betaFeedback

WorksPhysics and Astronomy

Fluid drags on a sphere at finite Reynolds number–a classical problem examined afresh

Abstract

Evaluating the drag coefficient Cd for a sphere moving in a fluid at finite Reynolds number Re has become a classical research subject in fluid dynamics, with publications appearing continuously by numerous authors over the years due to the unavailability of closed-form mathematical solutions. To date, mathematical formulas for calculating Cd as a function of Re can only be obtained by correlation curve fitting to experimental data. The consequence is the coexistence of various formulas with different mathematical fitting functions and an increasing number of fitting parameters. The present study focuses on examining the accuracy of Klyachko's 1934 formula for Cd with two common fractions as fitting parameters, which could enable closed form integration of a simplified equation governing rectilinear motion of aerosol spheres. The computed values of root-mean-square relative error from recently developed formulas show reasonable accuracy of the Klyachko formula for Re less than 1000 and its extension to Re less than 2e5 as a piecewise continuous function, with the possibility of significant improvement through a slight modification. Hence, a few demonstrative examples are presented using the extended Klyachko formula in the study of rectilinear acceleration, the calculation of the terminal velocity of aerosol spheres, and so forth, to shed fresh light on how this classical problem can be studied.

Introduction

As a fundamental canonical problem in fluid mechanics, flow past a sphere has been studied by numerous authors and discussed in well-known subject-matter books [14, 22, 15, 8, 6 Pruppacher] and Klett, 1997). Besides its theoretical convenience for the in-depth mathematical study of drag, boundary layer separation, and turbulence, the problem of a spherical body moving in a fluid often appears in a myriad of natural processes and engineering applications [8, 6, 18]. Despite the tremendous efforts by generations after generations of mathematically talented scientists, closed-form analytical solutions for flow around rigid and fluid spheres are effectively limited to very low values of the Reynolds number, Re≡ud/νRe \equiv ud/\nu (where dd denotes the sphere’s diameter, uu and ν\nu the relative velocity and kinematic viscosity of the surrounding fluid). Knowledge about the situation at higher Reynolds numbers must resort to experimental measurements, numerical computations, as well as approximations based on boundary-layer theory [6].

Dictated by the laws of physics, a sphere (or any object of another shape) will experience a drag force—as the surface integral over its whole surface of all stresses arising from the relative fluid flow—whenever there is a relative motion of it in the surrounding fluid [15]. In scientific literature, a dimensionless version of the drag force—the drag coefficient Cd≡8Fd/(πd2ρu2)C_d \equiv8F_d/(\pi d^2\rho u^2) (with FdF_d denoting the drag force and ρ\rho the density of surrounding fluid)—has commonly been utilized for convenience of analytical description [22, 15, 8, 6, 18, 20, 9, 17].

The drag coefficient of a sphere in steady motion, in an incompressible Newtonian fluid, CdC_d is often presented as a graph versus ReRe known as the ”Rayleigh curve” or “standard drag curve” [6], which has been expressed in terms of mathematical functions with adjustable fitting parameters by various authors via empirical or semiempirical means with persistent correlation efforts [e.g. 21, 12, 16, 4, 5], including the usage of a genetic programming procedure [2] and machine learning with a symbolic regression algorithm [11]. In view of the Cheng [5] comparisons with compiled experimental data from multiple sources, the reported values of the “average relative error” of correlation formulas from different authors, with vastly different mathematical complexities, are mostly within ∼5%\sim5\%, e.g., 3.019% of Clift, Grace, and Weber [6], 3.236% of Brown and Lawler [4], 2.469% of Cheng [5]. and so on so forth.

Valuable as it might be for a percent or so improvement in the correlation relative error, publications by many recent authors [4, 5, 2, 13, 11] seem to focus only on sophisticated correlation formulas, often expressed in terms of transcendental functions with increasing number of the fitting parameters, for 0<Re<2×1050 < Re < 2 \times10^5, while completely omitting (or no more than barely mentioning) the much cleaner formula with only two fitting parameters, Cd=24(1+αReβ)/ReC_d = 24(1 + \alpha Re^\beta)/Re, suggested by Schiller and Naumann [21] (with α=0.15\alpha= 0.15 and β=0.687\beta= 0.687), Klyachko [12] (with common fractions α=1/6≈0.167\alpha= 1/6 \approx0.167 and β=2/3≈0.667\beta= 2/3 \approx0.667), among others. According to Fuchs [8], the Klyachko formula appears quite accurate, agreeing with the experimental data at least for 3<Re<4003 < Re < 400 (whereas Klyachko [12] applied it for 0<Re≤10000 < Re \le1000).

Interestingly, a formula very similar to that of Klyachko (with only the parameter α=1/6\alpha= 1/6 replaced by 0.158) was adopted by Serafini [23] (who claimed to agree with experimental values for 0.5<Re<5000.5 < Re < 500 but without citing the previous work of Klyachko), in the derivation of a closed-form analytical solution. Later, Putnam (1961) extended the validity of the Klyachko formula to Re<1000Re < 1000 (also without citing prior publications by either Klyachko or Serafini), when demonstrating the existence of an analytical solution (which had already been used by Klyachko in 1934). What has been implemented in the computational fluid dynamics (CFD) software package OpenFOAM® for the “sphere drag” calculation is in terms of a piecewise continuous function

Cd(Re)={24Re(1+16Re2/3),for 0<Re<Rem=100024Rem(1+16Rem2/3)=0.424,for Re≥Rem=1000(1)C_d(Re) = \begin{cases} \frac{24}{Re}\left(1+\frac{1}{6}Re^{2/3}\right), & \text{for } 0 < Re < Re_m = 1000 \\ \frac{24}{Re_m}\left(1+\frac{1}{6}Re_m^{2/3}\right)=0.424, & \text{for } Re \ge Re_m = 1000 \end{cases} \tag*{(1)}

following the KIVA-II work of Amsden, O’Rourke, and Butler [1] (which would actually be invalid beyond Re∼2×105Re \sim2 \times10^5 when the abrupt “drag crisis” occurs). It is effectively an extension of the Klyachko formula for ReRe up to ∼2×105\sim2 \times10^5 (though Klyachko suggested ∼3×105\sim3 \times10^5 as the extension upper limit, which is realistically over-aggressive and should be ignored) by continuously connecting it to a constant of 4(6+Rem2/3)/Rem4(6+Re_m^{2/3})/Re_m at Re=RemRe=Re_m, where the value of RemRe_m is chosen as 1000 by Amsden et al. in (1).

The mathematical function of (1) for 0<Re<10000 < Re < 1000, though significantly nonlinear, was shown to enable the derivation of a closed-form analytical solution to the differential equation governing the rectilinear motion of a sphere in a fluid [23, 19], when forces (such as added mass and Basset-Boussinesq history force) other than the steady-flow drag are negligible, such as situations often encountered with aerosol particles [8]. A closed-form analytical solution in fluid dynamics of a moving sphere is generally considered intractable, and highly appreciated for its mathematical elegance even though it is useful only for special circumstances in a limited interval of ReRe values. Therefore, it deserves special attention in the scientific community.

In what follows, we start by examining the numerical differences between (1) and those of the more recent formulas, to shed light on how good (or bad) the extended Klyachko formula can be, thereby closing a knowledge gap in the scientific literature. Then, a few examples are presented for practical applications using the Cd(Re)C_d(Re) given by (1) or its improved version.

Comparison of Cd(Re)C_d(Re) with a reference formula

To measure the difference (or “error”) between the extended Klyachko formula (1) and a recently developed correlation formula, such as that of Brown and Lawler [4] or others, the root-mean-square relative error (RMSRE):

RMSRE⁡=1b−a∫ab[f(x)−F(x)F(x)]2 dx(2)\operatorname{RMSRE} = \sqrt{\frac{1}{b-a}\int_a^b \left[\frac{f(x)-F(x)}{F(x)}\right]^2 \,\mathrm{d}x} \tag*{(2)}

is used here, where f(x)f(x) denotes the test function and F(x)F(x) the reference function.

Table 1 presents the values of RMSRE (2) with f(x)f(x) for the Cd(Re)C_d(Re) in (1) and the reference function F(x)F(x) for the correlation formulas for 480 points “subcritical” experimental data for Re<2×105Re < 2 \times10^{5} by recent authors, i.e.,

Cd=24Re(1+0.15Re0.681)+0.407ReRe+8710C_d = \frac{24}{Re}(1 + 0.15Re^{0.681}) + \frac{0.407Re}{Re+8710}

by Brown and Lawler [4],

Cd=24Re(1+0.27Re)0.43+0.47[1−exp⁡(−0.04Re0.38)]C_d = \frac{24}{Re}(1 + 0.27Re)^{0.43} + 0.47[1-\exp(-0.04Re^{0.38})]

by Cheng [5],

Cd=5.4856×109tanh⁡(4.3774×10−9/Re)+0.0709tanh⁡(700.6575/Re)+0.3894tanh⁡(74.1539/Re)−0.1198tanh⁡(7429.0843/Re)+1.7174tanh⁡[9.9851/(Re+2.3384)]+0.4744\begin{aligned} C_d ={}& 5.4856 \times10^{9}\tanh(4.3774 \times10^{-9}/Re) + 0.0709\tanh(700.6575/Re) \\ &+0.3894\tanh(74.1539/Re) - 0.1198\tanh(7429.0843/Re) \\ &+1.7174\tanh[9.9851/(Re+2.3384)] + 0.4744 \end{aligned}

by Barati, Neyshabouri, and Ahmadi [2], integrated from Re=0.01Re = 0.01 to Re=1000Re = 1000, 900, 800, and 500 as well as Re=1000Re = 1000 to Re=2×105Re = 2 \times10^{5}. Apparently, for 0.01<Re<10000.01 < Re < 1000, the RMSRE value is consistently lowest for (1) versus the correlation formula of Cheng [5], while the differences between (1) and all three correlation formulas of Brown and Lawler [4], Cheng [5], and Barati, Neyshabouri, and Ahmadi [2] have RMSRE values less than 5%. Most of the differences appear to come from the interval of 800<Re<1000800 < Re < 1000, because a close examination shows that (1) exhibits a gradual deviation from the “standard curve” of CdC_d versus ReRe for Re>500Re > 500 up to by almost 8% underestimation around Re=1000Re = 1000 where Cd∼0.46C_d \sim0.46 has been consistently measured in experiments. But for 1000<Re<2×1051000 < Re < 2 \times10^{5} (where the experimental data vary from Cd∼0.40C_d \sim0.40 in 2000<Re<1042000 < Re < 10^{4} to Cd∼0.47C_d \sim0.47 in 4×104<Re<2×1054 \times10^{4} < Re < 2 \times10^{5}), the RMSRE values for the simple approximation Cd=0.424C_d = 0.424 versus those of the sophisticated correlation formulas become slightly over 10%, coming mostly from the interval of 105<Re<2×10510^{5} < Re < 2 \times10^{5}. Nonetheless, these few percentage points of differences between (1) and more recent correlation formulas appear quite insignificant when viewed in the graph of Figure 1 (where the relatively low RMSRE case of Cheng is omitted without loss of illustrative effectiveness).

Figure 1. Graphic comparison between Klyachko’s formula expressed as Eq. (1) and more recent correlation formulas for drag coefficient [4, 2]

It should be noted that Table 1 also reveals the fact that a significant reduction of the RMSRE values can be achieved by simply moving the upper limit of the Klyachko formula in (1) from Rem=1000Re_m = 1000 to 900, for a modified formula with improved accuracy (especially for Re>1000Re > 1000 up to Re∼2×105Re \sim2 \times10^{5} with the constant value of CdC_d for Re>RemRe > Re_m effectively increased from 0.424 to ≈0.441\approx0.441),

ReRe intervalBrown and LawlerChengBarati et al.
(0.01,1000)(0.01, 1000)0.04340.04110.0464
(0.01,900)(0.01, 900)0.03730.03320.0396
(0.01,800)(0.01, 800)0.03150.02560.0337
(0.01,500)(0.01, 500)0.01980.01120.0285
(1000,105)(1000, 10^{5})0.07530.07110.0734
(1000,2×105)(1000, 2 \times10^{5})0.11090.12060.1155
(Rem=900,105)(Re_m = 900, 10^{5})0.05390.04660.0525
(Rem=900,2×105)(Re_m = 900, 2 \times10^{5})0.05290.05860.0562

Table 1. RMSRE values for Cd(Re)C_d(Re) of Klyachko’s formula expressed as Eq. (1) versus that obtained by Brown and Lawler [4], Cheng [5], and Barati, Neyshabouri, and Ahmadi [2].

Cd(Re)={24Re(1+16Re2/3),for 0<Re<Rem=9004(6+Rem2/3)Rem≈0.441,for Rem=900≤Re<2×105(3)C_d(Re) = \begin{cases} \frac{24}{Re}\left(1+\frac{1}{6}Re^{2/3}\right), & \text{for } 0 < Re < Re_m = 900 \\ \frac{4(6+Re_m^{2/3})}{Re_m} \approx0.441, & \text{for } Re_m = 900 \le Re < 2 \times10^{5} \end{cases} \tag*{(3)}

As a comparative reference, the RMSRE values calculated for Brown and Lawler versus Barati et al. and Cheng versus Barati et al. in the interval from Re=0.01Re = 0.01 to 1000 are ∼0.0228\sim0.0228 and 0.0161, respectively. Obviously, a few percent of disagreements among even those recent correlation formulas appear inevitable. Thus, the accuracy of either (1) or (3) can be quite acceptable.

In view of numerical differences among various available correlation formulas and the noisiness of experimental data, the comparison between (1) and recently developed “more accurate” correlation formulas appears quite reasonable for practical real-world applications, especially for Re<1000Re < 1000. The “accuracy” of (1) is shown here to further improve with a slight intuitive modification like (3),

There seems to be no agreed-upon best metric for evaluating the correlation accuracy; different authors sometimes choose different sets of metrics in presenting correlation results. Even when RMSRE is used, the formula (2) is but just one form for straightforward calculations, while it can have a corresponding discrete version as:

RMSRE=1N∑i=1N[f(xi)−F(xi)F(xi)]2\mathrm{RMSRE} = \sqrt{\frac{1}{N}\sum_{i=1}^{N}\left[\frac{f(x_i)-F(x_i)}{F(x_i)}\right]^2}

the value of which could be sensitive to how those NN data points are distributed. If so desired, those NN data points can be evenly distributed in a logarithmic scale of ReRe as shown in Figure 1 for discrete points of “Brown and Lawler” and “Barati et al.” Then, the discrete RMSRE values of ReRe in (0.01,1000)(0.01,1000) become 0.0290, 0.0345, 0.0375 for (1) versus “Brown and Lawler”, “Cheng”, “Barati et al.”, respectively. For ReRe in (1000,105)(1000,10^{5}), those discrete RMSRE values are 0.0685, 0.0608, 0.0687. Despite slight numerical variations, whether to use (2) or its discrete version does not seem to make a fundamental difference in the present case when comparing (1) with other correlation formulas.

In terms of the discrete RMSRE, the value for the Brown and Lawler formula versus their set of N=480N = 480 compiled experimental data is calculated as 0.0411, and that for the Cheng formula as 0.0346. On the other hand, [4, 5, 2] also presented their correlation results using the same set of experimental data in terms of the “average relative error”, 1N∑(∣Cd,calc−Cd,exp∣/Cd,exp)\frac{1}{N}\sum(\lvert C_{d,\mathrm{calc}}-C_{d,\mathrm{exp}}\rvert/C_{d,\mathrm{exp}}), calculated as 0.0324, 0.0247, 0.0223. No matter which metric is employed, the apparent accuracy improvements over the years seem to have approached their limit, differing by no more than a few percent from inevitably noisy experimental data. Therefore, those recent correlation formulas can be reasonably regarded as trustworthy surrogates for the actual experimental data.

Particle rectilinear acceleration

For the case of quasistatic drag as the only dominant force controlling the rectilinear motion of a spherical particle (with particle density ρp\rho_p and velocity u(t)u(t)), Newton’s law of motion based on (3) for the “initial particle Reynolds number” Re0≡u(0)d/ν<2×105Re_0 \equiv u(0)d/\nu< 2 \times10^5 can be written as

du~dt~={−3ρ(6+Rem2/3)ρpRemu~2≈0.331ρρpu~2,for 900≤Re0u~<2×105,−18ρRe0ρp(1+Re02/36u~2/3)u~,for 0<Re0u~<900,(4)\frac{d\tilde{u}}{d\tilde{t}} = \begin{cases} -\frac{3\rho(6+Re_m^{2/3})}{\rho_p Re_m}\tilde{u}^2 \approx\frac{0.331\rho}{\rho_p}\tilde{u}^2, & \text{for } 900 \le Re_0\tilde{u} < 2 \times10^5,\\ -\frac{18\rho}{Re_0\rho_p}\left(1+\frac{Re_0^{2/3}}{6}\tilde{u}^{2/3}\right)\tilde{u}, & \text{for } 0 < Re_0\tilde{u} < 900, \end{cases} \tag*{(4)}

where t~=tu(0)/d\tilde{t}=tu(0)/d and u~=u/u(0)\tilde{u}=u/u(0).

For the case of Re0>Rem=900Re_0 > Re_m = 900, a straightforward integration of (4a) for u~(0)=1\tilde{u}(0)=1 yields

u~(t~)=[1+(0.331ρρp)t~]−1,for t~≤t~m=ρp0.331ρ(Re0−RemRem).(5)\tilde{u}(\tilde{t})=\left[1+\left(\frac{0.331\rho}{\rho_p}\right)\tilde{t}\right]^{-1},\quad\text{for }\tilde{t}\le\tilde{t}_m=\frac{\rho_p}{0.331\rho}\left(\frac{Re_0-Re_m}{Re_m}\right). \tag*{(5)}

Then, for t~>t~m\tilde{t}>\tilde{t}_m, it turns out (though may not be immediately obvious) that (4b) can be rearranged as

d(u~−2/3)dt~=12ρRe0ρp(u~−2/3+Re02/36),\frac{d\left(\tilde{u}^{-2/3}\right)}{d\tilde{t}}=\frac{12\rho}{Re_0\rho_p}\left(\tilde{u}^{-2/3}+\frac{Re_0^{2/3}}{6}\right),

which yields an analytical solution for u~(t~)<u~(t~m)=Rem/Re0\tilde{u}(\tilde{t})<\tilde{u}(\tilde{t}_m)=Re_m/Re_0, similar to that shown by Serafini [23] and Putnam [19] (equivalent to the case of Rem=Re0Re_m=Re_0 such that t~m=0\tilde{t}_m=0 and Rem/Re0=1Re_m/Re_0=1),

u~(t~)=[(Re02/3Rem2/3+Re02/36)exp⁡(12ρRe0ρpΔt~)−Re02/36]−3/2=66Re0Φ(t~)3,(6)\tilde{u}(\tilde{t})=\left[\left(\frac{Re_0^{2/3}}{Re_m^{2/3}}+\frac{Re_0^{2/3}}{6}\right)\exp\left(\frac{12\rho}{Re_0\rho_p}\Delta\tilde{t}\right)-\frac{Re_0^{2/3}}{6}\right]^{-3/2}=\frac{6\sqrt{6}}{Re_0\Phi(\tilde{t})^3}, \tag*{(6)}

where Δt~≡t~−t~m\Delta\tilde{t}\equiv\tilde{t}-\tilde{t}_m and

Φ(t~)≡(6Rem2/3+1)exp⁡(12ρRe0ρpΔt~)−1.\Phi(\tilde{t})\equiv\sqrt{\left(\frac{6}{Re_m^{2/3}}+1\right)\exp\left(\frac{12\rho}{Re_0\rho_p}\Delta\tilde{t}\right)-1}.

As Serafini [23] pointed out, it is the exponent 2/32/3 in (4b) that enables closed-form integration. (In other words, the formula of Schiller and Naumann [21], though very similar to that of Klyachko in mathematical form but with an exponent β\beta slightly deviating from 2/32/3, cannot facilitate a closed-form integration in a rigorous sense.)

By integrating u~(t~)\tilde{u}(\tilde{t}) given by (5) from 0 to t~\tilde{t} (if t~≤t~m\tilde{t}\le\tilde{t}_m for Re0>RemRe_0>Re_m) and (6) from t~m\tilde{t}_m to t~\tilde{t} (when t~>t~m\tilde{t}>\tilde{t}_m), the particle flight distance (in units of dd) as a function of flight time can also be derived as

z~(t~)={ρp0.331ρln⁡(1+0.331ρt~ρp),for t~≤t~m,ρp0.331ρln⁡(1+0.331ρt~mρp)+6ρpρ{1Φ(t~m)−1Φ(Δt~)+tan⁡−1[Φ(t~m)]−tan⁡−1[Φ(Δt~)]},for t~>t~m.(7)\tilde{z}(\tilde{t}) = \begin{cases} \frac{\rho_p}{0.331\rho}\ln\left(1+\frac{0.331\rho\tilde{t}}{\rho_p}\right), & \text{for }\tilde{t}\leq\tilde{t}_m,\\[1em] \frac{\rho_p}{0.331\rho}\ln\left(1+\frac{0.331\rho\tilde{t}_m}{\rho_p}\right)+\sqrt{6}\frac{\rho_p}{\rho}\left\{\frac{1}{\Phi(\tilde{t}_m)}-\frac{1}{\Phi(\Delta\tilde{t})}+\tan^{-1}[\Phi(\tilde{t}_m)]-\tan^{-1}[\Phi(\Delta\tilde{t})]\right\}, & \text{for }\tilde{t}>\tilde{t}_m. \end{cases} \tag*{(7)}

Then, the stop distance (namely, the maximum achievable flight distance as t~→∞\tilde{t}\rightarrow\infty), often used to evaluate possible nozzle-to-substrate standoff in drop-on-demand inkjet printing for an ejected droplet, when Re0>RemRe_0>Re_m becomes

z~(∞)=ρp0.331ρln⁡(1+0.331ρt~mρp)+6ρpρ{1Φ(t~m)+tan⁡−1[Φ(t~m)]−π2}.(8)\tilde{z}(\infty)=\frac{\rho_p}{0.331\rho}\ln\left(1+\frac{0.331\rho\tilde{t}_m}{\rho_p}\right)+\sqrt{6}\frac{\rho_p}{\rho}\left\{\frac{1}{\Phi(\tilde{t}_m)}+\tan^{-1}[\Phi(\tilde{t}_m)]-\frac{\pi}{2}\right\}. \tag*{(8)}

For the case of Re0≤RemRe_0\leq Re_m, we should have Rem/Re0Re_m/Re_0 replaced by 1 and t~m=0\tilde{t}_m=0. Thus (8) can be written as Serafini, [23]; Putnam [19]

z~(∞)=6ρpρ[Re02/36+tan⁡−1(6Re02/3)−π2],(9)\tilde{z}(\infty)=\sqrt{6}\frac{\rho_p}{\rho}\left[\sqrt{\frac{Re_0^{2/3}}{6}}+\tan^{-1}\left(\sqrt{\frac{6}{Re_0^{2/3}}}\right)-\frac{\pi}{2}\right], \tag*{(9)}

because Φ(0)=6Re02/3\Phi(0)=\sqrt{\frac{6}{Re_0^{2/3}}}.

It should be noted that (4) and subsequent analytical formulas (5)-(9) are valid only when ρp≫ρ\rho_p\gg\rho (as in the case of aerosols), such that the effects of time-dependent components of drag (such as added mass and Basset-Boussinesq history force included in the discussion of Landau and Lifshitz, 1959 as well as Fuchs, 1964) are negligible, and also when the effect of gravity could be ignored, i.e., the terminal velocity due to gravity is much smaller than the sphere velocity magnitude of interest, as in the study of dust particles in dust-recollecting devices Klyachko, [12], droplet impingement onto an airfoil Serafini, [23], fuel droplet spray in a combustion chamber [19, 1], among others.

Particle terminal settling velocity

With a given Cd(Re)C_d(Re) for a sphere moving in a fluid, the terminal velocity vv of a spherical particle with density ρp\rho_p settling under the influence of gravity can be readily calculated based on force balance:

πd2ρv2Cd8=πd3(ρp−ρ)g6or3Re2Cd4=d3gν2(ρpρ−1),(10)\frac{\pi d^2\rho v^2 C_d}{8}=\frac{\pi d^3(\rho_p-\rho)g}{6}\quad\text{or}\quad\frac{3Re^2C_d}{4}=d^3\frac{g}{\nu^2}\left(\frac{\rho_p}{\rho}-1\right), \tag*{(10)}

where gg denotes gravitational acceleration. Noteworthy here is that 3Re2Cd/43Re^{2}C_d/4 is the Archimedes number (also called the Best number), while Re2CdRe^{2}C_d is known as the Davies number, often used by atmospheric scientists in studying the terminal velocity of water droplets in air [7, 3].

If the particle diameter and velocity are nondimensionalized as

d∗=d[gν2(ρpρ−1)]1/3=(3CdRe24)1/3and(11)d_* = d\left[\frac{g}{\nu^2}\left(\frac{\rho_p}{\rho}-1\right)\right]^{1/3} = \left(\frac{3C_dRe^2}{4}\right)^{1/3} \quad\text{and} \tag*{(11)}
u∗=u[gν(ρpρ−1)]−1/3=(4Re3Cd)1/3,(12)u_* = u\left[g\nu\left(\frac{\rho_p}{\rho}-1\right)\right]^{-1/3} = \left(\frac{4Re}{3C_d}\right)^{1/3}, \tag*{(12)}

we have u∗×d∗=Reu_* \times d_* = Re as suggested by Brown and Lawler [4].

With CdC_d given by (1) or (3), d∗d_* and u∗u_* can be easily calculated as functions of ReRe, as

d∗(Re)={[Re(18+3Re2/3)]1/3,for 0<Re<Rem,(18+3Rem2/3Rem)1/3Re2/3,for Rem≤Re<2×105(13)d_*(Re) = \begin{cases} \left[Re\left(18+3Re^{2/3}\right)\right]^{1/3}, & \text{for } 0<Re<Re_m,\\ \left(\frac{18+3Re_m^{2/3}}{Re_m}\right)^{1/3}Re^{2/3}, & \text{for } Re_m\le Re<2\times10^5 \end{cases} \tag*{(13)}

and then u∗(Re)=Re/d∗(Re)u_*(Re)=Re/d_*(Re),

where we have Rem=1000Re_m=1000 with (1) or Rem=900Re_m=900 with (3), if so chosen.

However, most authors present curve-fitting formulas for directly calculating the terminal velocity from a given particle size, namely, u∗(d∗)u_*(d_*) [4, 5]. In view of the fact that Re=u∗×d∗Re=u_*\times d_* and CdC_d as functions of ReRe, (11a) is equivalent to d∗=fd(u∗d∗)d_*=f_d(u_*d_*) and (11b) u∗=fu(u∗d∗)u_*=f_u(u_*d_*), none of which could readily be converted to an explicit function of u∗(d∗)u_*(d_*). If (11) and (12) are used as a parametric representation for generating tabulated data or plot, complicated numerical inversion can be avoided. Occasionally though, explicit solutions of u∗(d∗)u_*(d_*) can be obtained when a very simplified form of Cd(Re)C_d(Re) was adopted, like that used by Zigrang and Sylvester [25], or the formula of (12) for Re>RemRe>Re_m which yields

u∗=Remd∗18+3Rem2/3,u_*=\sqrt{\frac{Re_m d_*}{18+3Re_m^{2/3}}},

e.g., u∗≈1.7733d∗u_*\approx1.7733\sqrt{d_*} for Rem=1000Re_m=1000 or u∗≈1.7389d∗u_*\approx1.7389\sqrt{d_*} for Rem=900Re_m=900 where the difference seems negligible with either choice.

Obviously, calculating d∗d_* and u∗u_* from a given ReRe according to (12) can be much more straightforward and less cumbersome; only a mindset is needed here for the calculation procedure starting with a given value of ReRe. In generating a lookup table or graph of u∗u_* versus d∗d_* for practical usage, which calculation procedure to follow usually does not make any difference.

Table 2 and figure 2 exemplify tabulated values of d∗d_*, u∗u_* generated based on (12) for Rem=1000Re_m=1000 in comparison with the correlation formulas:

u∗=[(18d∗2)0.898(0.936d∗+1)/(d∗+1)+(0.317d∗)0.449]−1.114u_*=\left[\left(\frac{18}{d_*^2}\right)^{0.898(0.936d_*+1)/(d_*+1)}+\left(\frac{0.317}{d_*}\right)^{0.449}\right]^{-1.114}

by Brown and Lawler [4],

ReRed∗d_\astu∗u_\astBrown and LawlerCheng
0.010.566070.0176660.01903470.0177829
0.030.8186700.0366450.03926480.0371126
0.11.230830.0812460.08566510.0832422
0.31.797050.1669410.1722530.173591
12.758920.3624600.3656450.380909
34.174000.7187340.7203000.743474
106.834541.4631561.4940561.454615
3011.210652.6760262.8130762.629949
10020.217064.9463195.2261175.001796
30035.760938.3890448.4387028.609237
100068.2562414.6506713.1545214.07184
3000141.978221.1300020.2318321.4360
1×1041 \times10^{4}316.816331.5640331.1079331.49032
3×1043 \times10^{4}659.004445.5232145.3365644.28541
1×1051 \times10^{5}1470.53168.0026568.0361565.02726

Table 2. Parametrically calculated values of d∗d_\ast and u∗u_\ast based on Eq. (13) for Rem=1000Re_m = 1000 from given ReRe compared with the correlation formulas of u∗(d∗)u_\ast(d_\ast) obtained by Brown and Lawler [4], and Cheng [5].

u∗=4d∗41296(1+0.022d∗3)0.54+1.41d∗3[1−exp⁡(−0.15d∗0.45)]u_\ast= \sqrt{\frac{4d_\ast^{4}}{1296(1+0.022d_\ast^{3})^{0.54}+1.41d_\ast^{3}\left[1-\exp\left(-0.15d_\ast^{0.45}\right)\right]}}

by Cheng [5], and the corresponding graph of u∗u_\ast versus d∗d_\ast. Again not surprisingly, the results of d∗(Re)d_\ast(Re) given by (12) for Rem=1000Re_m = 1000, based on as a slightly less accurate Cd(Re)C_d(Re) formula (1), are still in reasonable agreement with those of more recent correlation formulas.

One of the most studied cases of terminal velocity is a water droplet falling in air, with ρp/ρ=1000/1.2\rho_p/\rho= 1000/1.2, ν=1.5×10−5 (m2 s−1)\nu= 1.5 \times10^{-5}\,(\mathrm{m}^{2}\,\mathrm{s}^{-1}), and g=9.81 (m s−2)g = 9.81\,(\mathrm{m}\,\mathrm{s}^{-2}), because of its direct relevance to cloud physics and rainfall rate in meteorology. Thus, we have d=0.030205×d∗d = 0.030205 \times d_\ast (mm) and u=0.49661×u∗u = 0.49661 \times u_\ast (m/s), according to (11a) and (11b). For example, at Re=270Re = 270 (or 70), eq. (12) yields d∗∼33.82d^\ast\sim33.82 (or 16.90) corresponding to d∼1.02d \sim1.02 (or 0.51) mm, and u∗∼270/33.82=7.98u^\ast\sim270/33.82 = 7.98 (or 70/16.90=4.1470/16.90 = 4.14) with u∼3.96u \sim3.96 (or 2.06) m/s, fairly close to the measured values tabulated by Gunn and Kinzer [10], e.g., for d=1d = 1 (or 0.5) mm, u=403u = 403 (or 2.06) m/s with calculated Re=269Re = 269 (or 68.7).

Another important problem closely related to terminal velocity is the dynamical process of the acceleration of a sphere toward its terminal velocity, often needed for evaluating the required free fall length in an experimental setup (Wang and Pruppacher [24]), among other applications. Similar to that with (4), assuming quasistatic drag but with the addition of a term representing the gravitational force in the equation for rectilinear motion (for Re<RemRe < Re_m):

Graphic comparison between parametrically calculated $u_\ast$ versus $d_\ast$ based on Eq. (12) for $Re_m=1000$ and correlation formulas by Brown and Lawler and Cheng

Figure 2. Graphic comparison between parametrically calculated u∗u_\ast versus d∗d_\ast based on Eq. (13) for Rem=1000Re_m=1000 and more recent correlation formulas of u∗(d∗)u_\ast(d_\ast) by Brown and Lawler [4] and Cheng [5]

du~dt~=3ρ(6+Re2/3)ρpRe(1−6u~+Re2/3u~5/36+Re2/3),with Re≡u(∞)dν(14)\frac{d\tilde{u}}{d\tilde{t}}=\frac{3\rho\left(6+Re^{2/3}\right)}{\rho_p Re}\left(1-\frac{6\tilde{u}+Re^{2/3}\tilde{u}^{5/3}}{6+Re^{2/3}}\right),\quad\text{with } Re\equiv\frac{u(\infty)d}{\nu} \tag*{(14)}

denoting the particle Reynolds number for terminal velocity u(∞)u(\infty), where dd is the sphere diameter, with 0≤t~=tu(∞)/d<∞0\leq\tilde{t}=tu(\infty)/d<\infty and 0≤u~=u/u(∞)≤10\leq\tilde{u}=u/u(\infty)\leq1. Here,

3ρ(6+Re2/3)ρpRe=ρgdρpu(∞)2(ρpρ−1)(15)\frac{3\rho\left(6+Re^{2/3}\right)}{\rho_p Re}=\frac{\rho g d}{\rho_p u(\infty)^2}\left(\frac{\rho_p}{\rho}-1\right) \tag*{(15)}

reflects the fact of gravitational-force-driven motion.

Although differing from (4) only by a constant term, (13) does not admit a closed-form analytical solution. However, it would not be difficult to compute numerical solutions of u~(t~)\tilde{u}(\tilde{t}) by 4th-order Runge-Kutta integration of (13), and then z~(t~)\tilde{z}(\tilde{t}) by integration of u~\tilde{u}, from 0 to t~\tilde{t}, for generating results for u~\tilde{u} (in units of u(∞)u(\infty)) versus z~\tilde{z} (in units of dd).

In the case of Re>RemRe>Re_m, (13) needs to be rewritten as

du~dt~=3ρ(6+Rem2/3)ρpRem×{1−(RemRe)(6u~+Re2/3u~5/36+Rem2/3),for 0<t~≤t~m,1−u~2,for t~≥t~m,(16)\frac{d\tilde{u}}{d\tilde{t}}=\frac{3\rho\left(6+Re_m^{2/3}\right)}{\rho_p Re_m}\times \begin{cases} 1-\left(\frac{Re_m}{Re}\right)\left(\frac{6\tilde{u}+Re^{2/3}\tilde{u}^{5/3}}{6+Re_m^{2/3}}\right), & \text{for } 0<\tilde{t}\leq\tilde{t}_m,\\ 1-\tilde{u}^{2}, & \text{for }\tilde{t}\geq\tilde{t}_m, \end{cases} \tag*{(16)}
(17)\text{} \tag*{(17)}

where (17) for t~>t~m\tilde{t} > \tilde{t}_{m} leads to an analytical solution

u~(t~)=tanh⁡[3ρ(6+Rem2/3)ρpRem(t~−t~m)+tanh⁡−1(RemRe)].(18)\tilde{u}(\tilde{t}) = \tanh\left[\frac{3\rho(6+Re_{m}^{2/3})}{\rho_{p}Re_{m}}(\tilde{t}-\tilde{t}_{m})+\tanh^{-1}\left(\frac{Re_{m}}{Re}\right)\right]. \tag*{(18)}

Here, the value of t~m\tilde{t}_{m} is determined by numerically integrating (16) with respect to t~\tilde{t} until u~=Rem/Re\tilde{u}=Re_{m}/Re. Thus, at t~=t~m\tilde{t}=\tilde{t}_{m} we have du~/dt~=3ρ(6+Rem2/3)(Re2−Rem2)/(ρpRemRe2)d\tilde{u}/d\tilde{t}=3\rho(6+Re_{m}^{2/3})(Re^{2}-Re_{m}^{2})/(\rho_{p}Re_{m}Re^{2}).

Integrating the u~(t~)\tilde{u}(\tilde{t}) of (18) from t~m\tilde{t}_{m} to t~\tilde{t} yields the particle flight distance

z~(t~)=z~(t~m)+ρpRem3ρ(6+Rem2/3)ln⁡{cosh⁡[3ρ(6+Rem2/3)ρpRem(t~−t~m)+tanh⁡−1(RemRe)]cosh⁡[tanh⁡−1(Rem/Re)]}.(19)\tilde{z}(\tilde{t})=\tilde{z}(\tilde{t}_{m})+\frac{\rho_{p}Re_{m}}{3\rho(6+Re_{m}^{2/3})}\ln\left\{\frac{\cosh\left[\frac{3\rho(6+Re_{m}^{2/3})}{\rho_{p}Re_{m}}(\tilde{t}-\tilde{t}_{m})+\tanh^{-1}\left(\frac{Re_{m}}{Re}\right)\right]}{\cosh\left[\tanh^{-1}(Re_{m}/Re)\right]}\right\}. \tag*{(19)}

Because du~/dt~=u~du~/dz~d\tilde{u}/d\tilde{t}=\tilde{u}d\tilde{u}/d\tilde{z}, an analytical formula explicitly relating z~\tilde{z} to u~\tilde{u} (or u~\tilde{u} to z~\tilde{z}) can also be obtained, for t~>t~m\tilde{t}>\tilde{t}_{m} when u~>Rem/Re\tilde{u}>Re_{m}/Re, based on (17), e.g.,

z~(u~)=z~(Rem/Re)+ρpRem6ρ(6+Rem2/3)ln⁡[1−(Rem/Re)21−u~2],(20)\tilde{z}(\tilde{u})=\tilde{z}(Re_{m}/Re)+\frac{\rho_{p}Re_{m}}{6\rho(6+Re_{m}^{2/3})}\ln\left[\frac{1-(Re_{m}/Re)^{2}}{1-\tilde{u}^{2}}\right], \tag*{(20)}

where the value of z~(Rem/Re)\tilde{z}(Re_{m}/Re) should be determined by numerical integration of (16) from t~=0\tilde{t}=0 to t~m\tilde{t}_{m} though.

Figure 3 provides but one graphic example of u~\tilde{u} versus z~\tilde{z} for Re=70Re=70, 270270, and 10001000 (for Rem=900Re_{m}=900), corresponding to d∼0.51d\sim0.51, 1.021.02, and 2.092.09 mm of spherical particles of density ρp∼1000 kg m−3\rho_p \sim1000\ \mathrm{kg}\ \mathrm{m}^{-3} falling in air with terminal velocity u(∞)∼2.06u(\infty) \sim2.06, 3.96, and 7.18 m/s, agreeing well with experimental data for water droplets falling in air [10, 3], especially for d<1d < 1 mm when droplet deformations are negligible.

Graphic display of droplet velocity versus flight distance

Figure 3. Graphic display of u~(t~)\tilde{u}(\tilde{t}) versus z~(t~)\tilde{z}(\tilde{t}) generated by numerically integrating (14), for Re=70Re=70 and 270270 (when Re<RemRe<Re_{m}), and (15), for Re=1000>Rem=900Re=1000>Re_{m}=900, and then integrating u~(t~)\tilde{u}(\tilde{t}) for z~(t~)\tilde{z}(\tilde{t})

In a strictly theoretical sense, it takes infinite time and length for a particle to reach its terminal velocity. But evaluating a finite falling length for u~\tilde{u} to reach 0.99 or 0.95, sometimes denoted as z~99\tilde{z}_{99} or z~95\tilde{z}_{95}, could be more useful for practical applications [24]. Thus, in the present context (for Re=70Re = 70, 270, and 1000), we have z~99∼2139.58\tilde{z}_{99} \sim2139.58, 3765.22, and 5092.97, corresponding to z99∼1.09z_{99} \sim1.09, 3.85, and 10.64 m (for spherical particles of d∼0.51d \sim0.51, 1.02, and 2.09 mm of density ρp∼1000 kg m−3\rho_p \sim1000\ \mathrm{kg}\ \mathrm{m}^{-3}). As expected, the computed values of z99z_{99} here for d<1d < 1 mm are in good agreement with those presented by Wang and Pruppacher [24] shown with a curve plotted in their Fig. 3 for d∼0.5d \sim0.5 and 1 mm, z99∼1z_{99} \sim1 and 4 m. If u~(z~95)=0.95\tilde{u}(\tilde{z}_{95}) = 0.95 is practically acceptable, we could have considerably reduced fall distances for measurement setups, as z95∼0.635z_{95} \sim0.635, 2.24, and 6.46 m (corresponding to z~95∼1243.43\tilde{z}_{95} \sim1243.43, 2195.26, and 3092.04). Apparently, we have z95/z99<0.61z_{95}/z_{99} < 0.61 in this case.

Summary

Out of scientific curiosity about the less-known 1934 Klyachko formula for the sphere drag coefficient, which exhibits an exceptionally clean mathematical expression, an overdue relative-error evaluation is performed by comparing it with more recent correlation formulas, for quantitatively understanding how good (or bad) this early curve-fitting formula, as well as its extension to the Reynolds number Re<2×105Re < 2 \times10^{5} as a piecewise continuous function, can be for practical applications. The results indicate quantitatively rather insignificant differences between the extended Klyachko formula Cd(Re)C_d(Re) and other apparently more sophisticated correlation formulas, for Re<1000Re < 1000 (or, better yet, for Re<900Re < 900).

Once the confidence is established for its reasonable accuracy, the extended Klyachko formula Cd(Re)C_d(Re) is briefly shown, similar to Serafini [23] and Putnam [19], to yield closed-form analytical solutions (covering the applicable range of 0<Re<2×1050 < Re < 2 \times10^{5}) for the study of rectilinear acceleration of aerosol spheres, when the effect of gravity can be ignored. Then, a demonstration of using the extended Klyachko formula Cd(Re)C_d(Re) is presented for conveniently calculating the spherical particle diameter and terminal settling velocity as functions of ReRe, yielding results in good agreement with recent correlation formulas of Brown and Lawler [4] and Cheng [5]. As expected, the dynamical process of an aerosol sphere accelerating toward its terminal velocity can also be described by solutions of ordinary differential equations based on the extended Klyachko formula, with comparable results to such relevant data of Wang and Pruppacher [24] for water droplet falling in air when droplet deformations are negligible. These demonstrative examples are expected to enhance confidence in the extended Klyachko formula for conveniently studying the classical problem of aerosol (sphere) dynamics, among other applications. The derived closed-form analytical formulas can also be applied in modern CFD code development and validation.

Acknowledgments

The author would like to thank Prof. Rodney Kubesh (of St. Cloud State University) for helpful discussions and Prof. Carlos Ramírez (of University of Puerto Rico) for providing a copy of the original 1934 Russian article by Klyachko.

Funding

None

Ethics declarations

Conflict of interests

The author has no conflicts to declare

ORCID iD

James Q Feng 0000-0003-4041-3179

References

  1. [1]Amsden, A A, P J O’Rourke, and T D Butler (1989). KIVA-II: A computer program for chemically reactive flos with sprays. Tech. rep. LA-11560-MS; ON: DE89012805. Los Alamos National Lab.DOI
  2. [2]Barati, R, S A A S Neyshabouri, and G Ahmadi (2014). “Development of empirical models with high accuracy for estimation of drag coefficient of flow around a smooth sphere: An evolutionary approach”. Powder Technol. 257, pp. 11–19. DOI: 10.1016/j.powtec.2014.02.045.DOI
  3. [3]Beard, K V (1976). “Terminal velocity and shape of cloud and precipitation drops aloft”. J. Atmos. Sci. 33.5, pp. 851–864. DOI: 10.1175/1520-0469(1976)033<0851:TVASOC>2.0.CO.doi.org/10.1175/1520-0469(1976)033<0851:tvasoc>2.0.co;2
  4. [4]Brown, P P and D F Lawler (2003). “Sphere drag and setting velocity revisited”. J. Environ. Eng. 129.3, pp. 222–231. DOI: 10.1061/(ASCE)0733-9372(2003)129:3(222).DOI
  5. [5]Cheng, N-S (2009). “Comparison of formulas for drag coefficient and settling velocity of spherical particles”. Powder Technol. 189, pp. 395–398. DOI: 10.1016/j.powtec.2008.07.006.DOI
  6. [6]Clift, R, J R Grace, and M E Weber (1978). Bubbles, Drops, and Particles. New York: Dover.
  7. [7]Davies, C D (1945). “Definitive equations for fluid resistance of spheres”. Proc. Phys. Soc. London A57.4, pp. 259–270. DOI: 10.1088/0959-5309/57/4/301.DOI
  8. [8]Fuchs, N A (1964). The Mechanics of Aerosols. New York: Dover.
  9. [9]Goossens, W R A (2019). “Review of the empirical correlations for the drag coefficient of rigid spheres”. Powder Technology 352, pp. 350–359. DOI: 10.1016/j.powtec.2019.04.075.DOI
  10. [10]Gunn, R and G D Kinzer (1949). “The terminal velocity of fall for water droplets in stagnant air”. Journal of Meteorology 6, pp. 243–248. DOI: 10.1175/1520-0469(1949)006<0243:TVOFF>2.0.CO.doi.org/10.1175/1520-0469(1949)006<0243:ttvoff>2.0.co;2
  11. [11]Hasadi, Y M F El and J T Padding (2023). “Do logarithmic terms exist in the drag coefficient of a single sphere at high Reynolds numbers?” Chemical Engineering Science 265, p. 118195. DOI: 10.1016/j.ces.2022.118195.
  12. [12]Klyachko, L S (1934). “Equations of motion for dust particles in dust-collecting devices”. USSR Journal Otoplenie i Ventilyatsiya (Heating and Ventilation) No. 4, pp. 27–29. DOI: 10.48550/arXiv.2610.07315.
  13. [13]Kramer, O J I, P J de Moel, S K R Raaghav, E T Baars, W H van Vugt, W-P Breugem, J T Padding, and J P van der Hoek (2021). “Can terminal settling velocity and drag of natural particles in water ever be predicted accurately?” Drinking Water Engineering and Science 14.1, 53=71. DOI: 10.5194/dwes-14-53-2021.DOI
  14. [14]Lamb, H (1932). Hydrodynamics (6th Edition). Cambridge: Cambridge University Press.
  15. [15]Landau, L D and E M Lifshitz (1959). Fluid Mechanics. London: Pergamon.
  16. [16]Lapple, C E and C B Shepherd (1940). “Calculation of Particle Trajectories”. Ind. Eng. Chem. 32.5, pp. 605–617.DOI
  17. [17]Michaelides, E E (2026). “The equation of motion of particles in fluids—An historical perspective”. Powders 5.1, p. 5010005. DOI: 10.3390/powders5010005.DOI
  18. [18]Pruppacher, H R and J D Klett (1997). Microphysics of Clouds and Precipitation (revised and enlarged edition). Dordrecht: Kluwer Academic Publishers.
  19. [19]Putnam, A (1961). “Integratable form of droplet drag coefficient”. J. Am. Rochet Soc. 31.10, pp. 1467–1468.
  20. [20]Ramirez, C A (2017). “Summary of frictional drag coefficient relationships for spheres: Evolving solution strategies applied to an old problem”. Chem. Eng. Sci. 168, pp. 339–343. DOI: 10.1016/j.ces.2017.04.037.DOI
  21. [21]Schiller, L and A Naumann (1933). “Über die grundlegenden Berechnungen bei der Schw-erkraftaufbereitung”. Zeitschrift des Vereins Deutscher Ingenieure 77.12, pp. 318–320.
  22. [22]Schlichting, H (1955). Boundary-Layer Theory. New York: McGraw-Hill Book Company.
  23. [23]Serafini, J S (1954). Impingement of water droplets on wedges and double-wedge airfoils at supersonic speeds. Tech. rep. NASA-TP-1159. NASA Lewis Flight Propulsion Laboratory.
  24. [24]Wang, P K and H R Pruppacher (1977). “Acceleration to terminal velocity of cloud and raindrops”. J. Appl. Meteor. 16, pp. 275–280. DOI: 10.1175/1520-0450(1977)016<0275:ATTVOC>2.0.CO.doi.org/10.1175/1520-0450(1977)016<0275:attvoc>2.0.co;2
  25. [25]Zigrang, D J and N D Sylvester (1981). “An explicit equation for particle settling velocities in solid-liquid systems”. AIChE J. 27.6, pp. 1043–1044. DOI: 10 . 1002 / aic . 690270629.DOI

Paper details

vlnc:W_ccv7khq3
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