Yamagishi Complexity Torus LabYamagishi Lab

[Paper 1127] In Slow Two-Dimensional Flow the Drag on a Cylinder Is Not Fixed Because the Same Force Passes Every Shell and the Velocity Difference Piles Up as the Logarithm of the Radius

Abstract

In slow flow governed by viscosity alone, a sphere in three dimensions feels the drag 6πμUa, but for a cylinder in two dimensions there is no solution that both sticks to the cylinder and becomes a uniform stream far away (the Stokes paradox, Stokes 1851). Paper 36 showed that the drag coefficient in d dimensions carries a factor d−2 and vanishes at d = 2. This paper applies the shell counting of Paper 1106 to a flow that has a direction, and computes what remains in two dimensions. It claims no new theorem or law. In the numerical computations the force and the viscosity are set to 1 for the flow of a point force, and the speed, the radius and the viscosity are set to 1 for the flow around a body. The force in two dimensions is per unit length of the cylinder.

For the flow of a point force (the Stokeslet), in two, three and four dimensions alike, the same force passes through every shell of radius 0.1 to 1000 (differences below 10⁻¹⁴), and the stress thins out as the (1−d)th power of the radius. The velocity averaged over a shell is the sum of a radial term and a directional term. The directional term is 1/d, the shell average of cos² of the angle to the flow. In three dimensions the two terms share the same 1/r and combine into (1 + 1/3)/(8πr) = 1/(6πr). The velocity of a sphere equals this shell average (the potential dipole of Paper 36 that is added to the flow of a sphere averages to 0 over a shell), which gives the 6π of the sphere. The coefficient of Paper 36 is this form after simplification, and its values for d = 3 to 7 also agree to 30 digits. In two dimensions the radial term becomes a logarithm; the average is (−ln r + 1/2)/(4π), it drops by the same ln10/(4π) = 0.183234 for every tenfold increase of the radius, and it never stops. The density of dissipation goes as the −2(d−1)th power of the radius, so on the map of Paper 1106 it sits at n = 2d − 2, and only d = 2 lies on the border n = d.

Cutting off the outside brings the drag back. For a cylinder inside a fixed coaxial cylinder of radius R, solving symbolically gives the drag 4πμU/(ln(R/a) − (R² − a²)/(R² + a²)), and moving the wall from 1000 to a million radii away changes the drag only by a factor 0.460985. In a spherical container in three dimensions the drag is 6πμUa times 1 + (9/4)(a/R) + …, and approaches 6πμUa as the wall moves away. Without a wall, inertia cuts off the outside. Near the point, the point-force solution of Oseen’s equation approaches the two-dimensional Stokeslet cut off at the length L = 4e^(−γ)ν/U = 2.245838 ν/U, and joining the two gives Lamb’s formula 4πμU/(ln(L/a) + 1/2) for the drag on the cylinder. The wall and inertia only fix one and the same length L. Even at aU/ν = 10⁻⁶ the constant beside the logarithm moves the drag by a relative amount of 0.0948.

In two dimensions, shape and orientation enter only as a constant added beside the logarithm. When the force on the surface of an elliptic cylinder is distributed in proportion to the distance from the centre to the tangent line (the same distribution as the homoeoid of Paper 1116), the velocity on the cylinder is the same everywhere and 4πμU/F = ln(2L/(α + β)) + α/(α + β), where α is the semi-axis along the motion. Written with half-angles this follows as a formula, and the values added up numerically on the surface also differ from it by less than 10⁻¹⁹. For a flat plate, the ratio of the broadside to the edgewise drag (broadside divided by edgewise) approaches 1 as 1.333808, 1.131563 and 1.068924 when the outer length L is 10, 10³ and 10⁶ times the half-length of the plate. A disc in three dimensions has the drag 16μUa broadside and (32/3)μUa edgewise, and the ratio 3/2 does not depend on its size. The separator is whether the radial term of the shell average is a power or a logarithm. For a power, the outside is a small correction and shape and orientation enter as a multiplying factor. For a logarithm, the outer length enters the answer, and shape and orientation are demoted to a constant beside the logarithm.

Keywords: Stokes paradox, Stokeslet, Oseen approximation, drag on a cylinder, shell counting, logarithm, separator.

Scope (what is not claimed)

This paper claims no new theorem or law. The Stokes paradox (Stokes 1851), Oseen’s equation (Oseen 1910) and the drag on a cylinder obtained from it (Lamb 1911), the treatment that joins the expansions near and far (Kaplun 1957; Proudman and Pearson 1957), the Stokeslet, the correction for the walls of a container, and the drag on elliptic cylinders and discs (Happel and Brenner 1983; Kim and Karrila 1991) are all known. It also appears to be known (the literature has not been checked here) that, at the first order of the approximation, the drag on a two-dimensional body sees its shape only through one length, and that in three dimensions the Stokeslet averaged over a sphere is F/(6πμa) (Faxén’s law). Lamb’s formula has not been reread in the original; the formula and the numbers were checked by the computations of this paper (the near-field form of Oseen’s point-force solution). The solutions in a container and the formula for the elliptic cylinder were solved here and checked numerically, and have not been compared with formulas in other literature. The accuracy of Oseen’s approximation, higher-order inertial terms, fast flow with shed vortices, flow that changes in time, cylinders of finite length, and the sideways force on an ellipse moving obliquely are not treated. The ratio of the drags on a flat plate belongs to the first term in joining the near and the far regions; there the far solution is fixed by a point force along the flow alone, so L is the same for edgewise and broadside motion (the joining follows Kaplun 1957 and Proudman and Pearson 1957, and Oseen’s equation has not been solved directly for a plate). What is added is: counting, in two, three and four dimensions, that even for a flow with a direction the same force passes every shell and the shell average is the sum of a radial term and a directional term, checking in three, four and five dimensions with the potential dipole of Paper 36 that the velocity of a sphere equals that shell average, and rereading the coefficient of Paper 36 in this form; placing the two-dimensional paradox on the border of the map of Paper 1106 through the exponent of the density of dissipation; lining up the two ways of cutting off the outside (a container wall and the inertial length) as ways of fixing one length L; and counting, with elliptic cylinders, flat plates and discs, that shape and orientation enter as a multiplying factor in three dimensions but as a constant beside the logarithm in two. Paper 36 computed the closed form of the drag in d dimensions, showed that the factor d−2 of the coefficient vanishes at d = 2 and that in two dimensions only the solution with the fluid at rest remains, and wrote in one sentence that the true two-dimensional Stokeslet is logarithmic. This paper computes what that sentence contains (the coefficient of the logarithm, the outer length and the constant beside it). Paper 37 showed that the coefficient carries (d−2)/(d−1) both for the sticking condition and for the slipping condition; this paper treats only the sticking condition. Paper 1106 made the map on which the exponent of the weight of one shell and the dimension decide where an infinity sits, and placed the planar vortex and the straight current on the border; this paper adds the point of a flow with a direction. Paper 1098 showed that in a lattice sum whose exponent equals the dimension, shape enters as the mean logarithm of the distance to the edge. The length (α + β)/2 of the ellipse here is a different length from the harmonic mean of the ellipse in Paper 1098; what they share is the pattern that, on the border, shape enters beside the logarithm. Paper 1116 computed the shell whose thickness is proportional to the distance to the tangent plane (the homoeoid); the force on the surface of the elliptic cylinder here is the two-dimensional version of that distribution. Paper 1084 computed the longitudinal share 1/3 on a sphere, and Paper 492 showed numerically that the drag on a cylinder in a flow without viscosity is 0. This paper sits beside them.

1. Only in two dimensions is the drag not fixed

Slow flow in which viscosity acts strongly and inertia does not (Stokes flow) obeys $\mu\nabla^2\mathbf u=\nabla p$ and $\nabla\cdot\mathbf u=0$. When a sphere of radius $a$ moves at speed $U$ in three dimensions, the drag is $6\pi\mu Ua$ (Stokes 1851). For a cylinder in two dimensions (a long cylinder perpendicular to the page, seen in cross-section) there is no solution that satisfies both the condition that the fluid sticks to the surface and the condition of a uniform stream far away. This is the Stokes paradox.

Paper 36 computed the drag in $d$ dimensions as $F=\frac{d(d-2)}{d-1}|S^{d-1}|\mu Ua^{d-2}$ ($|S^{d-1}|$ is the area of the unit sphere), and showed that the coefficient vanishes at $d=2$ and that in two dimensions only the solution with the fluid at rest remains. Paper 37 showed that the coefficient carries $(d-2)/(d-1)$ both for the sticking condition and for the slipping condition. Paper 492 showed numerically that, in a flow without viscosity, the drag on a cylinder is 0. This paper is on the side of a flow with viscosity alone.

Paper 1106 made a map: when a density thins out as the inverse $n$th power of the radius, the weight of one shell is $r^{d-1-n}$, and $n=d$ is the border. The points placed on it were quantities without a direction, or quantities whose directions had already been summed. This paper asks whether this shell counting still works for a flow that has a direction, and computes what remains in two dimensions.

In the numerical computations, the flow of a point force (Sections 2 to 4) uses the force $F=1$ and the viscosity $\mu=1$, the body in a container (Section 5) uses the speed $U=1$, the radius $a=1$ and $\mu=1$, and Oseen’s equation (Section 6) uses $U=1$, $\nu=1/2$ ($k=1$) and density 1. The force in two dimensions is per unit length of the cylinder. Symbolic computations used sympy, and numerical ones used Gaussian and trapezoidal quadrature in numpy and 20-digit and 30-digit arithmetic in mpmath.

2. The same force passes every shell

The flow of a point force $\mathbf F$ (the Stokeslet) is, in two dimensions,

$$u_i=\frac{F_j}{4\pi\mu}\Bigl(-\delta_{ij}\ln r+\frac{x_ix_j}{r^2}\Bigr),\qquad p=\frac{F_jx_j}{2\pi r^2},$$

and in three or more dimensions

$$u_i=\frac{F_j}{2(d-2)|S^{d-1}|\mu}\Bigl(\delta_{ij}r^{2-d}+(d-2)\frac{x_ix_j}{r^d}\Bigr),\qquad p=\frac{F_jx_j}{|S^{d-1}|\,r^d}$$

(the form with $1/(8\pi\mu)$ in three dimensions). In two, three and four dimensions it was checked symbolically that these satisfy the Stokes equations everywhere except at the point and have no sources. The velocity has a direction, and the stress carries directional weights such as $x_ix_j$. Even so, adding up the force through a shell of radius $r$ (the force that the outer fluid exerts on the inner fluid) gives the same value at every radius.

dimension $d$force through the shells (radius 0.1, 1, 10, 1000)largest difference from $(-1,0,\dots)$largest stress on the shell $\times r^{d-1}$ ($=d/|S^{d-1}|$)
2$-1$ (all four)$2.22\times10^{-16}$0.3183098862 ($1/\pi$)
3$-1$ (all four)$8.44\times10^{-15}$0.2387324146 ($3/(4\pi)$)
4$-1$ (all four)$8.44\times10^{-15}$0.2026423673 ($2/\pi^2$)

The density of the force on the shell is $-(d/|S^{d-1}|) n_in_1 r^{1-d}$ ($\mathbf n$ being the outward unit normal of the shell), and its magnitude is largest at the pole $\mathbf n=\mathbf e_1$ along the flow (the right-hand column of the table). Adding the directional weight $n_in_1$ over the shell gives $\delta_{i1}|S^{d-1}|r^{d-1}/d$, so the force through the shell is exactly $-1$ at every radius. The directions cancel inside the coefficient, and only the exponent $1-d$ of the radius remains. The force does not decrease as it passes each shell, and the area of the shell grows as $r^{d-1}$, so the stress thins out as $r^{1-d}$. This is the same counting as Gauss’s law, by which the electric flux is kept from shell to shell. Even for a flow with a direction, the total force through a shell and the exponent of the radius can be read by shell counting.

3. The shell average is a radial term plus a directional term

The velocity of a sphere (a circle in two dimensions) equals the velocity of the flow of a point force placed at its centre, averaged over its surface (the reason in three or more dimensions is given later in this section, and the circle in two dimensions is computed in Section 7). Averaging the component of the Stokeslet along the flow over a shell of radius $r$ gives

$$\langle u_1\rangle=\frac{F\,r^{2-d}}{2|S^{d-1}|\mu}\Bigl(\frac{1}{d-2}+\frac1d\Bigr).$$

Here $1/(d-2)$ is the radial term coming from $\delta_{ij}r^{2-d}$, and $1/d$ is the directional term coming from the shell average of $x_1^2/r^2$ (the average of $\cos^2$ of the angle to the flow). In three and four dimensions this agreed within $10^{-12}$ with the value obtained by adding over the shell numerically.

In three dimensions both terms carry $1/r$, so they combine into $(1+1/3)/(8\pi r)=1/(6\pi r)$. For a sphere of radius $a$ this average is the velocity $U$ of the sphere, and the drag is $6\pi\mu Ua$.

Why does the velocity of a sphere equal the shell average? The flow outside a sphere in $d$ dimensions is the Stokeslet plus the potential dipole $\delta_{i1}r^{-d}-d x_ix_1r^{-d-2}$ (Construction 2.1 of Paper 36). Taking the strength of the dipole as $Fa^2/(2d|S^{d-1}|\mu)$ removes the $x_ix_1$ term on the sphere, and what remains is a uniform velocity equal to the shell average. Since the dipole averages to 0 over a shell, the velocity of the sphere equals the shell average of the Stokeslet alone. In three, four and five dimensions it was checked symbolically that this flow satisfies the Stokes equations and has no sources, and to 30 digits that the velocity at five points on the sphere equals the shell average (differences below $10^{-32}$).

The directional term $1/3$ is the same number as the average of $\cos^2$ on a sphere that Paper 1084 called the longitudinal share. In $d$ dimensions

$$\frac{F}{\mu Ua^{d-2}}=\frac{2|S^{d-1}|}{1/(d-2)+1/d}=\frac{d(d-2)}{d-1}\,|S^{d-1}|,$$

which is the same form as the coefficient of Paper 36 after simplification (the values for $d=3$ to 7 also agree to 30 digits).

$d$radial term $1/(d-2)$directional term $1/d$$F/(\mu Ua^{d-2})$ (value of Paper 36)
311/318.8495559215 ($6\pi$)
41/21/452.6378901391 ($16\pi^2/3$)
51/31/598.6960440109 ($10\pi^2$)
61/41/6148.830128065 ($24\pi^3/5$)
71/51/7192.927943789 ($56\pi^3/9$)

The factor $d-2$ that Paper 36 showed to vanish at $d=2$ is the denominator of the radial term. At $d=2$ the radial term takes the form $r^0/0$ and is replaced by $-\ln r$. The shell average in two dimensions is

$$\langle u_1\rangle=\frac{F}{4\pi\mu}\Bigl(-\ln r+\frac12\Bigr),$$

so the directional term $1/2$ remains, but the radial term becomes a logarithm. At radii 1, 10, 100 and 1000 the average was 0.0397887358, $-0.1434451639$, $-0.3266790637$ and $-0.5099129634$, dropping by the same $\ln10/(4\pi)=\mathbf{0.183234}$ for every tenfold increase of the radius. In three dimensions it was 0.0530516477, 0.0053051648 and 0.0005305165 at radii 1, 10 and 100, falling to a tenth at each tenfold step and coming to rest. In two dimensions it keeps dropping by the same amount however far one goes, so the difference between the velocity of the body and that of the distant flow is decided by where the outside ends. A question that does not state where the outside ends has no answer. This is the Stokes paradox seen through shell counting.

4. The position on the map

The dissipation of the Stokeslet (the density $2\mu e_{ij}e_{ij}$ of work turned into heat by viscosity, $e_{ij}$ being the symmetric part of the velocity gradient) became $10^{-2}$, $10^{-4}$ and $10^{-6}$ times smaller in two, three and four dimensions when the point was moved ten times farther away. The exponent on the map of Paper 1106 is $n=2d-2$, and only $d=2$ lies on the border $n=d$.

In two dimensions the dissipation in one shell is $1/(4\pi r)$ (multiplied by $r$ it was 0.0795774715 at radii 1, 10 and 100 alike), and every bundle of shells that increases the radius tenfold carries the same dissipation $\ln10/(4\pi)=0.183234$. In three dimensions the dissipation in one shell is $1/(4\pi r^2)$; the bundle from radius 1 to 10 carries 0.071620 and the bundle from 10 to 100 carries 0.007162, a tenth for each bundle.

The point $(d,n)=(2,2)$ of the two-dimensional cylinder is the same point at which Paper 1106 placed the planar vortex and the straight current, where the sum grows logarithmically at both ends. The inner end is cut off by the radius $a$ of the cylinder, so what remains is the outer end. The point $(3,4)$ in three dimensions is the same as that of the field energy of a point charge, which overflows at the centre and stops outside. The inner end is cut off by the radius $a$ of the sphere, so the sum of the dissipation stops in proportion to $a^{d-n}=a^{2-d}$. The work $FU$ done in moving the body equals the sum of the dissipation (checked numerically for the two-dimensional container in Section 5), and the sum of the dissipation is proportional to $F^2$, so $F$ is proportional to $a^{d-2}$. Only the exponent is read here. The coefficient also contains the part of the dipole in the flow of a sphere: in three dimensions, the dissipation outside the sphere is $9\pi\mu U^2a$ for the Stokeslet alone and $FU=6\pi\mu U^2a$ for the true flow of the sphere (ratio 3/2). The exponent of $a^{d-2}$ in Paper 36 is the map’s $d-n=2-d$ with its sign reversed.

5. Cutting off the outside with a container wall

Let a cylinder of radius $a$ move at speed $U$ inside a fixed coaxial cylinder of radius $R$. Writing the stream function as $\psi=(Ar^3+Br\ln r+Cr+D/r)\sin\theta$, the four coefficients were fixed by the sticking conditions on the two surfaces, and it was checked symbolically that, with the pressure $p=\mu(8Ar-2B/r)\cos\theta$, the Stokes equations are satisfied. Adding up the stress on the surface of the cylinder gives

$$F=\frac{4\pi\mu U}{\ln(R/a)-(R^2-a^2)/(R^2+a^2)},$$

which equals $-4\pi\mu B$ from the coefficient of $r\ln r$ in the stream function (the strength of the Stokeslet). At $R/a=10$ the total dissipation in the fluid was 9.5027929939, agreeing with the product of the drag and the speed to a relative difference of order $10^{-14}$.

In three dimensions, for a sphere of radius $a$ moving inside a spherical container of radius $R$, the same steps with the stream function $\psi=(Ar^4+Br^2+Cr+D/r)\sin^2\theta$ gave

$$\frac{F}{6\pi\mu Ua}=\frac{4(1+\lambda+\lambda^2+\lambda^3+\lambda^4)}{(1-\lambda)^3(4+7\lambda+4\lambda^2)},\qquad \lambda=\frac aR$$

(agreeing at $R/a=10$ with the force 24.2442298095 obtained by adding up the stress directly). For small $\lambda$ this is $1+\frac94\lambda+\cdots$.

$R/a$2D $F/(4\pi\mu U)$2D $F (\ln(R/a)-1)/(4\pi\mu U)$3D $F/(6\pi\mu Ua)$
$10$0.7562080.9850261.286196
$10^2$0.2773640.9999451.023015
$10^3$0.1692691.0000001.002255
$10^6$0.0780301.0000001.000002

In three dimensions the drag approaches $6\pi\mu Ua$ as the wall moves away, and the effect of the wall dies out as $a/R$. The drag stays within 1 percent above its value without a wall when $R/a$ exceeds 227.2451. In two dimensions the drag goes to 0 along $4\pi\mu U/(\ln(R/a)-1)$, and moving the wall from 1000 to a million radii away changes the drag only by a factor of 0.460985. In three dimensions the wall is a small correction; in two dimensions the wall is part of the answer.

Writing the formula for a distant wall in the form $\ln(L/a)+1/2$ of the shell average of Section 3 gives $L=e^{-3/2}R=0.223130 R$. The wall fixes one length $L$ at which the outside is cut off.

6. Cutting off the outside with the inertial length

Even without a wall, inertia cuts off the outside. Oseen wrote the inertial term as $U\partial_x\mathbf u$, since far away the flow is nearly uniform (Oseen 1910). With $\nu=\mu/\rho$ and $k=U/(2\nu)$, the solution of Oseen’s equation for a point force $F$ can be written as

$$\mathbf u=\nabla(\phi+\chi)-2k\chi\,\mathbf e_x,\qquad \phi=-\frac{F}{2\pi\rho U}\ln r,\qquad \chi=-\frac{F}{2\pi\rho U}e^{kx}K_0(kr)$$

($K_0$ is a modified Bessel function). It was checked to 30 digits that $\chi$ satisfies $(\nabla^2-2k\partial_x)\chi=0$, and that $\mathbf u$, with the pressure $p=Fx/(2\pi r^2)$, satisfies $U\partial_x\mathbf u-\nu\nabla^2\mathbf u+\nabla p/\rho=0$ and $\nabla\cdot\mathbf u=0$ (residuals at most of order $10^{-30}$).

Comparing this solution near the point (where $kr$ is small) with the two-dimensional Stokeslet cut off at the length $L$, $(-\ln(r/L) \delta_{ij}+x_ix_j/r^2)F/(4\pi\mu)$, the largest difference of the velocities multiplied by $4\pi\mu/F$ (over 720 angles) shrank to 0.0262, 0.000491 and $7.22\times10^{-6}$ at $kr=10^{-2}$, $10^{-4}$ and $10^{-6}$, of the order of $kr |\ln(kr)|$ (the largest is at the angle 0 along the flow), when

$$L=\frac2k\,e^{-\gamma}=4e^{-\gamma}\frac{\nu}{U}=\mathbf{2.245838}\,\frac{\nu}{U}$$

($\gamma$ is Euler’s constant). Inertia cuts off the outside at a length of the order of $\nu/U$. Joining this length to the form of the shell average of Section 3 (the first term in joining the near and the far regions) gives

$$F=\frac{4\pi\mu U}{\ln(L/a)+1/2}=\frac{4\pi\mu U}{\ln\bigl(4\nu/(aU)\bigr)+1/2-\gamma}.$$

This is Lamb’s formula (Lamb 1911). What this paper computed are the two sides being joined (the far form $\ln(L/a)+1/2$ of the flow near the cylinder, and the near form of Oseen’s point-force solution). That setting the two equal where they overlap is correct at the first term rests on the treatment that joins the near and far expansions in Kaplun 1957 and in Proudman and Pearson 1957.

$aU/\nu$$F/(4\pi\mu U)$ (Lamb’s formula)logarithm only $1/\ln(\nu/(aU))$relative difference
$10^{-1}$0.2768810.4342940.5685
$10^{-2}$0.1690830.2171470.2843
$10^{-3}$0.1217010.1447650.1895
$10^{-6}$0.0661170.0723820.0948

The constant added beside the logarithm is $\ln4+1/2-\gamma=1.309079$. Even at $aU/\nu=10^{-6}$ this constant moves the drag by a relative amount of 0.0948. Because the logarithm grows slowly, the effect of the constant beside it is slow to die out.

The wall’s $L=e^{-3/2}R$ and inertia’s $L=4e^{-\gamma}\nu/U$ are equal when $R=4e^{3/2-\gamma}\nu/U=10.0651 \nu/U$. The wall and inertia are different things physically, but in the formula for the drag they occupy the same seat (the outer length $L$). In three dimensions inertia only adds a small correction of the order of $aU/\nu$ to $6\pi\mu Ua$ (Oseen 1910; not computed here), and the outside never enters the core of the answer.

7. Shape and orientation enter only as a constant beside the logarithm

Consider an elliptic cylinder whose semi-axis along the motion is $\alpha$ and whose sideways semi-axis is $\beta$. Distribute the force on its surface in proportion to the distance $p$ from the centre to the tangent line. This is the two-dimensional version of the distribution that Paper 1116 called the homoeoid; writing the points of the surface as $(\alpha\cos t, \beta\sin t)$, it is uniform in the parameter $t$. Adding up numerically the velocity on the surface of the cylinder in the flow produced by this force (the Stokeslet cut off at the length $L$, added over the surface), at four points on the surface of the ellipses $(2,1)$ and $(1,0.2)$ and the circle $(1,1)$, gave everywhere

$$\frac{4\pi\mu U}{F}=\ln\frac{2L}{\alpha+\beta}+\frac{\alpha}{\alpha+\beta},$$

with a largest difference from the formula of $6.44\times10^{-20}$ (the sideways velocity is 0). This also follows as a formula. Measuring from the surface point $t_0$ and writing with the half-angles $\sigma=(t+t_0)/2$ and $s=(t-t_0)/2$, the difference of the two points is $(2\alpha\sin\sigma\sin s, -2\beta\cos\sigma\sin s)$; the average of the logarithm is $\langle\ln|2\sin s|\rangle+\langle\frac12\ln(\alpha^2\sin^2\sigma+\beta^2\cos^2\sigma)\rangle=0+\ln((\alpha+\beta)/2)$, the average of the directional weight is $\langle\alpha^2\sin^2\sigma/(\alpha^2\sin^2\sigma+\beta^2\cos^2\sigma)\rangle=\alpha/(\alpha+\beta)$, and the average of the sideways weight is 0, none of them depending on $t_0$ (for the three shapes, within $10^{-15}$ of the values added up numerically over $\sigma$). Since the velocity on the cylinder is uniform, this is the flow of a cylinder moving without changing its shape, and the force on the surface is the force on the cylinder. For the circle $\alpha=\beta=a$ it is $\ln(L/a)+1/2$, returning to the shell average of Section 3.

A flat plate (half-length $b$) is the case $\beta\to0$; the density of the force becomes $1/(\pi\sqrt{b^2-s^2})$, and the average of the logarithm on the plate is $\ln(b/2)$ at every point of the plate ($-0.693147180559945$ at four points with $b=1$). Moving along the plate (edgewise) gives $4\pi\mu U/F=\ln(2L/b)+1$, and pushing the plate (broadside) gives $\ln(2L/b)$. They are listed with the length $r_e$ that puts each into the form $\ln(L/r_e)+1/2$ (for a circular cylinder, the radius itself).

body and direction of motion$4\pi\mu U/F$length $r_e$
circular cylinder (radius $a$)$\ln(L/a)+1/2$$a$
ellipse $(2,1)$, along the long axis$\ln(2L/3)+2/3$1.269723
ellipse $(2,1)$, along the short axis$\ln(2L/3)+1/3$1.772041
flat plate (half-length $b$), edgewise$\ln(2L/b)+1$$0.303265 b$
flat plate (half-length $b$), broadside$\ln(2L/b)$$0.824361 b$

Both shape (the length $(\alpha+\beta)/2$ in the logarithmic term of the formula) and orientation ($\alpha/(\alpha+\beta)$) enter only as a constant added beside the logarithm. Since at the first term the outer length $L$ does not depend on the orientation (see the scope note), the ratio $(\ln(2L/b)+1)/\ln(2L/b)$ of the broadside to the edgewise drag of a flat plate (broadside divided by edgewise) approaches 1 with the outer length, being 1.333808, 1.131563 and 1.068924 at $L/b=10$, $10^3$ and $10^6$. The effect of orientation thins out as the logarithm grows.

Three dimensions are different. Distributing the force density $1/\sqrt{a^2-\rho^2}$ on a disc of radius $a$ gives the same velocity at every point of the disc. From a point at distance $\rho_0$ from the centre, the inner integral along the direction $\varphi$ out to the rim closes as $\pi/2-\arcsin\bigl(\rho_0\cos\varphi/\sqrt{a^2-\rho_0^2\sin^2\varphi}\bigr)$, and the second term changes sign under $\varphi\to\pi-\varphi$, so it cancels when added over the directions. The drag is $F=16\mu Ua$ broadside and $F=(32/3)\mu Ua$ edgewise (checked with the closed form at the four points $\rho_0=0$, 0.3, 0.7 and 0.95, and with numerical integration at 0, 0.3 and 0.7). The ratio 3/2 depends neither on the size nor on the outer length. In three dimensions shape and orientation enter as a multiplying factor. A sphere has $6\pi\mu Ua=18.849556 \mu Ua$, 1.178097 times the broadside drag of the disc.

Paper 1098 showed that in a lattice sum whose exponent equals the dimension, shape enters as the mean logarithm of the distance to the edge. The length $(\alpha+\beta)/2$ of the ellipse here is a different length from the harmonic mean of the ellipse in Paper 1098. What they share is the pattern that, on the border, shape enters as a constant beside the logarithm.

8. Shared root or distinct roots

The quantities of this paper are sorted with the test of Paper 300.

(1) The logarithm of the two-dimensional cylinder and the logarithm of the planar vortex and the straight current on the map of Paper 1106: both lie on the border $(d,n)=(2,2)$ and come from the weight of one shell being $1/r$. When the dimension is moved, the dissipation of the Stokeslet and the field energy of a point source move along the same line $n=2d-2$ (checked for $d=2$ to 5), and they coincide with the point $(2,2)$ of the planar vortex and the straight current in two dimensions and with the point $(3,4)$ of the point charge in three (the test of Paper 300 of whether the agreement persists when the variable is moved). Both become logarithmic in two dimensions because the quantity kept from shell to shell (the circulation of the vortex, the current, the force on the cylinder) spreads over a circle of length $2\pi r$, so the field goes as $1/r$ (in $d$ dimensions it spreads over the area $r^{d-1}$ of the shell, the field goes as $r^{1-d}$, and the exponent of the density of its square is $2d-2$). They share a root. The logarithm of the Stokeslet is the two-dimensional form of what Observation 2.3 of Paper 36 wrote about $r^{2-d}$ (the Stokeslet inherits the fundamental solution of the Laplace equation).

(2) The factor $d-2$ of the coefficient in Paper 36 and the two-dimensional logarithm here: both are the radial term $r^{2-d}/(d-2)$ as $d\to2$, so they share a root.

(3) The container wall and the inertial length: the wall is a boundary condition and inertia is a term of the equation, so physically their roots are distinct. In the formula for the drag both only fix one outer length $L$, and they occupy the same seat.

(4) The directional term $1/3$ and the longitudinal share $1/3$ of Paper 1084: both are the average of $\cos^2$ on a sphere, and up to that average they share a root. Paper 1084 counted how the $4\pi$ of a field is divided, while this paper averages a velocity, so they are used for different things.

(5) The homoeoid distribution of Paper 1116 and the force on the surface of the elliptic cylinder here: they are the same distribution. That the logarithmic term is constant on the surface is the same fact as the sum of logarithms being constant inside a two-dimensional homoeoid (zero force inside): for the ellipse $(2,1)$ it was $\ln(3/2)$ at four inside points, and larger at two outside points. That the directional term $\alpha/(\alpha+\beta)$ is constant comes from a different integral (the half-angle average of Section 7). Whether this comes from the same reason as the vanishing force inside the three-dimensional ellipsoid of Paper 1116 is not shown here.

9. The separator

The separator is whether the radial term of the shell average is a power or a logarithm. If the exponent $n=2d-2$ of the density of dissipation is larger than the dimension $d$ ($d\ge3$), the radial term is the power $r^{2-d}/(d-2)$, the outside becomes a small correction (in a three-dimensional container it dies out as $a/R$), and the directional term $1/d$ carries the same power as the radial term, so the two combine into a multiplying factor. Shape and orientation also enter as a multiplying factor (the 3/2 of the disc). If $n=d$ ($d=2$), the radial term is a logarithm, the length $L$ at which the outside is cut off ($e^{-3/2}R$ for a wall, $4e^{-\gamma}\nu/U$ for inertia) enters the answer, and both the directional term $1/2$ and the shape are demoted to constants added beside the logarithm. A two-dimensional question that does not state where the outside ends has no answer; once it is stated, the answer comes back as one over a logarithm.

Closing

In slow two-dimensional flow the drag on a cylinder is not fixed because the force that the cylinder exerts on the fluid passes every shell undiminished, and the shell-averaged velocity piles up by the same ln10/(4π) for every tenfold increase of the radius and never stops. Shell counting still works for a flow with a direction. The direction enters the shell average as a term 1/d; in three dimensions it combines into the factor that makes the 6π of the sphere, and in two dimensions it remains as 1/2 beside the logarithm. Cutting off the outside at a container wall or at the inertial length brings the drag back as 4πμU divided by a logarithm, and the wall and inertia only fix one and the same length L. Shape and orientation enter as a multiplying factor in three dimensions, and as a constant beside the logarithm in two.

References

  1. Happel, J., and Brenner, H. (1983). Low Reynolds Number Hydrodynamics. Martinus Nijhoff.
  2. Kaplun, S. (1957). Low Reynolds number flow past a circular cylinder. Journal of Mathematics and Mechanics.
  3. Kim, S., and Karrila, S. J. (1991). Microhydrodynamics: Principles and Selected Applications. Butterworth-Heinemann.
  4. Lamb, H. (1911). On the uniform motion of a sphere through a viscous fluid. Philosophical Magazine.
  5. Oseen, C. W. (1910). Über die Stokes’sche Formel und über eine verwandte Aufgabe in der Hydrodynamik. Arkiv för Matematik, Astronomi och Fysik.
  6. Proudman, I., and Pearson, J. R. A. (1957). Expansions at small Reynolds numbers for the flow past a sphere and a circular cylinder. Journal of Fluid Mechanics.
  7. Stokes, G. G. (1851). On the effect of the internal friction of fluids on the motion of pendulums. Transactions of the Cambridge Philosophical Society.
  8. Yamagishi, Y. (2026). The $\pi$ in Stokes’ $6\pi$ Is a Solid Angle [Paper 36]. YOI: https://www.yamagishiyuuki.com/yoi/36; DOI: 10.5281/zenodo.22091319
  9. Yamagishi, Y. (2026). Is $6\pi$ Really $4\pi+2\pi$? [Paper 37]. YOI: https://www.yamagishiyuuki.com/yoi/37; DOI: 10.5281/zenodo.22091500
  10. Yamagishi, Y. (2026). The Three Hundredth Census: Whether Two Things Share a Root Is Decidable [Paper 300]. YOI: https://www.yamagishiyuuki.com/yoi/300; DOI: 10.5281/zenodo.22171914
  11. Yamagishi, Y. (2026). One and the Same Flow Makes the Drag $0$ and the Lift $3$ [Paper 492]. YOI: https://www.yamagishiyuuki.com/yoi/492; DOI: 10.5281/zenodo.22760083
  12. Yamagishi, Y. (2026). 4π Splits into a Longitudinal 4π/3 and a Transverse 8π/3, and the 21 cm Line of Hydrogen Measures the Transverse Part [Paper 1084]. YOI: https://www.yamagishiyuuki.com/yoi/1084
  13. Yamagishi, Y. (2026). In a Lattice Sum Whose Exponent Equals the Dimension, Shape Enters as the Mean Logarithm of the Distance to the Edge [Paper 1098]. YOI: https://www.yamagishiyuuki.com/yoi/1098
  14. Yamagishi, Y. (2026). A Physical Infinity Is Set by How Fast Things Thin Out against the Dimension, Not by the Mass [Paper 1106]. YOI: https://www.yamagishiyuuki.com/yoi/1106
  15. Yamagishi, Y. (2026). A Uniform Ellipsoid Pulls from Outside Like a Flat Focal Disc, and a Sphere Pulling Like a Point Is the Case in Which That Disc Shrinks to One Point [Paper 1116]. YOI: https://www.yamagishiyuuki.com/yoi/1116

Reference

Yamagishi, Y. (2026). In Slow Two-Dimensional Flow the Drag on a Cylinder Is Not Fixed Because the Same Force Passes Every Shell and the Velocity Difference Piles Up as the Logarithm of the Radius [Paper 1127]. YOI: https://www.yamagishiyuuki.com/yoi/1127
ORCID: 0009-0006-8587-3450

PDF (English) · PDF (Japanese) · Date proof of the English PDF (OpenTimestamps .ots) · SHA256SUMS

The record page for this paper (in Japanese, with the PDFs in each language and the date proof): YOI 1127. This paper is a preprint and has not been peer reviewed.