions. We note here for clarity that throughout the paper new solutions are designated with x(t)x(t), corresponding energies with E(t)E(t), and both are shown as solid red curves in all figures, solutions obtained by the approach used in [14] are designated with x~(t)\tilde{x}(t), corresponding energies with E~(t)\tilde{E}(t), and both are shown as solid blue curves in all figures, while numerical solutions and numerically obtained energies are shown as dotted black curves in all figures. At the end of each of § 3, § 4, and § 6 we discuss the results for the type of damping considered. In Section VI, we summarize the important findings of the paper and provide an outlook for the follow-up paper that will cover the cases of combined linear and nonlinear damping.

New equations for the amplitude decay, initial amplitude, and initial phase

We consider the paradigmatic example of a damped harmonic oscillator shown in Figure 1, i.e. a block of mass mm that oscillates back and forth along a horizontal surface under the influence of the restoring force of an ideal spring Fres(t)=−kx(t)F_{\mathrm{res}}(t)=-kx(t), where kk is the stiffness of the spring and x(t)x(t) is the displacement of the block from the equilibrium position (set to x=0x=0), and under the influence of the damping force

Schematic representation of a block-spring system with a restoring force and a total damping force

Figure 1. Schematic representation of a block-spring system with a restoring force Fres(t)F_{\mathrm{res}}(t) and a (total) damping force Fd(t)F_d(t). Here we show the time instant at which x(t)>0x(t)>0 and v(t)<0v(t)<0, i.e. at which Fres(t)<0F_{\mathrm{res}}(t)<0 and Fd(t)>0F_d(t)>0. See text for details.

Fd(t)=−sgn⁡[v(t)]μmg−bv(t)−Dv(t)∣v(t)∣(1)F_d(t)=-\operatorname{sgn}[v(t)]\mu mg-bv(t)-Dv(t)|v(t)| \tag*{(1)}

where μ>0\mu>0, b>0b>0 and D>0D>0 are the corresponding damping constants, v(t)=dx(t)/dtv(t)=dx(t)/dt is the velocity, and

sgn⁡[v(t)]={1if v(t)>00if v(t)=0−1if v(t)<0(2)\operatorname{sgn}[v(t)] = \begin{cases} 1 & \text{if } v(t)>0 \\ 0 & \text{if } v(t)=0 \\ -1 & \text{if } v(t)<0 \end{cases} \tag*{(2)}

is the sign function. The corresponding equation of motion is

mx¨(t)=−sgn⁡[v(t)]μmg−bv(t)−Dv(t)∣v(t)∣−kx(t)(3)m\ddot{x}(t)=-\operatorname{sgn}[v(t)]\mu mg-bv(t)-Dv(t)|v(t)|-kx(t) \tag*{(3)}

where a(t)=d2x(t)/dt2a(t)=d^2x(t)/dt^2 is the acceleration of the block. The first term on the right-hand side of equation (3) models the force of sliding friction [5, 6, 7, 18, 19], while the second and third terms model, e.g., the influence of air resistance, since air resistance generally depends on terms proportional to velocity and square of velocity [11, 13, 15, 20]. For simplicity, here we assume that the static and dynamic coefficients of friction are the same, i.e. both are equal to μ\mu. The energy (potential plus kinetic) of the block-spring system is given by

E(t)=kx2(t)2+mv2(t)2.(4)E(t)=\frac{kx^2(t)}{2}+\frac{mv^2(t)}{2}. \tag*{(4)}

If we put μ=b=D=0\mu=b=D=0 in (3), i.e. for Fd(t)=0F_d(t)=0, we get the equation of the undamped harmonic oscillator [3], with general solution

xHO(t)=A0cos⁡(ω0t+φ0),(5)x_{HO}(t)=A_0\cos(\omega_0t+\varphi_0), \tag*{(5)}

where ω0=k/m\omega_0=\sqrt{k/m} is the angular frequency of the undamped system, while A0A_0 and φ0\varphi_0 are constants (amplitude and initial phase) that are determined from initial conditions (x0,v0)(x_0,v_0), i.e.

A0=x02+(v0ω0)2,φ0=arctan⁡(−v0ω0x0).(6)A_0=\sqrt{x_0^2+\left(\frac{v_0}{\omega_0}\right)^2},\qquad\varphi_0=\arctan\left(-\frac{v_0}{\omega_0x_0}\right). \tag*{(6)}

The undamped system oscillates with conserved, i.e. constant, energy [3]. For damped systems, i.e. for Fd(t)≠0F_d(t)\ne0, the energy is not conserved due to the power of the damping force Pd(t)=Fd(t)v(t)P_d(t)=F_d(t)v(t), and the energy dissipation rate is given by dE(t)/dt=Pd(t)dE(t)/dt=P_d(t), i.e.

dE(t)dt=−μmg∣v(t)∣−bv2(t)−D∣v(t)∣3,(7)\frac{dE(t)}{dt}=-\mu mg|v(t)|-bv^2(t)-D|v(t)|^3, \tag*{(7)}

where we used sgn⁡[v(t)]v(t)=∣v(t)∣\operatorname{sgn}[v(t)]v(t)=|v(t)| and v2(t)∣v(t)∣=∣v(t)∣3v^2(t)|v(t)|=|v(t)|^3. Thus, equation (7) tells us that the energy of the damped system decreases monotonically with time, i.e. dE(t)/dt≤0dE(t)/dt\le0 for all t≥0t\ge0, and dE(t)/dt=0dE(t)/dt=0 holds at the turning points, i.e. at instants when v(t)=0v(t)=0.

In a recent paper [14], approximate analytical solutions of harmonic oscillator weakly damped by a force (1) were derived starting from the simple ansatz

x~(t)=A0f~(t)cos⁡(ω0t+φ0),(8)\tilde{x}(t)=A_0\tilde{f}(t)\cos(\omega_0t+\varphi_0), \tag*{(8)}

where A0A_0, ω0\omega_0 and φ0\varphi_0 are the same as in the undamped case, i.e. as in (5), while the unknown function f~(t)\tilde{f}(t) describes the slow decay of oscillations with time. By slow decay we mean that the rate of change of the function f~(t)\tilde{f}(t) over time is significantly slower than the rate of change of the function cos⁡(ω0t+φ0)\cos(\omega_0t+\varphi_0). The approach in [14] is further based on taking

v~(t)=−ω0A0f~(t)sin⁡(ω0t+φ0),(9)\tilde{v}(t)=-\omega_0A_0\tilde{f}(t)\sin(\omega_0t+\varphi_0), \tag*{(9)}

for velocity, i.e. in neglecting the term with df~(t)/dtd\tilde{f}(t)/dt in dx~(t)/dtd\tilde{x}(t)/dt, which leads to energy of the form

E~(t)=mω02A02f~2(t)2.(10)\tilde{E}(t)=\frac{m\omega_0^2 A_0^2\tilde{f}^2(t)}{2}. \tag*{(10)}

These steps are justified in the case of weak damping, i.e. when the frequency and initial phase remain approximately the same as in the undamped case and ∣df~(t)/dt∣ω0−1≪1|d\tilde{f}(t)/dt|\omega_0^{-1}\ll1 holds for all t≥0t\geq0 [14]. Simple first order differential equation for the function f~(t)\tilde{f}(t) is obtained by inserting velocity (9) and energy (10) into the energy dissipation rate (7) and taking the time average of trigonometric functions over half periods [14]. It was shown (for a particular choice of mm, kk and A0A_0) that the approximate analytical solutions obtained by this approach are in excellent agreement with the numerical solutions for a certain limited range of values of the damping constants (μ\mu, bb and DD) and that this agreement also depends on the type of initial conditions considered (i.e. on φ0\varphi_0) [14]. In particular, the agreement of approximate analytical solutions of the form (8) with numerical solutions is better in the case of initial conditions (x0=0x_0=0, v0≠0v_0\ne0) with purely kinetic initial energy than in the case of initial conditions (x0≠0x_0\ne0, v0=0v_0=0) with purely potential initial energy [14].

Here we improve the approach used in [14] by starting from the ansatz of the form

x(t)=Af(t)cos⁡(ω0t+φ),(11)x(t)=Af(t)\cos(\omega_0t+\varphi), \tag*{(11)}

where A≠A0A\ne A_0 and φ≠φ0\varphi\ne\varphi_0 are unknown constants and f(t)f(t) is the unknown function that describes the amplitude decay over time. The corresponding velocity is

v(t)=dx(t)dt=−ω0Af(t)sin⁡(ω0t+φ)+Adf(t)dtcos⁡(ω0t+φ).(12)v(t)=\frac{dx(t)}{dt}=-\omega_0Af(t)\sin(\omega_0t+\varphi)+A\frac{df(t)}{dt}\cos(\omega_0t+\varphi). \tag*{(12)}

Although we have not yet determined the function f(t)f(t), without loss of generality, we can take f(0)=1f(0)=1, and using initial conditions (x(0)=x0x(0)=x_0, v(0)=v0v(0)=v_0) we obtain equations

x0=Acos⁡(φ)(13)x_0=A\cos(\varphi) \tag*{(13)}

and

v0=−ω0Asin⁡(φ)+Adf(t)dt∣t=0cos⁡(φ)(14)v_0=-\omega_0A\sin(\varphi)+A\left.\frac{df(t)}{dt}\right|_{t=0}\cos(\varphi) \tag*{(14)}

from which we will determine the constants AA and φ\varphi once we determine df(t)/dtdf(t)/dt at t=0t=0. The energy corresponding to (11) and (12) is

E(t)=mω02A22(f2(t)−f(t)df(t)dtω0−1sin⁡(2ω0t+2φ)+(df(t)dt)2ω0−2cos⁡2(ω0t+φ))(15)E(t)=\frac{m\omega_0^2A^2}{2}\left(f^2(t)-f(t)\frac{df(t)}{dt}\omega_0^{-1}\sin(2\omega_0t+2\varphi)+\left(\frac{df(t)}{dt}\right)^2\omega_0^{-2}\cos^2(\omega_0t+\varphi)\right) \tag*{(15)}

We note here that keeping the terms of order (df(t)/dt)ω0−1(df(t)/dt)\omega_0^{-1} and (df(t)/dt)2ω0−2(df(t)/dt)^2\omega_0^{-2} in velocity (12) and energy (15) is necessary in modeling the influence of damping on AA and φ\varphi and also to ensure that our modeled displacement (11), velocity (12), and energy (15) can be matched to the initial conditions, but in modeling the function f(t)f(t) itself the simplest way is to proceed similarly as in [14], i.e. we approximate the left hand side of the energy dissipation rate (7) by

dE(t)dt≈ddt(mω02A2f2(t)2)=mω02A2f(t)df(t)dt(16)\frac{dE(t)}{dt}\approx\frac{d}{dt}\left(\frac{m\omega_0^2A^2f^2(t)}{2}\right)=m\omega_0^2A^2f(t)\frac{df(t)}{dt} \tag*{(16)}

and for the velocity appearing on the right hand side of (7) we use

v(t)≈−ω0Af(t)sin⁡(ω0t+φ).(17)v(t)\approx-\omega_0Af(t)\sin(\omega_0t+\varphi). \tag*{(17)}

Thus, using (16) and (17) in (7) we get

df(t)dt=−μgω0A∣sin⁡(ω0t+φ)∣−bmf(t)sin⁡2(ω0t+φ)−Dω0Amf2(t)∣sin⁡(ω0t+φ)∣3.(18)\frac{df(t)}{dt}=-\frac{\mu g}{\omega_0A}|\sin(\omega_0t+\varphi)|-\frac{b}{m}f(t)\sin^2(\omega_0t+\varphi)-\frac{D\omega_0A}{m}f^2(t)|\sin(\omega_0t+\varphi)|^3. \tag*{(18)}

Since, by the assumption of weak damping, the rate of change of f(t)f(t) is significantly slower than the rate of change of trigonometric functions over time intervals of the order ΔT=T0=2π/ω0\Delta T=T_0=2\pi/\omega_0, we can make an additional approximation in equation (18) by averaging the trigonometric functions over time intervals ΔT/2\Delta T/2, i.e.

df(t)dt=−μgω0A⟨∣sin⁡(ω0t+φ)∣⟩−bmf(t)⟨sin⁡2(ω0t+φ)⟩−Dω0Amf2(t)⟨∣sin⁡(ω0t+φ)∣3⟩,(19)\frac{df(t)}{dt}=-\frac{\mu g}{\omega_0A}\langle|\sin(\omega_0t+\varphi)|\rangle-\frac{b}{m}f(t)\langle\sin^2(\omega_0t+\varphi)\rangle-\frac{D\omega_0A}{m}f^2(t)\langle|\sin(\omega_0t+\varphi)|^3\rangle, \tag*{(19)}

where we use the notation ⟨F(t)⟩=(T0/2)−1∫tt+T0/2F(t′)dt′\langle F(t)\rangle= (T_0/2)^{-1}\int_{t}^{t+T_0/2} F(t')dt'. Since ⟨∣sin⁡(ω0t+φ)∣⟩=2/π\langle|\sin(\omega_0t+\varphi)|\rangle= 2/\pi, ⟨sin⁡2(ω0t+φ)⟩=1/2\langle\sin^2(\omega_0t+\varphi)\rangle= 1/2, and ⟨∣sin⁡(ω0t+φ)∣3⟩=4/(3π)\langle|\sin(\omega_0t+\varphi)|^3\rangle= 4/(3\pi) we get

df(t)dt=−(d2Af2(t)+d1f(t)+d0A)(20)\frac{df(t)}{dt}=-\left(d_2Af^2(t)+d_1f(t)+\frac{d_0}{A}\right) \tag*{(20)}

where

d0=2μgπω0,d1=b2m,d2=4Dω03πm,(21)d_0=\frac{2\mu g}{\pi\omega_0},\quad d_1=\frac{b}{2m},\quad d_2=\frac{4D\omega_0}{3\pi m}, \tag*{(21)}

as the final differential equation for the function f(t)f(t). We are now in a position to write the final equations from which we determine the constants AA and φ\varphi. Combining (13) and (14), along with (20), we get

v0+(d2A+d1+d0A)x0=−ω0A2−x02(22)v_0+\left(d_2A+d_1+\frac{d_0}{A}\right)x_0=-\omega_0\sqrt{A^2-x_0^2} \tag*{(22)}

as the equation from which we determine AA, and

φ=arctan⁡(−v0+(d2A+d1+d0A)x0ω0x0)(23)\varphi=\arctan\left(-\frac{v_0+\left(d_2A+d_1+\frac{d_0}{A}\right)x_0}{\omega_0x_0}\right) \tag*{(23)}

as the equation from which we can calculate the initial phase (once we determined AA from (22)). The first order differential equation (20) can be solved by separation of variables and integration, i.e. we have to solve

∫f(0)f(t)(d2Af2(t′)+d1f(t′)+d0A)−1df(t′)=−∫0tdt′.(24)\int_{f(0)}^{f(t)}\left(d_2Af^2(t')+d_1f(t')+\frac{d_0}{A}\right)^{-1}df(t')=-\int_0^t dt'. \tag*{(24)}

Thus, f(t)f(t) in our ansatz (11) is obtained by solving (24). We note here that f~(t)\tilde{f}(t) that appears in ansatz (8) is obtained by solving the integral of the same form as (24), but in which A0A_0 is in place of AA [14]. Thus, once f(t)f(t) is obtained, one easily obtains the corresponding f~(t)\tilde{f}(t) by replacing AA in f(t)f(t) with A0A_0. Furthermore, for initial conditions (x0=0,v0>0)(x_0=0,v_0>0) (22) gives A=A0=v0/ω0A=A_0=v_0/\omega_0 and (23) gives φ=φ0=−π/2\varphi=\varphi_0=-\pi/2, i.e. the same values as in the undamped case. Thus, for this particular type of initial conditions, approximate analytical solutions (8) and (11) are the same, but the corresponding velocities (9) and (12), as well as the energies (10) and (15), are not the same.

For an easier comparison of approximate analytical solutions obtained by the new approach presented here with the solutions obtained by the approach used in [14], we will consider a block-spring system with the same parameters as in [14], i.e. we consider a block of mass m=1m=1 kg attached to a spring of stiffness k=30k=30 N/m, and we use g=9.81g=9.81 m/s2s^2. The corresponding undamped angular frequency is ω0=5.48 s−1\omega_0=5.48\,s^{-1} and the period of the undamped system is T0=1.15T_0=1.15 s. These values are along the lines of experiments performed on block-spring systems, e.g. see [8]. Throughout the paper, as in [14], we consider initial conditions that correspond to undamped solutions with A0=0.2A_0=0.2 m and various values of φ0\varphi_0, e.g., for (x0=A0,v0=0)(x_0=A_0,v_0=0) we have φ0=0\varphi_0=0, for (x0=0,v0=ω0A0)(x_0=0,v_0=\omega_0A_0) we have φ0=−π/2\varphi_0=-\pi/2, etc.

For this choice of the parameters of block-spring system and initial conditions, solutions of the form (8), obtained in [14], describe well the dynamics governed by the equation (3) for a range of values of the damping constants μ\mu, bb and DD that correspond to

0<d0A0≲0.03ω0,0<d1≲0.1ω0,0<d2A0≲0.1ω0.(25)0<\frac{d_0}{A_0}\lesssim0.03\omega_0,\quad0<d_1\lesssim0.1\omega_0,\quad0<d_2A_0\lesssim0.1\omega_0. \tag*{(25)}

Values of d0/A0d_0/A_0, d1d_1, and d2A0d_2A_0 are suitable to quantify the damping strength, since the weak damping limit can be characterized by the relations d0/A0≪ω0d_0/A_0\ll\omega_0, d1≪ω0d_1\ll\omega_0, and d2A0≪ω0d_2A_0\ll\omega_0 [14]. In what follows, we will compare the approximate analytical solutions of the equation of motion (3) (with all three types of damping present) that are more accurate than the solutions recently presented in [14], but since the approach that includes the determination of constants AA and φ\varphi using equations (22) and (23) has not yet been investigated in the cases where only one type of nonlinear damping is present, e.g. for harmonic oscillator damped only by a force quadratic in velocity or only by sliding friction, we will first focus on those simpler cases.

Damping quadratic in velocity

If D>0D > 0 and μ=b=0\mu= b = 0 (i.e. d2>0d_2 > 0 and d0=d1=0d_0 = d_1 = 0), equation (22) becomes a quadratic equation in AA and we easily get

A=2x0v0d2+4x02v02d22+4(ω02−x02d22)(v02+ω02x02)2(ω02−x02d22)(26)A = \frac{2x_0v_0d_2 + \sqrt{4x_0^2v_0^2d_2^2 + 4\left(\omega_0^2 - x_0^2d_2^2\right)\left(v_0^2 + \omega_0^2x_0^2\right)}}{2\left(\omega_0^2 - x_0^2d_2^2\right)} \tag*{(26)}

while equation (23) simplifies to

φ=arctan⁡(−v0+x0d2Aω0x0).(27)\varphi= \arctan\left(-\frac{v_0 + x_0d_2A}{\omega_0x_0}\right). \tag*{(27)}

In this case, solving the integral (24) gives

f(t)=11+d2At.(28)f(t) = \frac{1}{1+d_2At}. \tag*{(28)}

Thus, in case of damping quadratic in velocity, our approximate solution (11) becomes

x(t)=(A1+d2At)cos⁡(ω0t+φ),(29)x(t) = \left(\frac{A}{1+d_2At}\right)\cos(\omega_0t+\varphi), \tag*{(29)}

where AA and φ\varphi are given by (26) and (27). In this case, the approach used in [14] gives

x~(t)=(A01+d2A0t)cos⁡(ω0t+φ0),(30)\tilde{x}(t) = \left(\frac{A_0}{1+d_2A_0t}\right)\cos(\omega_0t+\varphi_0), \tag*{(30)}

where A0A_0 and φ0\varphi_0 are defined in (6).

Initial conditions (x0>0x_0 > 0, v0=0v_0 = 0)

For this type of initial conditions (26) and (27) reduce to

A=x01−(d2x0/ω0)2A = \frac{x_0}{\sqrt{1-(d_2x_0/\omega_0)^2}}

and

φ=arctan⁡(−d2Aω0),(31)\varphi= \arctan\left(-\frac{d_2A}{\omega_0}\right), \tag*{(31)}

while A0=x0A_0 = x_0 and φ0=0\varphi_0 = 0. As noted earlier, throughout the paper we always choose the initial conditions that correspond to A0=0.2 mA_0 = 0.2\,\mathrm{m}, i.e. we take here x0=0.2 mx_0 = 0.2\,\mathrm{m}. In Figure 2(a)–(d) we show solutions x(t)x(t) given by (29), solutions x~(t)\tilde{x}(t) given by (30) and the corresponding numerical solutions for D/(kg/m)={1.18,2.95,5.89,8.84}D/(\mathrm{kg}/\mathrm{m}) = \{1.18, 2.95, 5.89, 8.84\}, i.e. for d2A0/ω0={0.1,0.25,0.5,0.75}d_2A_0/\omega_0 = \{0.1, 0.25, 0.5, 0.75\}, with initial conditions (x0=0.2 m,v0=0)(x_0 = 0.2\,\mathrm{m}, v_0 = 0). The ode45 MATLAB function has been utilized to obtain the corresponding numerical solutions of the equation (3). We can see that the solutions x(t)x(t) agree better with the numerical results than the solutions x~(t)\tilde{x}(t) for all chosen values of the ratio d2A0/ω0d_2A_0/\omega_0. In Figure 2(a) we have shown a shorter time span compared to Figure 2(b)–(d), in order to make the difference between x(t)x(t) and x~(t)\tilde{x}(t) more visible.

Four panels showing analytical, approximate, and numerical solutions $x(t)$

Figure 2. Solid red curves show the solutions x(t)x(t) with initial conditions (x0=0.2m,v0=0)(x_0 = 0.2m, v_0 = 0) and various values of the ratio d2A0/ω0d_2A_0/\omega_0. Solid blue curves show the solutions x~(t)\tilde{x}(t) and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratio d2A0/ω0d_2A_0/\omega_0. See text for details.

In Figure 3(a)–(d) we show the energies E(t)E(t) given by (15), the energies E~(t)\tilde{E}(t) given by (10), and the corresponding numerically obtained energies for d2A0/ω0={0.1,0.25,0.5,0.75}d_2A_0/\omega_0 = \{0.1, 0.25, 0.5, 0.75\} with initial conditions (x0=0.2 m,v0=0)(x_0 = 0.2\,\mathrm{m}, v_0 = 0). For the chosen parameters of the block-spring system and the chosen values of the initial conditions, we have E(0)=E~(0)=0.6 JE(0) = \tilde{E}(0) = 0.6\,\mathrm{J}.

Four panels showing solid red, solid blue, and black dotted energy curves for different damping ratios

Figure 3. Solid red curves show the energies E(t)E(t) with initial conditions (x0=0.2 m,v0=0)(x_0 = 0.2\,\mathrm{m}, v_0 = 0) for the same values of the ratio d2A0/ω0d_2 A_0/\omega_0 as used in Figure 2(a)–(Figure 2d). Solid blue curves show the energies E~(t)\widetilde{E}(t) and black dotted curves show the numerically obtained energies with the same initial conditions and for the corresponding values of the ratio d2A0/ω0d_2 A_0/\omega_0. See text for details.

Initial conditions (x0=0x_0 = 0, v0>0v_0 > 0)

For this type of initial conditions, (26) and (27) simplify to A=A0=v0/ω0A = A_0 = v_0/\omega_0 and φ=φ0=−π/2\varphi= \varphi_0 = -\pi/2, i.e. (29) and (30) become

x(t)=x~(t)=v0ω0(1+d2v0t/ω0)sin⁡(ω0t).(32)x(t) = \tilde{x}(t) = \frac{v_0}{\omega_0(1+d_2v_0t/\omega_0)}\sin(\omega_0t). \tag*{(32)}

We take here v0=ω0A0=1.10m/sv_0 = \omega_0A_0 = 1.10m/s. In Figure 4(a) and (b) we show solutions (32) and the corresponding numerical solutions with initial conditions (x0=0,v0=1.10m/s)(x_0 = 0, v_0 = 1.10m/s) for d2A0/ω0={0.25,0.5}d_2A_0/\omega_0 = \{0.25, 0.5\}. In Figure 5(a) and (b) we show the corresponding energies given by (15) and (10), i.e. E(t)E(t) and E~(t)\tilde{E}(t), and numerically obtained energies. Therefore, as we commented earlier, although the new approach and the approach used in [14] give the same approximate analytical solutions in the case of initial conditions (x0=0,v0≠0)(x_0 = 0, v_0 \ne0), the corresponding energies differ even in this case, i.e. the energy is in general more realistically modeled in the new approach.

Two plots of $x(t)$ for $d_2A_0=0.25\omega_0$ and $d_2A_0=0.5\omega_0$

Figure 4. Solid red curves show the solutions x(t)x(t) with initial conditions (x0=0,v0=1.10m/s)(x_0 = 0, v_0 = 1.10m/s) for two values of the ratio d2A0/ω0d_2A_0/\omega_0. Dotted black curves show the numerical solutions for the same initial conditions and the corresponding values of the ratio d2A0/ω0d_2A_0/\omega_0. See text for details.

Two plots of the energies $E(t)$ and $\widetilde{E}(t)$ for $d_2A_0=0.25\omega_0$ and $d_2A_0=0.5\omega_0$

Figure 5. Solid red curves show the energies E(t)E(t) with initial conditions (x0=0,v0=1.1m/s)(x_0 = 0, v_0 = 1.1m/s) for the same values of the ratio d2A0/ω0d_2A_0/\omega_0 as used in Figure 4(a) and (b). Solid blue curves show the energies E~(t)\widetilde{E}(t) and black dotted curves show the numerically obtained energies with the same initial conditions and for the corresponding values of the ratio d2A0/ω0d_2A_0/\omega_0. See text for details.

Initial conditions (x0≠0,v0≠0)(x_0 \ne0, v_0 \ne0)

As an example, here we take x0=A02/2=0.14mx_0 = A_0\sqrt{2}/2 = 0.14m and v0=−ω0x0=−0.77m/sv_0 = -\omega_0x_0 = -0.77m/s. For this choice, initial potential energy is equal to initial kinetic energy, and φ0=π/4\varphi_0 = \pi/4. In Figure 6a and (b) we show solutions x(t)x(t) given by (29), solutions x~(t)\tilde{x}(t) given by (30), and the corresponding numerical solutions for d2A0/ω0={0.25,0.5}d_2A_0/\omega_0 = \{0.25, 0.5\} with chosen initial conditions. In Figure 7a and (b) we show the corresponding energies given by (15) and (10), i.e. E(t)E(t) and E~(t)\tilde{E}(t), and numerically obtained energies.

Comments on results for harmonic oscillator with damping quadratic in velocity

The results presented in this section show that the new approximate analytical solutions, i.e. solutions (29) with initial amplitude (26) and initial phase (27), agree excellently with the numerical solutions in describing the free oscillations of a harmonic oscillator with damping quadratic in velocity for all types of initial conditions and for a range of values 0<D≲5.89 kg/m0 < D \lesssim5.89\,\mathrm{kg}/\mathrm{m} of the corresponding damping constant, i.e. for 0<d2A0/ω0≲0.50 < d_2 A_0/\omega_0 \lesssim0.5. For d2A0/ω0>0.5d_2 A_0/\omega_0 > 0.5 the deviations of the solutions (29) from the numerical solutions become more pronounced, as the example in Figure 2(d), with D=8.84 kg/mD = 8.84\,\mathrm{kg}/\mathrm{m} (i.e. d2A0/ω0=0.75d_2 A_0/\omega_0 = 0.75), clearly shows.

Although the differential equation (20), which gives us the function f(t)f(t), was derived under the assumption of weak damping, i.e. under the assumption ∣df(t)/dt∣≪ω0\lvert df(t)/dt\rvert\ll\omega_0, we see that solutions (29) describe the dynamics well for damping that we could characterize (at least) as moderately strong. For example, in [21] the elliptic harmonic balance method has been used to obtain approximate analytical solutions of the harmonic oscillator with strong quadratic damping, and the parameters used were such that the amplitude of the obtained approximate analytical solution, as well as the corresponding numerical solution, after the first period drops to somewhat less than 50% of the initial value (see Figure 6 in [21]). We can see, in Figure 2(b), that for our choice of parameters, this happens for d2A0/ω0=0.25d_2 A_0/\omega_0 = 0.25, and for d2A0/ω0=0.5d_2 A_0/\omega_0 = 0.5 the amplitude after the first period drops to slightly less than 25% of the initial value, as shown in Figure 2(c). Therefore, although our derivation of the function f(t)f(t) is based on the assumption that it changes slowly with time relative to the oscillatory (trigonometric) part of the dynamics, which is strictly fulfilled only in the limit of weak damping, the solution we obtained describes the dynamics well even for rather strong damping.

Solution $x(t)$ curves for $d_2A_0=0.25\omega_0$

(a)

Solution $x(t)$ curves for $d_2A_0=0.5\omega_0$

(b)

Figure 6. Solid red curves show the solutions x(t)x(t) with initial conditions (x0=0.14m,v0=−0.77m/s)(x_0=0.14m, v_0=-0.77m/s) for two values of the ratio d2A0/ω0d_2A_0/\omega_0. Solid blue curves show the solutions x~(t)\widetilde{x}(t) and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratio d2A0/ω0d_2A_0/\omega_0. See text for details.

Furthermore, an ansatz of similar form as (11) is used in [22], but with an unknown frequency (i.e. with ω≠ω0\omega\ne\omega_0) and, by a more complex mathematical procedure, which also includes the time averaging over the trigonometric part of the ansatz, approximate solutions of the harmonic oscillator with damping quadratic in velocity were obtained with a frequency that is reduced in relation to the frequency of the undamped case. More precisely, the obtained frequency is ω=ω0−Δω0\omega= \omega_0 - \Delta\omega_0 [22], where Δω0>0\Delta\omega_0 > 0 is the term which depends on the constant of quadratic damping DD and on magnitude of the initial amplitude. A brief numerical analysis of the free oscillations of the harmonic oscillator with damping quadratic in velocity presented in [14] shows, for initial condition (x0>0,v0=0)(x_0 > 0, v_0 = 0) and 0<d2A0/ω0≤0.50 < d_2A_0/\omega_0 \le0.5, that the quadratic damping has the greatest effect on the shift of the instants of the first zero crossing and the first turning point, while the time interval between the subsequent turning points remains approximately equal to the half-period of the undamped oscillator. Thus, the dynamics can be described well by keeping the frequency the same as in the undamped case, and taking into account that the initial phase can change compared to the undamped case, as we have done here.

Energy $E(t)$ curves for $d_2A_0=0.25\omega_0$

(a)

Energy $E(t)$ curves for $d_2A_0=0.5\omega_0$

(b)

Figure 7. Solid red curves show the energies E(t)E(t) with initial conditions (x0=0.14,v0=−0.77m/s)(x_0=0.14, v_0=-0.77m/s) for the same values of the ratio d2A0/ω0d_2A_0/\omega_0 as used in Figure 6a and (b). Solid blue curves show the energies E~(t)\widetilde{E}(t) and black dotted curves show the numerically obtained energies with the same initial conditions and for the corresponding values of the ratio d2A0/ω0d_2A_0/\omega_0. See text for details.

Coulomb damping

If μ>0\mu>0 and D=b=0D=b=0 (i.e. d0>0d_0>0 and d1=d2=0d_1=d_2=0), equation (22) becomes a (depressed) quartic equation in AA, i.e.

ω02A4−(v02+ω02x02)A2−2x0v0d0A−d02x02=0(33)\omega_0^2 A^4-\left(v_0^2+\omega_0^2x_0^2\right)A^2-2x_0v_0d_0A-d_0^2x_0^2=0 \tag*{(33)}

while equation (23) simplifies to

φ=arctan⁡(−v0+x0d0/Aω0x0).(34)\varphi=\arctan\left(\frac{-v_0+x_0d_0/A}{\omega_0x_0}\right). \tag*{(34)}

In this case solving integral (24) gives

f(t)=1−d0At.(35)f(t)=1-\frac{d_0}{A}t. \tag*{(35)}

The function (35) monotonically decreases from the value f(0)=1f(0)=1 to f(τ)=0f(\tau)=0, where τ=A/d0\tau=A/d_0. For t>τt>\tau, (35) becomes increasingly negative, which corresponds to an increase in the amplitude of oscillations and is not physical. Thus, in the case of Coulomb damping, our approximate solution (11) can be written as

x(t)=(A−d0t)θ(t)cos⁡(ω0t+φ),(36)x(t)=(A-d_0t)\theta(t)\cos(\omega_0t+\varphi), \tag*{(36)}

where

θ(t)={1if 0≤t≤τ,0if t>τ,(37)\theta(t)=\begin{cases}1 & \text{if }0\leq t\leq\tau,\\0 & \text{if }t>\tau,\end{cases} \tag*{(37)}

while AA is obtained by solving (33) and φ\varphi calculated from (34). The corresponding velocity, i.e. (12), can be written as

v(t)=(−ω0(A−d0t)sin⁡(ω0t+φ)−d0cos⁡(ω0t+φ))θ(t),(38)v(t)=\left(-\omega_0(A-d_0t)\sin(\omega_0t+\varphi)-d_0\cos(\omega_0t+\varphi)\right)\theta(t), \tag*{(38)}

Thus, the function θ(t)\theta(t) is simply added by hand to exclude the non-physical part of the solution. In this case, the approach used in [14] gives

x~(t)=(A0−d0t)θ~(t)cos⁡(ω0t+φ0),(39)\tilde{x}(t)=(A_0-d_0t)\tilde{\theta}(t)\cos(\omega_0t+\varphi_0), \tag*{(39)}

where θ~(t)\tilde{\theta}(t) is the same as (37) but with τ~=A0/d0\tilde{\tau}=A_0/d_0 instead of τ\tau. It is well known that the exact solutions of a harmonic oscillator with Coulomb damping halt at some finite time instant thaltt_{\mathrm{halt}} and, in general, at some displacement xhalt≠0x_{\mathrm{halt}}\ne0 with property ∣xhalt∣≤μmg/k\lvert x_{\mathrm{halt}}\rvert\leq\mu mg/k, e.g., see [5, 6, 7, 14, 23]. The approximate solution (36) describes the dynamics up to the instant t=τt=\tau, at which x(τ)=0x(\tau)=0 and the corresponding velocity, given by (38), is v(τ)≠0v(\tau)\ne0. Thus, if we take τ\tau as an approximation of thaltt_{\mathrm{halt}}, in addition to an error in the halting position, our modeled solution (36) has an error in velocity. In case of solution (39), there is also an error in the halting position, since x~(τ~)=0\tilde{x}(\tilde{\tau})=0, but the corresponding velocity, given by (9), is v~(τ~)=0\tilde{v}(\tilde{\tau})=0, i.e it has the correct value at least from a modeling point of view. We will first show examples with solutions (36) and (39) as they are, and then introduce an alternative approach that corrects the aforementioned shortcomings of solution (36).

Since the exact solutions of the harmonic oscillator with Coulomb damping are piecewise continuous and have already been studied in detail elsewhere, e.g., see [5, 6, 7, 14, 23], for simplicity, we will again check the validity of our approximate solutions by comparing them with numerically obtained solutions.

Initial conditions (x0>0x_0>0, v0=0v_0=0)

For this type of initial conditions, equation (33) reduces to a quadratic equation in A2A^2 and we easily get

A=x02+x04+4(d0x0/ω0)22andφ=arctan⁡(−d0ω0A).(40)A=\sqrt{\frac{x_0^2+\sqrt{x_0^4+4(d_0x_0/\omega_0)^2}}{2}}\quad\text{and}\quad\varphi=\arctan\left(-\frac{d_0}{\omega_0A}\right). \tag*{(40)}

In Figure 8(a) and (b) we show solutions x(t)x(t) given by (36), solutions x~(t)\tilde{x}(t) given by (39) and the corresponding numerical solutions with initial conditions (x0=0.2m,v0=0)(x_0=0.2m,v_0=0) for μ={0.048,0.096}\mu=\{0.048,0.096\}, i.e. for d0/(ω0A0)={0.05,0.1}d_0/(\omega_0A_0)=\{0.05,0.1\}. In case d0/(ω0A0)=0.05d_0/(\omega_0A_0)=0.05, relations (40) give A=0.2002m≈A0A=0.2002m\approx A_0 and φ=−0.05≈φ0\varphi=-0.05\approx\varphi_0. Thus, solutions (36) and (39) almost completely overlap in Figure 8(a). Even for a rather strong Coulomb damping with d0/(ω0A0)=0.1d_0/(\omega_0A_0)=0.1, shown in Figure 8(b), the amplitude A=0.201m≈A0A=0.201m\approx A_0, but the phase becomes more clearly visible, i.e. φ=−0.1\varphi=-0.1, and we see that the solution x(t)x(t) agrees somewhat better with the numerical solution than x~(t)\tilde{x}(t) when considering the overall dynamics, but the improvement is not significant.

Solutions $x(t)$, $\tilde{x}(t)$, and numerical solutions for two values of $d_0/(A_0\omega_0)$

Figure 8. Solid red curves show the solutions x(t)x(t) with initial conditions (x0=0.2 m,v0=0)(x_0 = 0.2\,\mathrm{m}, v_0 = 0) for two values of the ratio d0/(A0ω0)d_0/(A_0\omega_0). Solid blue curves show the solutions x~(t)\tilde{x}(t) and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratio d0/(A0ω0)d_0/(A_0\omega_0). The dashed horizontal lines in each figure are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k\lvert x(t)\rvert\le\mu mg/k holds. See text for details.

The dashed horizontal lines in Figure 8(a) and (b) indicate the region within which k∣x(t)∣≤μmgk\lvert x(t)\rvert\leq\mu mg holds, i.e. the region in which the restoring force is of smaller magnitude than the friction force. In both Figure 8(a) and (b) we see that the approximate solutions have the last turning point within the area marked by the two dashed lines. After the last turning point, our approximate solutions describe the motion that continues until the block reaches the equilibrium position. In addition to the error in the halting positions, in Figure 8(a) and (b) we see that solutions x(t)x(t) and x~(t)\tilde{x}(t) also have errors in the instants of the zero crossings compared to the numerical solutions, and we can see that these errors increase with time, therefore the largest error is for the last zero crossing (the fact that solutions of the form (39) have errors in instants of zero crossings compared to exact analytical solutions of the harmonic oscillator with Coulomb damping is argued in [14], and the same argument can be applied to solutions of the form (36)). One way to make our description of the dynamics more physical would be simply to take that the dynamics halts when the approximate solutions reach the last turning point. In the examples shown in Figure 8(a) and (b), we see that this would somewhat reduce the error in the halting position, but would not affect the rather large error in the instant of the last zero crossing. Thus, to improve the agreement between the solutions of the form (36) and the numerical solutions, we will apply a slightly different strategy.

In Figure 8(b), as an example, the red arrow points to (t0,x(t0))=(1.12 s,0.076 m)(t_0,x(t_0)) = (1.12\,\mathrm{s}, 0.076\,\mathrm{m}), i.e. to the last turning point of the solution x(t)x(t) for which k∣x(t0)∣>μmgk\lvert x(t_0)\rvert> \mu mg holds. We can take that solution (36) describes the dynamics for 0≤t≤t00 \le t \le t_0, and to describe the dynamics for t>t0t > t_0, we exactly solve the equation of motion with the initial conditions (x(t0)≠0,v(t0)=0)(x(t_0) \ne0, v(t_0) = 0). We get

x(t)={(A−d0t)cos⁡(ω0t+φ)if 0≤t≤t0(x(t0)−sgn⁡[x(t0)]μmgk)cos⁡(ω0(t−t0))+sgn⁡[x(t0)]μmgkif t0<t≤t0+πω0−x(t0)+2sgn⁡[x(t0)]μmgkif t>t0+πω0(41)x(t)= \begin{cases} (A-d_0t)\cos(\omega_0t+\varphi) & \text{if } 0\le t\le t_0 \\ \left(x(t_0)-\operatorname{sgn}[x(t_0)]\frac{\mu mg}{k}\right)\cos(\omega_0(t-t_0))+\operatorname{sgn}[x(t_0)]\frac{\mu mg}{k} & \text{if } t_0<t\le t_0+\frac{\pi}{\omega_0} \\ -x(t_0)+2\operatorname{sgn}[x(t_0)]\frac{\mu mg}{k} & \text{if } t>t_0+\frac{\pi}{\omega_0} \end{cases} \tag*{(41)}

as the improved approximate solution, and

v(t)={−ω0(A−d0t)sin⁡(ω0t+φ)−d0cos⁡(ω0t+φ)if 0≤t≤t0−ω0(x(t0)−sgn⁡[x(t0)]μmgk)sin⁡(ω0(t−t0))if t0<t≤t0+πω00if t>t0+πω0(42)v(t)= \begin{cases} -\omega_0(A-d_0t)\sin(\omega_0t+\varphi)-d_0\cos(\omega_0t+\varphi) & \text{if } 0\le t\le t_0 \\ -\omega_0\left(x(t_0)-\operatorname{sgn}[x(t_0)]\frac{\mu mg}{k}\right)\sin(\omega_0(t-t_0)) & \text{if } t_0<t\le t_0+\frac{\pi}{\omega_0} \\ 0 & \text{if } t>t_0+\frac{\pi}{\omega_0} \end{cases} \tag*{(42)}

as the corresponding velocity. Therefore, our strategy is as follows: knowing the function (A−d0t)cos⁡(ω0t+φ)(A-d_0t)\cos(\omega_0t+\varphi), we can easily read out, for any choice of initial conditions and any value of the ratio d0/(ω0A0)d_0/(\omega_0A_0), the values of t0t_0 and x(t0)x(t_0), i.e. the values corresponding to the last turning point with the property k∣x(t0)∣>μmgk\lvert x(t_0)\rvert> \mu mg, and use these values in relations (41) and (42). In Figure 9(a) and (b) we show improved solutions x(t)x(t) given by (41), solutions x~(t)\tilde{x}(t) given by (39) and the corresponding numerical solutions for the same initial conditions and ratios d0/(ω0A0)d_0/(\omega_0A_0) as used in Figure 8(a) and (b). We see that solution (41) is in excellent agreement with the numerical solution, i.e. it gives an excellent approximation of the halting position, and the time t0+π/ω0t_0+\pi/\omega_0 is an excellent approximation of the duration of motion. In Figure 10(a) and (b) we show the energy E(t)=kx(t)2/2+mv(t)2/2E(t)=kx(t)^2/2+mv(t)^2/2 obtained using (41) and (42), energy E˙(t)=mω02(A0−d0t)2θ˙(t)/2\dot{E}(t)=m\omega_0^2(A_0-d_0t)^2\dot{\theta}(t)/2 corresponding to solution (39), and the numerically obtained energy for the same initial conditions and ratios d0/(ω0A0)d_0/(\omega_0A_0) as used in Figure 9(a) and (b). We can see that energy E(t)E(t), in addition to modeling the energy behavior more realistically compared to energy E~(t)\tilde{E}(t), is in excellent agreement with the numerically obtained energy.

Solutions $x(t)$, $\tilde{x}(t)$, and numerical solutions for two values of $d_0/(A_0\omega_0)$

Figure 9. Solid red curves show the solutions (41) with initial conditions (x0=0.2m,v0=0)(x_0=0.2m, v_0=0) for two values of the ratio d0/(A0ω0)d_0/(A_0\omega_0). Solid blue curves show the solutions x~(t)\tilde{x}(t), given by (39), and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratio d0/(A0ω0)d_0/(A_0\omega_0). The dashed horizontal lines in each figure are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)|\leq\mu mg/k holds. See text for details.

Energies $E(t)$, $\tilde{E}(t)$, and numerical energies for two values of $d_0/A_0$

Figure 10. Solid red curves show the energies E(t)E(t), obtained using (41) and (42), for the same initial conditions and values of the ratio d2A0/ω0d_2A_0/\omega_0 as used in Figure 9(a) and (b). Solid blue curves show the energies E~(t)=mω02(A0−d0t)2θ~(t)/2\tilde{E}(t)=m\omega_0^2(A_0-d_0t)^2\tilde{\theta}(t)/2 corresponding to solutions (39), and black dotted curves show the numerically obtained energies, both for the same initial conditions and values of the ratio d2A0/ω0d_2A_0/\omega_0 as used for E(t)E(t). See text for details.

Our approximate analytical solution (41) is piecewise continuous, but it is necessary to consider only two time intervals, i.e. 0≤t≤t00\le t\le t_0 and t0<t≤t0+π/ω0t_0<t\le t_0+\pi/\omega_0, while the exact solution of the harmonic oscillator with Coulomb damping describes the dynamics over half periods [5, 6, 7, 14, 23]. For example, in Figure 11(a) we show the solutions (41), (39) and the corresponding numerical solutions for d0/(ω0A0)=0.03d_0/(\omega_0A_0)=0.03 (i.e. for μ=0.029\mu=0.029). In this case, an exact analytical solution would require considering the dynamics over eleven time intervals, i.e. eleven half-periods, while, as we said, solution (41) considers only two time intervals and agrees excellently with the numerically obtained solution. In Figure 11(b) we show the energies corresponding to solutions shown in Figure 11(a).

Solutions and energies for initial conditions $x_0=0.2m$, $v_0=0$

Figure 11. (a) Solid red curves show the solution (41) with initial conditions (x0=0.2mx_0 = 0.2m, v0=0v_0 = 0) and d0/(A0ω0)=0.03d_0/(A_0\omega_0) = 0.03. Solid blue curves show the corresponding solution (39) and black dotted curves show the corresponding numerical solution. The dashed horizontal lines are positioned at ±μmg/k\pm\mu mg/k, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. (b) The energies corresponding to solutions shown in figure (a). See text for details.

Initial conditions (x0=0,v0>0)(x_0=0, v_0>0)

For this type of initial conditions, equation (33) simplifies to A=A0=v0/ω0A=A_0=v_0/\omega_0 and we get φ=φ0=−π/2\varphi=\varphi_0=-\pi/2, i.e. (36) and (39) become equal. The improved solution, i.e. (41), is, of course, also valid for these initial conditions, it is only necessary to determine the values (t0,x(t0))(t_0,x(t_0)) of the last turning point of the function (A0−d0t)cos⁡(ω0t+φ)(A_0-d_0t)\cos(\omega_0t+\varphi) with property ∣x(t0)∣>μmg/k|x(t_0)|>\mu mg/k, and insert these values into expression (41). For example, in Figure 12(a) we show the solutions (41), (39) and the corresponding numerical solutions for this choice of the initial conditions and d0/(ω0A0)=0.05d_0/(\omega_0A_0) = 0.05. In this case, solutions (41) and (39) overlap for 0<t≤t00 < t \leq t_0, and for t>t0t > t_0 the solution (41) agrees significantly better with the numerical solution. In Figure 12(b) we show the energies corresponding to solutions shown in Figure 12(a).

Solutions and energies for initial conditions $x_0=0$, $v_0=\omega_0A_0$

Figure 12. (a) Solid red curves show the solution (41) with initial conditions (x0=0x_0 = 0, v0=ω0A0v_0 = \omega_0A_0) and d0/(ω0A0)=0.05d_0/(\omega_0A_0) = 0.05. Solid blue curves show the corresponding solution (39) and black dotted curves show the corresponding numerical solution. The dashed horizontal lines are positioned at ±μmg/k\pm\mu mg/k, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. (b) The energies corresponding to solutions shown in figure (a). See text for details.

Initial conditions (x0≠0x_0 \ne0, v0≠0v_0 \ne0)

As an example, here we choose (x0=A02/2x_0 = A_0\sqrt{2}/2, v0=ω0x0v_0 = \omega_0x_0) and d0/(ω0A0)=0.05d_0/(\omega_0A_0) = 0.05. For this type of initial conditions, equation (33) remains a quartic equation in AA, i.e. we have to solve

30A4−1.200A2−0.012A−6×10−5=0,30A^4 - 1.200A^2 - 0.012A - 6 \times10^{-5} = 0 ,

where we did not write the physical units for simplicity and the resulting AA is in meters. Instead of delving into solving this quartic equation analytically, we solve it using the roots MATLAB function. We get four solutions, i.e. A/m={0.205,−0.195,−0.005+i0.005,−0.005−i0.005}A/m = \{0.205, -0.195, -0.005 + i0.005, -0.005 - i0.005\}. We expected one of the solutions to be close to A0=0.2mA_0 = 0.2m, therefore we take the value A=0.205mA = 0.205m as our solution, and the corresponding phase (34) is φ=−0.809\varphi= -0.809. In Figure 13(a) we show the solutions (41), (39) and the numerical solutions corresponding to this choice of the initial conditions and value of the ratio d0/(ω0A0)d_0/(\omega_0A_0). In Figure 13(b) we show the energies corresponding to solutions shown in Figure 13(a).

Solutions and energies for the harmonic oscillator with Coulomb damping

Figure 13. (a) Solid red curves show the solution (41) with initial conditions (x0=A02/2,v0=ω0x0)(x_0 = A_0\sqrt{2}/2, v_0 = \omega_0x_0) and d0/(A0ω0)=0.05d_0/(A_0\omega_0) = 0.05. Solid blue curves show the corresponding solution (39) and black dotted curves show the corresponding numerical solution. The dashed horizontal lines are positioned at ±μmg/k\pm\mu mg/k, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. (b) The energies corresponding to solutions shown in figure (a). See text for details.

Comments on results for harmonic oscillator with Coulomb damping

The results presented in this section show that, in the case of Coulomb damping, the new approximate analytical solutions, i.e. solutions (36) with the corresponding AA and φ\varphi, do not provide a significantly better description of the dynamics compared to the solutions (39) which are provided by the approach used in [14]. In order to obtain a better agreement between the approximate analytical solutions (36) and the numerical solutions, we improved them by using (36) to describe the dynamics only up to the instant t0t_0, which corresponds to the last turning point with displacement ∣x(t0)∣>μmg/k|x(t_0)| > \mu mg/k, while the description of the dynamics for t>t0t > t_0 is obtained by solving exactly the equation of motion with the initial conditions (x(t0)≠0,v(t0)=0)(x(t_0) \ne0, v(t_0) = 0), i.e. we introduced improved approximate analytical solution (41). We showed that solution (41) gives excellent description of free oscillations of the harmonic oscillator with Coulomb damping for any type of initial conditions and for a range of corresponding damping constant 0<μ≤0.0480 < \mu\leq0.048, i.e. for 0<d0/(ω0A0)≤0.050 < d_0/(\omega_0A_0) \leq0.05.

As we have already mentioned, the case of free oscillations of the harmonic oscillator with Coulomb damping is exactly solvable, but the exact solution is piecewise continuous, i.e. it is necessary to analyze the dynamics over half-periods. Thus, the main advantage of the approximate analytical solution (41), compared to the exact solution, is that it requires us to consider only two time intervals, i.e. 0≤t≤t00 \leq t \leq t_0 and t0≤t≤t0+π/ω0t_0 \leq t \leq t_0 + \pi/\omega_0. This advantage is lost as we consider increasingly stronger Coulomb damping. For example, in Figure 9(b), we see that even for a rather strong damping with d0/(ω0A0)=0.1d_0/(\omega_0A_0) = 0.1, for which the amplitude after the first period drops to a value less than 50% of the initial amplitude, the approximate analytical solution (41) follows the numerical solution well, but in this example the advantage of solution (41) compared to the exact solution would not be great, i.e. solution (41) is written over two time intervals, while the corresponding exact solution would be written over three time intervals (three half-periods).

We note here that approximate analytical solutions (36) and (39), for the range of values 0<d0/(ω0A0)≤0.050 < d_0/(\omega_0A_0) \leq0.05, both begin to deviate significantly from the numerical solutions only after their corresponding energies, as well as the numerically obtained energies, drop to a small fraction of the initial energy, e.g. for d0/(ω0A0)=0.05d_0/(\omega_0A_0) = 0.05 to about 5% of initial energy, i.e. towards the very end of the dynamics. Thus, if we do not need high precision in describing the very end of the dynamics, both solutions (36) and (39) provide an elegant description of the overall dynamics and a solid estimate of the duration of the motion.

Combined Quadratic and Coulomb Damping

If μ>0\mu> 0, D>0D > 0 and b=0b = 0 (i.e. d0>0d_0 > 0, d2>0d_2 > 0 and d1=0d_1 = 0), equation (22) becomes a quartic equation in AA, i.e.

(ω02−x02d22)A4−2x0v0d2A3−(v02+ω02x02+2d0d2x02)A2−2x0v0d0A−d02x02=0(43)(\omega_0^2-x_0^2d_2^2)A^4-2x_0v_0d_2A^3-(v_0^2+\omega_0^2x_0^2+2d_0d_2x_0^2)A^2-2x_0v_0d_0A-d_0^2x_0^2=0 \tag*{(43)}

while the initial phase (23) becomes

φ=arctan⁡(−v0+(d2A+d0/A)x0ω0x0).(44)\varphi=\arctan\left(-\frac{v_0+(d_2A+d_0/A)x_0}{\omega_0x_0}\right). \tag*{(44)}

In this case, solving the integral (24) gives

f(t)=d0d2A2tan⁡(−d0d2 t+arctan⁡(d2A2d0)).(45)f(t)=\sqrt{\frac{d_0}{d_2A^2}}\tan\left(-\sqrt{d_0d_2}\,t+\arctan\left(\sqrt{\frac{d_2A^2}{d_0}}\right)\right). \tag*{(45)}

Similarly as in the case of purely Coulomb damping, the function (35) monotonically decreases from the value f(0)=1f(0)=1 to f(τ)=0f(\tau)=0, where

τ=1d0d2arctan⁡(d2A2d0).(46)\tau=\frac{1}{\sqrt{d_0d_2}}\arctan\left(\sqrt{\frac{d_2A^2}{d_0}}\right). \tag*{(46)}

For t>τt>\tau, (45) becomes increasingly negative. Thus, our approximate solution (11) becomes

x(t)=d0d2tan⁡(−d0d2 t+arctan⁡(d2A2d0))θ(t)cos⁡(ω0t+φ),(47)x(t)=\sqrt{\frac{d_0}{d_2}}\tan\left(-\sqrt{d_0d_2}\,t+\arctan\left(\sqrt{\frac{d_2A^2}{d_0}}\right)\right)\theta(t)\cos(\omega_0t+\varphi), \tag*{(47)}

where θ(t)\theta(t) is defined as in (37), but with τ\tau given by (46), while AA is obtained by solving (43) and φ\varphi calculated from (44). The velocity corresponding to (47) is

v(t)=(−ω0Af(t)sin⁡(ω0t+φ)+Adf(t)dtcos⁡(ω0t+φ))θ(t),(48)v(t)=\left(-\omega_0Af(t)\sin(\omega_0t+\varphi)+A\frac{df(t)}{dt}\cos(\omega_0t+\varphi)\right)\theta(t), \tag*{(48)}

where f(t)f(t) is given by (45), and the corresponding energy E(t)=kx(t)2/2+mv(t)2/2E(t)=kx(t)^2/2+mv(t)^2/2 is calculated with (47) and (48). In this case, the approach used in [14] gives f~(t)\tilde{f}(t) of the same form as (45), but with A0A_0 in place of AA, i.e. approximate solution (8) becomes

x~(t)=d0d2tan⁡(−d0d2 t+arctan⁡(d2A02d0))θ~(t)cos⁡(ω0t+φ0),(49)\tilde{x}(t)=\sqrt{\frac{d_0}{d_2}}\tan\left(-\sqrt{d_0d_2}\,t+\arctan\left(\sqrt{\frac{d_2A_0^2}{d_0}}\right)\right)\tilde{\theta}(t)\cos(\omega_0t+\varphi_0), \tag*{(49)}

where θ~(t)\tilde{\theta}(t) is the same as θ(t)\theta(t) but with τ~=arctan⁡(d2A02/d0)/d0d2\tilde{\tau}=\arctan(\sqrt{d_2A_0^2/d_0})/\sqrt{d_0d_2} instead of τ\tau. In the approach used in [14], the velocity corresponding to (49) is

v~(t)=−ω0A0f~(t)sin⁡(ω0t+φ0)θ~(t),(50)\tilde{v}(t)=-\omega_0A_0\tilde{f}(t)\sin(\omega_0t+\varphi_0)\tilde{\theta}(t), \tag*{(50)}

and the corresponding energy is E~(t)=kx~(t)2/2+mv~(t)2/2=mω02A02f~(t)2θ~(t)/2\tilde{E}(t)=k\tilde{x}(t)^2/2+m\tilde{v}(t)^2/2=m\omega_0^2A_0^2\tilde{f}(t)^2\tilde{\theta}(t)/2.

Initial conditions (x0>0x_0>0, v0=0v_0=0)

For this type of initial conditions, equation (43) reduces to a quadratic equation in A2A^2 and we easily get

A=(ω02+2d0d2)x02+(ω02+2d0d2)2x04+4d02d22(ω02−d22x02)2(ω02−d22x02)A=\sqrt{\frac{(\omega_0^2+2d_0d_2)x_0^2+\sqrt{(\omega_0^2+2d_0d_2)^2x_0^4+4d_0^2d_2^2(\omega_0^2-d_2^2x_0^2)}}{2(\omega_0^2-d_2^2x_0^2)}}

and

φ=arctan⁡(−d2A+d0/Aω0).(51)\varphi=\arctan\left(-\frac{d_2A+d_0/A}{\omega_0}\right). \tag*{(51)}

In Figure 14(a)-(Figure 14d) we show solutions x(t)x(t) given by (47), solutions x~(t)\tilde{x}(t) given by (49) and the corresponding numerical solutions for d0/(ω0A0)={0.01,0.03}d_0/(\omega_0 A_0) = \{0.01, 0.03\} and d2A0/ω0={0.15,0.25}d_2 A_0/\omega_0 = \{0.15, 0.25\}, with initial conditions (x0=0.2m,v0=0)(x_0 = 0.2m, v_0 = 0). In case d0/(ω0A0)=0.01d_0/(\omega_0 A_0) = 0.01, the error of our approximate solutions (47) and (49) in the halting position compared to the numerical solutions can be at most Δx=±μmg/k=±0.016x0\Delta x = \pm\mu mg/k = \pm0.016x_0, while in case d0/(ω0A0)=0.03d_0/(\omega_0 A_0) = 0.03 it can be at most Δx=±0.047x0\Delta x = \pm0.047x_0. In Figure 14(a) and (b), we can see that for most of the time, solutions x(t)x(t) overlap perfectly with the numerical solutions, that is, only towards the end of the motion are the deviations from the numerical solutions noticeable, mostly at the instants of the last zero crossings and, of course, at the instants and the positions of the halt. These discrepancies occur due to the presence of Coulomb damping, and in Figure 14(c) and (d) are more pronounced, due to stronger Coulomb damping. In Figure 15(a)-(Figure 15d) we show the energies corresponding to solutions shown in Figure 14(a)-(Figure 14d). We see that the energies corresponding to the solution (47) are in excellent agreement with the numerically obtained energies. Thus, the deviations of the solutions (47) from the numerical solutions towards the end of the dynamics, which we can see in Figure 14(a)-(Figure 14d), are negligible in the context of energy.

Four panels showing approximate, modified approximate, and numerical solutions

Figure 14. Solid red curves show the solutions x(t)x(t), given by (47), with initial conditions (x0=0.2m,v0=0)(x_0 = 0.2m, v_0 = 0) for d0/(ω0A0)={0.01,0.03}d_0/(\omega_0 A_0) = \{0.01, 0.03\} and d2A0/ω0={0.15,0.25}d_2 A_0/\omega_0 = \{0.15, 0.25\}. Solid blue curves show the solutions x~(t)\tilde{x}(t), given by (49), and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratios d0/(ω0A0)d_0/(\omega_0 A_0) and d2A0/ω0d_2 A_0/\omega_0. The dashed horizontal lines in each figure are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. See text for details.

Energies corresponding to solutions shown in Fig. 14(a)-(d)

Figure 15. Energies corresponding to solutions shown in Figure 14(a)-(Figure 14d). See text for details.

Still, in order to make our approximate description of dynamics, i.e. the solution (47), a better model of physical reality, we will proceed similarly as in the case with purely Coulomb damping. As an example, the red arrow in Figure 14(d) points to (t0,x(t0))=(2.37s,0.049x0)(t_0, x(t_0)) = (2.37s, 0.049x_0), i.e. to the last turning point of the solution (47) with the magnitude of displacement greater than μmg/k=0.047x0\mu mg/k = 0.047x_0. In this particular case, the displacement of the considered turning point, i.e. x(t0)x(t_0), is only 4% larger in magnitude than μmg/k\mu mg/k. For this reason, it is barely visible in Figure 14(d) that this turning point is above the dashed horizontal line. Similarly, in Figure 14(c), the red curve has a fifth turning point with a displacement just slightly less in magnitude than μmg/k\mu m g/k, thus the fourth turning point of the solid red curve shown in Figure 14(c) is the last turning point with a displacement greater in magnitude than μmg/k\mu m g/k. In any case, although, depending on the range shown, it may not be easy to visually determine which of the turning points is the last turning point with the displacement greater in magnitude than μmg/k\mu m g/k, this can certainly be easily done in various ways, e.g., simply by zooming in on the area of interest on the graph and reading out the coordinates of the required turning point or by using available functions in software packages, such as MATLAB, to filter these coordinates from the numerical data provided by the expression (47). Thus, for any choice of initial conditions and values d0d_0 and d2d_2, the coordinates (t0,x(t0))(t_0,x(t_0)) can be easily determined.

From the results shown in § 3, it is clear that the damping quadratic in velocity has the highest influence on the dynamics within the first few half-periods. Thus, in the case of combined quadratic and Coulomb damping, towards the very end of the dynamics, we expect Coulomb damping to dominate. Therefore, the dynamics for t>t0t>t_0 can be well described even if we completely ignore the contribution of the quadratic damping, i.e. if we take that the exact solution of equation of motion (3) with purely Coulomb damping, with initial conditions (x(t0)≠0,v(t0)=0)(x(t_0)\ne0,v(t_0)=0), describes the dynamics for t>t0t>t_0. In that case, we obtain

x(t)={Af(t)cos⁡(ω0t+φ)if 0≤t≤t0,(x(t0)−sgn⁡[x(t0)]μmgk)cos⁡(ω0(t−t0))+sgn⁡[x(t0)]μmgkif t0<t≤t0+πω0,−x(t0)+2sgn⁡[x(t0)]μmgkif t>t0+πω0.(52)x(t)=\begin{cases} Af(t)\cos(\omega_0t+\varphi) & \text{if }0\le t\le t_0,\\ \left(x(t_0)-\operatorname{sgn}[x(t_0)]\frac{\mu m g}{k}\right)\cos\left(\omega_0(t-t_0)\right)+\operatorname{sgn}[x(t_0)]\frac{\mu m g}{k} & \text{if }t_0<t\le t_0+\frac{\pi}{\omega_0},\\ -x(t_0)+2\operatorname{sgn}[x(t_0)]\frac{\mu m g}{k} & \text{if }t>t_0+\frac{\pi}{\omega_0}. \end{cases} \tag*{(52)}

as an improved approximate analytical solution, where f(t)f(t) is given by (45), and

v(t)={−ω0Af(t)sin⁡(ω0t+φ)+Adf(t)dtcos⁡(ω0t+φ)if 0≤t≤t0−ω0(x(t0)−sgn⁡[x(t0)]μmgk)sin⁡(ω0(t−t0))if t0<t≤t0+πω00if t>t0+πω0(53)v(t)= \begin{cases} -\omega_0 A f(t)\sin(\omega_0t+\varphi)+A\frac{\mathrm{d}f(t)}{\mathrm{d}t}\cos(\omega_0t+\varphi) & \text{if }0\le t\le t_0\\ -\omega_0\left(x(t_0)-\operatorname{sgn}[x(t_0)]\frac{\mu mg}{k}\right)\sin(\omega_0(t-t_0)) & \text{if }t_0<t\le t_0+\frac{\pi}{\omega_0}\\ 0 & \text{if }t>t_0+\frac{\pi}{\omega_0} \end{cases} \tag*{(53)}

as the corresponding velocity. In Figure 16(a) and (b) we show improved solutions x(t)x(t) given by (52), solutions x~(t)\tilde{x}(t) given by (49) and the corresponding numerical solutions for the same initial conditions and ratios d0/(ω0A0)d_0/(\omega_0A_0) and d2A0/ω0d_2A_0/\omega_0 as used in Figure 14(c) and (d). We see that solution (52) gives an excellent approximation of the instant and the position of the halt. In Figure 17(a) and (b) we show the energy E(t)=kx(t)2/2+mv(t)2/2E(t)=kx(t)^2/2+mv(t)^2/2 obtained using (52)

Improved approximate solutions and numerical solutions for two parameter choices

Figure 16. Solid red curves show the improved solutions x(t)x(t) given by (52), with initial conditions (x0=0.2m,v0=0)(x_0=0.2\mathrm{m},v_0=0) for d0/(ω0A0)=0.03d_0/(\omega_0A_0)=0.03 and d2A0/ω0={0.15,0.25}d_2A_0/\omega_0=\{0.15,0.25\}. Solid blue curves show the solutions x~(t)\tilde{x}(t) given by (49) and black dotted curves show the numerical solutions for the same initial conditions and the corresponding values of the ratios d0/(ω0A0)d_0/(\omega_0A_0) and d2A0/ω0d_2A_0/\omega_0. The dashed horizontal lines in each figure are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)|\le\mu mg/k holds. See text for details.

Energy curves for two parameter choices

Figure 17. Solid red curves show the energies E(t)E(t), obtained using (52) and (53), for the same initial conditions and values of the ratios d0/(ω0A0)d_0/(\omega_0 A_0) and d2A0/ω0d_2A_0/\omega_0 as used in Figure 16(a) and (b). Solid blue curves show the energies E~(t)\tilde{E}(t) corresponding to solutions (49), and black dotted curves show the numerically obtained energies, both for the same initial conditions and values of the ratios as used for E(t)E(t). See text for details.

and (53), the energy E~(t)\tilde{E}(t) corresponding to the solution (49), and the numerically obtained energies for the same initial conditions and ratios d0/(ω0A0)d_0/(\omega_0A_0) and d2A0/ω0d_2A_0/\omega_0 as used in Figure 16(a) and (b). We can see that the energies E(t)E(t), obtained with (52) and (53), agree excellently with the numerically obtained energies, as expected, since, in the context of energy alone, even the less precise approximate solution (47), and the corresponding velocity (48), resulted in excellent approximation of numerically obtained energies.

However, if, e.g., we are not satisfied with the precision with which the solution (52) describes the dynamics, we will show how it can be further improved. In the example of the solution (47) shown in Figure 14(d), the fourth turning point (t0,x(t0))(t_0,x(t_0)), marked by a red arrow, has a displacement x(t0)x(t_0) only slightly greater than μmg/k\mu mg/k, so the corresponding improved solution, i.e. (52), necessarily halts before reaching the equilibrium position, i.e. the last zero crossing occurred before the instant t0t_0, and therefore the improved solution (52), shown in Figure 16(b), does not make any correction to the instant of the last zero crossing. In Figure 14(d), the green arrow points to the third turning point of the solution (47), with the coordinates (t3,x(t3))=(1.74s,−0.155x0)(t_3,x(t_3))=(1.74\mathrm{s},-0.155x_0), where we used subscript 3 simply to denote the third turning point. If we ignore the quadratic damping for t>t3t>t_3, we can take

x(t)={Af(t)cos⁡(ω0t+φ)if 0≤t≤t3(x(t3)+μmgk)cos⁡(ω0(t−t0))−μmgkif t3<t≤t3+πω0(x(t3)+3μmgk)cos⁡(ω0(t−t0))+μmgkif t3+πω0<t≤t3+2πω0x(t3)+4μmgkif t>t3+2πω0(54)x(t)= \begin{cases} Af(t)\cos(\omega_0t+\varphi) & \text{if }0\le t\le t_3\\ \left(x(t_3)+\frac{\mu mg}{k}\right)\cos(\omega_0(t-t_0))-\frac{\mu mg}{k} & \text{if }t_3<t\le t_3+\frac{\pi}{\omega_0}\\ \left(x(t_3)+\frac{3\mu mg}{k}\right)\cos(\omega_0(t-t_0))+\frac{\mu mg}{k} & \text{if }t_3+\frac{\pi}{\omega_0}<t\le t_3+\frac{2\pi}{\omega_0}\\ x(t_3)+\frac{4\mu mg}{k} & \text{if }t>t_3+\frac{2\pi}{\omega_0} \end{cases} \tag*{(54)}

as a further improved approximate solution, where again f(t)f(t) is given by (45). We note here that (54) is derived for the specific example with x(t3)<0x(t_3)<0, and it is easy to derive expressions that apply to the case with opposite sign. In Figure 18 we show the solution (54) and we can see a better overall agreement with the numerical solution compared to the solution (52) shown in Figure 16(b). Therefore, in the cases when the last turning point with a displacement greater in magnitude than μmg/k\mu mg/k, i.e. (t0,x(t0))(t_0,x(t_0)), has only slightly greater displacement than μmg/k\mu mg/k, i.e. when ∣x(t0)∣≳μmg/k|x(t_0)| \gtrsim\mu mg/k, we can take the coordinates of the turning point that precedes it, calculate the exact solution of equation of motion (3) with purely Coulomb damping starting from that instant, and get a better overall description of the dynamics.

Displacement curves for the improved and approximate solutions

Figure 18. Solid red curve shows the solution x(t)x(t) given by (54), with initial conditions (x0=0.2 m,v0=0)(x_0=0.2\,\mathrm{m},v_0=0) for d0/(ω0A0)=0.03d_0/(\omega_0A_0)=0.03 and d2A0/ω0=0.25d_2A_0/\omega_0=0.25. Solid blue curve shows the solution x~(t)\tilde{x}(t) given by (49) and black dotted curve shows the numerical solution for the same initial conditions and the same values of the ratios d0/(ω0A0)d_0/(\omega_0A_0) and d2A0/ω0d_2A_0/\omega_0. The dashed horizontal lines are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)|\leq\mu mg/k holds. See text for details.

Initial conditions (x0=0,v0>0)(x_0=0,v_0>0)

For this type of initial conditions, equation (43) simplifies to A=A0=v0/ω0A=A_0=v_0/\omega_0 and we get φ=φ0=−π/2\varphi=\varphi_0=-\pi/2, i.e. (47) and (49) become equal. The improved solution, i.e. (52), is, of course, also valid for these initial conditions, it is only necessary to determine the values (t0,x(t0))(t_0, x(t_0)) of the last turning point of the function Af(t)cos⁡(ω0t+φ)A f(t) \cos(\omega_0 t + \varphi) with property ∣x(t0)∣>μmg/k|x(t_0)| > \mu mg/k, and insert these values into expression (52). As an example of the dynamics with these type of initial conditions, in Figure 19(a) we show the solutions (52), (49) and the corresponding numerical solutions for d0/(ω0A0)=0.03d_0/(\omega_0 A_0) = 0.03 and d2A0/ω0=0.25d_2 A_0/\omega_0 = 0.25. In this case, (t0,x(t0))(t_0, x(t_0)) is the fourth turning point, solutions (52) and (49) overlap for 0<t≤t00 < t \leq t_0, and for t>t0t > t_0 the solution (52) agrees significantly better with the numerical solution compared to (49). In Figure 19(b) we show the energies corresponding to solutions shown in Figure 19(a).

Solutions and corresponding energies

Figure 19. (a) Solid red curves show the solution (52) with initial conditions (x0=0,v0=ω0A0)(x_0 = 0, v_0 = \omega_0 A_0), d0/(ω0A0)=0.03d_0/(\omega_0 A_0) = 0.03 and d2A0/ω0=0.25d_2 A_0/\omega_0 = 0.25. Solid blue curves show the corresponding solution (49) and black dotted curves show the corresponding numerical solution. The dashed horizontal lines are positioned at ±μmg/k\pm\mu mg/k, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. (b) The energies corresponding to solutions shown in figure (a). See text for details.

Initial conditions (x0≠0,v0≠0)(x_0 \ne0, v_0 \ne0)

As an example, here we choose (x0=A02/2,v0=ω0x0)(x_0 = A_0\sqrt{2}/2, v_0 = \omega_0 x_0), d0/(ω0A0)=0.03d_0/(\omega_0 A_0) = 0.03 and d2A0/ω0=0.25d_2 A_0/\omega_0 = 0.25. For this type of initial conditions, equation (43) remains a quartic equation in AA. Again, instead of delving into solving the corresponding quartic equation analytically, we solve it using the roots MATLAB function. Of the four solutions we obtained, the closest to the value A0=0.2mA_0 = 0.2m is A=0.234mA = 0.234m, which corresponds to the initial phase φ=−0.922\varphi= -0.922. In Figure 20(a) we show the solutions (52), (49) and the numerical solutions corresponding to this choice of the initial conditions and values of the ratios d0/(ω0A0)d_0/(\omega_0 A_0) and d2A0/ω0d_2 A_0/\omega_0. In this case, (t0,x(t0))(t_0, x(t_0)) used in the solution (52) is the fifth turning point. In Figure 20(b) we show the energies corresponding to the solutions shown in Figure 20(a).

Solution and energy curves

Figure 20. (a) Solid red curves show the solution (52) with initial conditions (x0=A02/2,v0=ω0x0)(x_0 = A_0\sqrt{2}/2, v_0 = \omega_0 x_0), for d0/(A0ω0)=0.03d_0/(A_0\omega_0) = 0.03 and d2A0/ω0=0.25d_2A_0/\omega_0 = 0.25. Solid blue curves show the corresponding solution (49) and black dotted curves show the corresponding numerical solution. The dashed horizontal lines are positioned at ±μmg/k\pm\mu mg/k, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. (b) The energies corresponding to solutions shown in figure (a). See text for details.

Since the displacement of the fifth turning point of the solid red curve, shown in Figure 20(a), is only slightly greater in magnitude than μmg/k\mu mg/k, we will also consider the solution which we obtain if we completely ignore the quadratic damping from the instant of the fourth turning point (t4,x(t4))(t_4, x(t_4)) onward, i.e. for t>t4t > t_4. Thus, we can use again the solution of the form (54), but with (t4,x(t4))=(1.84s,−0.146x0)(t_4, x(t_4)) = (1.84s, -0.146x_0) in place of (t3,x(t3))(t_3, x(t_3)), and with the values of AA and φ\varphi corresponding to initial conditions used here. The results are shown in Figure 21. Again, similarly as in Figure 18, we see that we can completely ignore the quadratic damping in the last two half-periods and still get a good description of the dynamics.

Solution curves

Figure 21. Solid red curve shows the solution x(t)x(t) of the form (54), but with (t4,x(t4))=(1.84 s,−0.146x0)(t_4, x(t_4)) = (1.84\,\mathrm{s}, -0.146x_0) in place of (t3,x(t3))(t_3, x(t_3)), with initial conditions (x0=A02/2,v0=ω0x0)(x_0 = A_0\sqrt{2}/2, v_0 = \omega_0x_0) for d0/(A0ω0)=0.03d_0/(A_0\omega_0) = 0.03 and d2A0/ω0=0.25d_2A_0/\omega_0 = 0.25. Solid blue curve shows the solution x~(t)\tilde{x}(t) given by (49) and black dotted curve shows the numerical solution for the same initial conditions and the same values of the ratios d0/(ω0A0)d_0/(\omega_0A_0) and d2A0/ω0d_2A_0/\omega_0. The dashed horizontal lines are positioned at corresponding ±μmg/k\pm\mu mg/k values, i.e. they indicate the region within which ∣x(t)∣≤μmg/k|x(t)| \leq\mu mg/k holds. See text for details.

Comments on results for harmonic oscillator with combined quadratic and Coulomb damping

The results presented in this section show that the new approximate analytical solutions (47) with the corresponding AA and φ\varphi, provide a significantly better description of the dynamics compared to the solutions (49) which are provided by the approach used in [14]. For the range of damping constants μ\mu and DD such that 0<d0/(ω0A0)≤0.030 < d_0/(\omega_0 A_0) \leq0.03 and 0<d2A0/ω0≤0.250 < d_2 A_0/\omega_0 \leq0.25 hold, solutions (47) provide a good description of the dynamics for all types of initial conditions, unless we are concerned with high precision at the very end of the dynamics, e.g. for the last half-period or last two half-periods before the halt.

In case we need greater precision in the description of the last or the last two half-periods before the halt, we introduced solutions (52) and (54), which significantly correct the limitations of solution (47). The trick was to exploit the physics of the problem, i.e. to completely ignore the influence of damping quadratic in velocity during the last or last two half-periods of the motion, due to small velocity during these time intervals, and take that only Coulomb damping decreases the oscillations towards the end of the dynamics.

In Figure 14(d), we can see, in the case d0/(ω0A0)=0.03d_0/(\omega_0A_0) = 0.03 and d2A0/ω0=0.25d_2A_0/\omega_0 = 0.25, that the amplitude of the solid red curve after the first period drops to about 30% of the initial value. Thus, the new approximate analytical solutions, introduced in this section, describe the dynamics well for damping that we could characterize (at least) as moderately strong.

Conclusion and Outlook

By comparing our new analytical solutions with numerical solutions, we have shown that they provide an excellent description of the dynamics of a harmonic oscillator damped by a force quadratic in velocity, sliding friction, and by the combination of these two damping forces. Our approach is valid for any type of initial conditions and for a wide range of values of the damping constants, which makes it suitable for the description and analysis of free damped vibrations of single-degree-of-freedom systems in the context of experiments in physics and engineering applications. The most elegant result is obtained for the damping quadratic in velocity analyzed in Section III. In that case, the initial amplitude AA is easily obtained analytically for any type of initial conditions by solving a simple quadratic equation. In the remaining two cases, depending on the type of initial conditions, it is necessary to solve the quartic equation to obtain initial amplitude AA, which is not easy to do analytically, but can be very easily solved numerically with standard and accessible software packages. Once we have determined AA, the initial phase φ\varphi is easy to calculate in any case.

Equation (22) suggests that the important combination of damping linear in velocity and damping quadratic in velocity will also be easily handled by our approach, since in that case (22) becomes a quadratic equation and the initial amplitude AA can be easily determined analytically for any type of initial conditions. This case is particularly important because in order to fully model the influence of air resistance, both types of damping are needed [11, 13, 15, 20]. Furthermore, we can expect that the approach presented in this paper will bring significant improvements in the description of the dynamics compared to the solutions already obtained in [14] for this combination of damping forces, because both damping linear in velocity and damping quadratic in velocity contribute to a significant change in the initial phase compared to the initial phase of the undamped case. Finally, the dynamics of every real-world system is also, to a greater or lesser extent, influenced by a force of sliding friction. Therefore, for a full description of the damping of free vibrations in real-world systems, it is necessary to take into account that all three types of damping are active, and how much each of them affects the dynamics depends, of course, on the specifics of the particular system we are considering, but also on the initial conditions (due to the nonlinear terms in the corresponding equation of motion). We will analyze in detail the new approximate analytical solutions for all remaining combinations of the three types of damping in a follow-up paper (i.e. in Part II).

Acknowledgments

K.L. and R.P. acknowledge support from the project “Implementation of cutting-edge research and its application as part of the Scientific Center of Excellence for Quantum and Complex Systems, and Representations of Lie Algebras”, Grant No. PK.1.1.10.0004, co-financed by the European Union through the European Regional Development Fund - Competitiveness and Cohesion Programme 2021-2027.

References

  1. [1]Frank S. Crawford. Waves: Berkeley Physics Course. volume 3. McGraw-Hill, New York, 1968.
  2. [2]J.D. Cutnell and K.W. Johnson. Physics. John Wiley & Sons, 2009.
  3. [3]David Halliday, Robert Resnick, and Jearl Walker. Fundamentals of Physics. John Wiley & Sons, 2013.
  4. [4]Hugh D. Young and Roger A. Freedman. University Physics with Modern Physics. Pearson, 2020.
  5. [5]I. R. Lapidus. Motion of a harmonic oscillator with sliding friction. American Journal of Physics, 38(11):1360–1361, nov 1970.DOI
  6. [6]A. Marchewka, David. S. Abbott, and R. J. BeichnerKamela. Oscillator damped by a constant-magnitude friction force. American Journal of Physics, 72(4):477–483, apr 2004.DOI
  7. [7]A. Anastasios Adamopoulosa and N. Adamopoulos. Constant and quadratic damping of free oscillations: easy solutions. International Journal of Mathematical Education in Science and Technology, 53(11):3151–3161, jan 2022.DOI
  8. [8]M. Kamela. An oscillating system with sliding friction. The Physics Teacher, 45(2):110–113, feb 2007.DOI
  9. [9]B. R. Jr. Smith. The quadratically damped oscillator: A case study of a non-linear equation of motion. American Journal of Physics, 80(9), sep 2012.DOI
  10. [10]C. E. Mungan and T. C. Lipscombe. Oscillations of a quadratically damped pendulum. European Journal of Physics, 34(5):1243–1253, jul 2013.DOI
  11. [11]X. Wang, C. Schmitt, and M. Payne. Oscillations with three damping effects. European Journal of Physics, 23(2):155–164, jan 2002.DOI
  12. [12]K. Karlo Lelas, Nikola Poljak, and Dario Jukić. Damped harmonic oscillator revisited: The fastest route to equilibrium. American Journal of Physics, 91(10):767–775, 10 2023.arxiv.org/abs/2301.08222
  13. [13]P. T. Squire. Pendulum damping. American Journal of Physics, 54(11):984–991, nov 1986.DOI
  14. [14]K. Lelas and R. Pezer. The amplitude decay of a harmonic oscillator damped simultaneously by weak linear and nonlinear damping forces. European Journal of Physics, 46(4):045001, jun 2025.arxiv.org/abs/2502.07458
  15. [15]R. A. Nelson and M. G. Olsson. The pendulum—rich physics from a simple system. American Journal of Physics, 54(2):112–121, feb 1986.DOI
  16. [16]M. V. Vitorino, A. Vieira, and M. S. Rodrigues. Effect of sliding friction in harmonic oscillators. Scientific Reports, 7:3726, jun 2017.DOI
  17. [17]K. Lelas and R. Pezer. Modeling the amplitude and energy decay of a weakly damped harmonic oscillator using the energy dissipation rate and a simple trick. European Journal of Physics, 46(1):015004, dec 2024.arxiv.org/abs/2406.18488
  18. [18]P. F. Hinrichsen and C. I. Larnder. Combined viscous and dry friction damping of oscillatory motion. American Journal of Physics, 86(8):577–584, aug 2018.
  19. [19]B. K. Karthik, R. B. Shreesh, V. Shrik​​anth, and A. K. Gaonkar. Prediction of energy dissipation by analytical solution to combined viscous and coulomb damping. Journal of Sound and Vibration, 573:118216, 2024.DOI
  20. [20]M. E. Bacon and D. D. Nguyen. Real-world damping of a physical pendulum. European Journal of Physics, 26(4):651–655, may 2005.DOI
  21. [21]L. Cveticanin. Oscillator with strong quadratic damping force. Publications de l’Institut Mathématique (Beograd), 85(99):119–130, 2009.DOI
  22. [22]L. Cveticanin. Oscillators with nonlinear elastic and damping forces. Computers & Mathematics with Applications, 62(4):1745–1757, aug 2011.DOI
  23. [23]J. A. Rizcallah. Revisiting the coulomb-damped harmonic oscillator. European Journal of Physics, 40(5), aug 2019.DOI

Paper details

vlnc:W_qsxy95wd
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents