Dimensional Analysis

Central idea. A physical law cannot depend on the units chosen to measure its quantities. If an unknown $a_0$ depends on $n$ governing parameters, of which $k$ have independent dimensions, the $\Pi$ theorem reduces the law to a relation between $n-k+1$ dimensionless groups, $\Pi_0 = F(\Pi_1,\dots,\Pi_{n-k})$. In fluid mechanics the list of parameters is read from the conservation equations and their boundary conditions; the limits $\mathrm{Re}\gg 1$ and $\mathrm{Re}\ll 1$ remove one more parameter; and equal dimensionless groups in a model and in the real system (physical similarity) let a laboratory test predict the full-scale result.

1. Dimensions of a Physical Magnitude

The falling body

The chapter opens with a mass $m$ released from rest at a height $h$ under gravity $g$ (Notes.pdf, p. 57, Fig. 6.1). The unknowns are the collision time $t_c$ and the collision velocity $v_c$.

m h g vc at t = tc unknowns tc and vc, functions of (m, h, g): Eq. (6.1)
Figure 1. The fall of a point mass, Fig. 6.1 of the notes: a mass m released from rest at a height h falls under gravity g; the unknowns are the collision time t_c and the collision velocity v_c.

The exact solution is quick, and it will be used to check the dimensional result. Newton's second law $m\,d^2z/dt^2 = -mg$ with $z = h$ and $dz/dt = 0$ at $t = 0$ is integrated once, from $0$ to $t$, using Barrow's rule:

$$ \frac{dz}{dt}(t) - \frac{dz}{dt}(0) = \int_0^t (-g)\,dt' = \big[-g t'\big]_0^t = -g t \quad\Longrightarrow\quad \frac{dz}{dt} = -g t $$

and once more:

$$ z(t) - z(0) = \int_0^t (-g t')\,dt' = \left[-\frac{g t'^2}{2}\right]_0^t = -\frac{g t^2}{2} \quad\Longrightarrow\quad z = h - \frac{g t^2}{2} $$

Setting $z(t_c) = 0$ gives $t_c = \sqrt{2h/g}$, and then $v_c = g t_c = \sqrt{2gh}$ (Notes.pdf, p. 57).

What we would guess before solving

Before any integration, one guesses which quantities can influence the answer: the mass, the height and gravity (Notes.pdf, Eq. 6.1):

$$ t_c = t_c(m, h, g), \qquad v_c = v_c(m, h, g) \tag{6.1} $$

Units and dimensions

The numbers $t_c$, $v_c$, $h$, $m$, $g$ are the result of comparing each quantity with a unit. If the unit of length is made $L$ times smaller, the unit of time $T$ times smaller and the unit of mass $M$ times smaller, the numbers $h$, $t_c$ and $m$ grow by the factors $L$, $T$ and $M$, the velocity $v_c$ by $LT^{-1}$ and the acceleration $g$ by $LT^{-2}$. These factors are the dimensions of the quantities:

$$ [h] = L, \qquad [t_c] = T, \qquad [m] = M, \qquad [v_c] = LT^{-1}, \qquad [g] = LT^{-2} $$

The dimensions are always power-law monomials $L^{\alpha}T^{\beta}M^{\gamma}$, with $\alpha$, $\beta$, $\gamma$ rational numbers (Notes.pdf, p. 58).

The two dimensionless ratios

Changing the units of length and time changes $(h/g)^{1/2}$ by a factor $T$ and $(gh)^{1/2}$ by a factor $LT^{-1}$; check: $[(h/g)^{1/2}] = (L\cdot L^{-1}T^{2})^{1/2} = T$ and $[(gh)^{1/2}] = (LT^{-2}\cdot L)^{1/2} = LT^{-1}$. Hence the ratios (Notes.pdf, Eqs. 6.2-6.3)

$$ \Pi_1 = \frac{t_c}{(h/g)^{1/2}} \tag{6.2} $$
$$ \Pi_2 = \frac{v_c}{(gh)^{1/2}} \tag{6.3} $$

have dimension $[\Pi_1] = [\Pi_2] = 1$: their value is the same in every system of units. With (6.1) they become (Notes.pdf, Eqs. 6.4-6.5)

$$ \Pi_1 = \frac{t_c(m,h,g)}{(h/g)^{1/2}} = \Pi_1(m, h, g) \tag{6.4} $$
$$ \Pi_2 = \frac{v_c(m,h,g)}{(gh)^{1/2}} = \Pi_2(m, h, g) \tag{6.5} $$

Removing the parameters one by one

The key argument of the whole chapter (Notes.pdf, p. 59):

  1. Change only the unit of mass. The number $m$ changes arbitrarily while $h$ and $g$ stay fixed, and $\Pi_1$, $\Pi_2$ (dimensionless) cannot change. So $\Pi_1$ and $\Pi_2$ do not depend on $m$: $\Pi_1 = \Pi_1(h, g)$, $\Pi_2 = \Pi_2(h, g)$.
  2. Change only the unit of time. The number $g$ is multiplied by $T^{-2}$ while $h$ stays fixed; again $\Pi_1$, $\Pi_2$ do not change, so they do not depend on $g$.
  3. Change the unit of length. $h$ changes arbitrarily and the $\Pi$'s do not, so they do not depend on $h$ either.

What is left is (Notes.pdf, Eqs. 6.6-6.7)

$$ \frac{t_c}{(h/g)^{1/2}} = C_1 \tag{6.6} $$
$$ \frac{v_c}{(gh)^{1/2}} = C_2 \tag{6.7} $$

Dimensional analysis gives the form of the law, not the constants. The integration above gives $C_1 = C_2 = \sqrt{2}$; a single experiment (drop one object, measure $t_c$ and $v_c$) would give them too, and then (6.6)-(6.7) hold for any mass, any height and any planet.

2. Physical Quantities with Independent Dimensions

How many fundamental units

A geometric problem needs one unit (length), a kinematic problem two (length, time), mechanics three (mass, length, time); thermal problems add temperature $\Theta$, electromagnetic ones electric charge. The chapter works with mechanical problems, $[a] = L^{\alpha}T^{\beta}M^{\gamma}$ (Notes.pdf, p. 60).

Definition

The quantities $a_1,\dots,a_k$ have independent dimensions when the dimension of none of them can be written as a product of powers of the dimensions of the other $k-1$.

Example: $\rho$, $U$, $\mu$

Take $[\rho] = ML^{-3}$, $[U] = LT^{-1}$, $[\mu] = ML^{-1}T^{-1}$. Assume the opposite: that two numbers $x$, $y$ exist with $[\rho] = [U]^x[\mu]^y$ (Notes.pdf, Eq. 6.8). Written with the exponents of $(L, M, T)$:

$$ L^{-3}M^{1}T^{0} = \left(L^{1}M^{0}T^{-1}\right)^{x}\left(L^{-1}M^{1}T^{-1}\right)^{y} \tag{6.8} $$

Equating the exponents of $L$, $M$ and $T$ (Notes.pdf, Eq. 6.9):

$$ -3 = x - y, \qquad 1 = y, \qquad 0 = -x - y \tag{6.9} $$

The second equation gives $y = 1$; the first then gives $x = -3 + 1 = -2$; the third requires $0 = -x - y = 2 - 1 = 1$, which is false. No pair $(x, y)$ exists, so $\rho$, $U$, $\mu$ have independent dimensions. By contrast, $p$, $\rho$, $U$ do not: $[\rho][U]^2 = ML^{-3}\cdot L^{2}T^{-2} = ML^{-1}T^{-2} = [p]$, that is $x = 1$, $y = 2$.

The rank test

The equations (6.9) say that the exponent vector of $\rho$ is a combination of those of $U$ and $\mu$ (Notes.pdf, Eq. 6.10):

$$ (-3, 1, 0) = x\,(1, 0, -1) + y\,(-1, 1, -1) \tag{6.10} $$

so $a_1,\dots,a_k$ have independent dimensions if and only if the rank of the matrix of exponents equals $k$. Rows $\rho$, $U$, $\mu$; columns $L$, $M$, $T$ (Notes.pdf, Eq. 6.11):

$$ \begin{pmatrix} -3 & 1 & 0 \\ 1 & 0 & -1 \\ -1 & 1 & -1 \end{pmatrix} \tag{6.11} $$

Expanding along the first row: $\det = -3\,[0\cdot(-1) - (-1)\cdot 1] - 1\,[1\cdot(-1) - (-1)\cdot(-1)] + 0 = -3(1) - 1(-2) = -1 \neq 0$, so the rank is 3. For $\rho$, $U$, $p$ (Notes.pdf, Eq. 6.12):

$$ \begin{pmatrix} -3 & 1 & 0 \\ 1 & 0 & -1 \\ -1 & 1 & -2 \end{pmatrix} \tag{6.12} $$

$\det = -3\,[0\cdot(-2) - (-1)\cdot 1] - 1\,[1\cdot(-2) - (-1)\cdot(-1)] = -3(1) - 1(-3) = 0$, and the first two rows are clearly not proportional, so the rank is 2: the three quantities are not independent.

Exam practice

With three fundamental units the matrix has three columns, so at most three quantities can have independent dimensions. To find $k$, look for the largest set of rows with a non-zero determinant (a $3\times 3$ minor first; if all vanish, a $2\times 2$ one).

3. The Π Theorem (Vaschy-Buckingham Theorem)

The general problem

We look for an unknown quantity $a_0$ that depends on $n$ governing parameters (Notes.pdf, Eq. 6.13):

$$ a_0 = f(a_1,\dots,a_k,a_{k+1},\dots,a_n) \tag{6.13} $$

Let $k \le n$ be the largest number of parameters with independent dimensions, and $a_1,\dots,a_k$ one such set. The dimensions of every other parameter are products of powers of theirs (Notes.pdf, Eq. 6.14):

$$ [a_{k+1}] = [a_1]^{\alpha_{k+1}}\cdots[a_k]^{\gamma_{k+1}}, \quad [a_{k+2}] = [a_1]^{\alpha_{k+2}}\cdots[a_k]^{\gamma_{k+2}}, \quad \dots, \quad [a_n] = [a_1]^{\alpha_{n}}\cdots[a_k]^{\gamma_{n}} \tag{6.14} $$

In general $0 < k < n$; $k = n$ happens in the falling body. The same holds for the unknown (Notes.pdf, Eq. 6.15):

$$ [a_0] = [a_1]^{\alpha_0}\cdots[a_k]^{\gamma_0} \tag{6.15} $$
Two typing slips in the notes (p. 61)

In (6.14) the notes print the exponent of $[a_k]$ in the second line as $\gamma_{k+1}$; it must be $\gamma_{k+2}$, as written above. The sentence after (6.14) reads "or $k = n$, as occurs if all parameters $a_1,\dots,a_n$ are dimensionless"; for dimensionless parameters no unit can change them, so the intended case is $k = 0$.

The dimensionless groups

Dividing each quantity by the product of powers with its own dimensions gives $n - k + 1$ dimensionless numbers (Notes.pdf, Eq. 6.16):

$$ \Pi_0 = \frac{a_0}{a_1^{\alpha_0}\cdots a_k^{\gamma_0}}, \quad \Pi_1 = \frac{a_{k+1}}{a_1^{\alpha_{k+1}}\cdots a_k^{\gamma_{k+1}}}, \quad \Pi_2 = \frac{a_{k+2}}{a_1^{\alpha_{k+2}}\cdots a_k^{\gamma_{k+2}}}, \quad \dots, \quad \Pi_{n-k} = \frac{a_n}{a_1^{\alpha_n}\cdots a_k^{\gamma_n}} \tag{6.16} $$

Dividing (6.13) by $a_1^{\alpha_0}\cdots a_k^{\gamma_0}$ and writing $a_{k+1},\dots,a_n$ through the $\Pi$'s, $a_{k+j} = \Pi_j\,a_1^{\alpha_{k+j}}\cdots a_k^{\gamma_{k+j}}$ (Notes.pdf, Eq. 6.17):

$$ \Pi_0 = \frac{f(a_1,\dots,a_k,a_{k+1},\dots,a_n)}{a_1^{\alpha_0}\cdots a_k^{\gamma_0}} = \frac{f\left(a_1,\dots,a_k,\Pi_1 a_1^{\alpha_{k+1}}\cdots a_k^{\gamma_{k+1}},\dots,\Pi_{n-k}a_1^{\alpha_n}\cdots a_k^{\gamma_n}\right)}{a_1^{\alpha_0}\cdots a_k^{\gamma_0}} \tag{6.17} $$

that is, $\Pi_0$ is a function of $a_1,\dots,a_k$ and of the $\Pi$'s (Notes.pdf, Eq. 6.18):

$$ \Pi_0 = \Pi_0(a_1,\dots,a_k,\Pi_1,\dots,\Pi_{n-k}) \tag{6.18} $$

Worked example: pressure loss in a pipe

The pressure loss per unit length of a liquid flowing along a long circular pipe depends on the volume flux $Q$, the diameter $D$, the density $\rho$ and the viscosity $\mu$ (Notes.pdf, Eq. 6.19):

$$ P_l = -\frac{\partial p}{\partial x} = f(Q, D, \rho, \mu) \tag{6.19} $$

Here $n = 4$. Dimensions: $[Q] = L^3T^{-1}$, $[D] = L$, $[\rho] = ML^{-3}$, $[\mu] = ML^{-1}T^{-1}$, $[P_l] = [p]/L = ML^{-2}T^{-2}$. The set $(Q, D, \rho)$ is independent: $\rho$ is the only one with $M$, $Q$ the only one with $T$, and $D$ adds $L$; so $k = 3$.

Exponents of $\mu$. Write $[\mu] = [Q]^{a}[D]^{b}[\rho]^{c}$ and equate exponents:

$$ M:\; 1 = c, \qquad T:\; -1 = -a, \qquad L:\; -1 = 3a + b - 3c $$

so $c = 1$, $a = 1$ and $b = -1 - 3(1) + 3(1) = -1$ (Notes.pdf, Eq. 6.20):

$$ [\mu] = [Q]^{1}[D]^{-1}[\rho]^{1} \tag{6.20} $$

Exponents of $P_l$. With $[P_l] = [Q]^{a}[D]^{b}[\rho]^{c}$:

$$ M:\; 1 = c, \qquad T:\; -2 = -a, \qquad L:\; -2 = 3a + b - 3c $$

so $c = 1$, $a = 2$ and $b = -2 - 6 + 3 = -5$ (Notes.pdf, Eq. 6.21):

$$ [P_l] = [Q]^{2}[D]^{-5}[\rho]^{1} \tag{6.21} $$

The two dimensionless groups are (Notes.pdf, Eq. 6.22)

$$ \Pi_0 = \frac{P_l}{\rho Q^2/D^5} \qquad\text{and}\qquad \Pi_1 = \frac{\mu}{\rho Q/D} \tag{6.22} $$

and substituting $\mu = \Pi_1\,\rho Q/D$ into (6.19) gives the form (6.18) (Notes.pdf, Eq. 6.23):

$$ \Pi_0 = \frac{f(Q, D, \rho, \Pi_1\rho Q/D)}{\rho Q^2/D^5} = F(Q, D, \rho, \Pi_1) \tag{6.23} $$

Now the argument of Section 1, repeated:

  1. Make the unit of mass $M$ times smaller: $\rho$ grows by $M$, while $\Pi_0$, $\Pi_1$, $Q$, $D$ do not change. So $F$ does not depend on $\rho$: $\Pi_0 = F(Q, D, \Pi_1)$.
  2. Make the unit of time $T$ times smaller: the number $Q$ changes by the factor $T^{-1}$ (since $[Q] = L^3T^{-1}$), while $D$, $\Pi_0$, $\Pi_1$ do not change: $\Pi_0 = F(D, \Pi_1)$.
  3. Change the unit of length: $\Pi_0 = F(\Pi_1)$.

The result (Notes.pdf, Eq. 6.24):

$$ \frac{P_l}{\rho Q^2/D^5} = F\left(\frac{\mu}{\rho Q/D}\right) \tag{6.24} $$

The theorem

Repeating the argument on (6.18): a change of units can always modify $a_1$ alone, leaving $a_2,\dots,a_k$ and all the $\Pi$'s fixed, so $\Pi_0$ does not depend on $a_1$; one by one, it does not depend on $a_2,\dots,a_k$ either (Notes.pdf, Eq. 6.25):

$$ \Pi_0 = F(\Pi_1,\dots,\Pi_{n-k}) \tag{6.25} $$
The Π theorem (Notes.pdf, p. 62)

If an unknown $a_0$ depends on $n$ governing parameters $a_1,\dots,a_n$, the number of parameters can always be reduced to $n - k$, where $k$ is the number of governing parameters with independent dimensions. The $k$ quantities with independent dimensions are used to write the other $n-k$ parameters and the unknown in dimensionless form, which gives (6.25).

Why it matters: fewer experiments

For the pipe, the theorem reduces four parameters to one. A single series of experiments in one pipe with one fluid, varying $Q$ and measuring $P_l$, plotted as $\Pi_0$ against $\Pi_1$, gives the law for any fluid in any pipe. For a new case ($Q^*$, $\rho^*$, $\mu^*$, $D^*$): compute $\Pi_1^* = \mu^*/(\rho^*Q^*/D^*)$, read $\Pi_0^*$ on the curve, and solve $P_l^* = \Pi_0^*\,\rho^*(Q^*)^2/(D^*)^5$ (Notes.pdf, p. 63).

4. Taylor's Analysis of a Nuclear Explosion

Governing parameters

An atomic explosion releases a large energy $E$ almost instantaneously at one point; a strong spherical shock of radius $r_o$ grows with time $t$. In the first stages the pressure $p_o$ behind the shock is thousands of times the ambient pressure, so the ambient pressure does not enter, and (6.13) becomes (Notes.pdf, Eq. 6.26):

$$ r_o = f_1(E, \rho_a, t) \qquad\text{and}\qquad p_o = f_2(E, \rho_a, t) \tag{6.26} $$

Rank

Rows $E$, $\rho_a$, $t$; columns $L$, $M$, $T$, from $[E] = ML^2T^{-2}$, $[\rho_a] = ML^{-3}$, $[t] = T$ (Notes.pdf, Eq. 6.27):

$$ \begin{pmatrix} 2 & 1 & -2 \\ -3 & 1 & 0 \\ 0 & 0 & 1 \end{pmatrix} \tag{6.27} $$

Expanding along the third row: $\det = 1\cdot(2\cdot 1 - 1\cdot(-3)) = 5 \neq 0$. So $k = 3 = n$ and no governing parameter is left: $n - k = 0$.

The exponents

Radius. $[r_o] = L = [E]^{a}[\rho_a]^{b}[t]^{c}$:

$$ M:\; 0 = a + b, \qquad L:\; 1 = 2a - 3b, \qquad T:\; 0 = -2a + c $$

From the first, $b = -a$; the second becomes $1 = 2a + 3a = 5a$, so $a = 1/5$, $b = -1/5$; the third gives $c = 2a = 2/5$.

Pressure. $[p_o] = ML^{-1}T^{-2} = [E]^{a}[\rho_a]^{b}[t]^{c}$:

$$ M:\; 1 = a + b, \qquad L:\; -1 = 2a - 3b, \qquad T:\; -2 = -2a + c $$

With $b = 1 - a$: $-1 = 2a - 3 + 3a$, so $a = 2/5$, $b = 3/5$, and $c = -2 + 2a = -6/5$.

The solution is therefore (Notes.pdf, Eqs. 6.28-6.29)

$$ \frac{r_o}{E^{1/5}\rho_a^{-1/5}t^{2/5}} = C_1 \qquad\text{and}\qquad \frac{p_o}{E^{2/5}\rho_a^{3/5}t^{-6/5}} = C_2 \tag{6.28} $$
$$ r_o = C_1 E^{1/5}\rho_a^{-1/5}t^{2/5} \qquad\text{and}\qquad p_o = C_2 E^{2/5}\rho_a^{3/5}t^{-6/5} \tag{6.29} $$

$C_1$ and $C_2$ come from integrating the compressible-flow equations, or from measurements.

The shock slows down, and when the law stops holding

The shock speed is the time derivative of (6.29), with the power rule $\frac{d}{dt}t^{2/5} = \frac{2}{5}t^{2/5-1} = \frac{2}{5}t^{-3/5}$:

$$ \frac{dr_o}{dt} = \frac{2}{5}\,C_1 E^{1/5}\rho_a^{-1/5}t^{-3/5} $$

Both the speed and $p_o$ decrease with time. The law fails when $p_o$ falls to the order of the ambient pressure $p_a$. Setting $E^{2/5}\rho_a^{3/5}t^{-6/5} \sim p_a$ and solving for $t$: $t^{6/5} \sim E^{2/5}\rho_a^{3/5}/p_a$, and raising to $5/6$,

$$ t \sim \frac{E^{1/3}\rho_a^{1/2}}{p_a^{5/6}} $$

(Notes.pdf, p. 64). The same reasoning applies to the cylindrical shock of a lightning stroke, with $E$ the energy per unit length, $[E] = MLT^{-2}$.

5. Parametric Dependence of the Solution

The hard step is the list of parameters

Missing an important parameter, or adding spurious ones, gives a wrong result. In fluid mechanics the list need not be a guess: write the conservation equations with their initial and boundary conditions, and read the parameters from them (Notes.pdf, p. 64).

R x z y U p + ρgz = P∞ liquid of density ρ and viscosity μ g Fx
Figure 2. Fluid motion around a sphere, Fig. 6.2 of the notes, seen from the sphere: far away the liquid moves with velocity U and its pressure is hydrostatic, p + ρgz = P_∞; the unknown is the drag F_x.

A first guess for the sphere

A sphere of radius $R$ moves with constant velocity $U$ in a stagnant liquid of density $\rho$ and viscosity $\mu$; far away the pressure is hydrostatic, $p + \rho g z = P_\infty$. A first guess for the drag (Notes.pdf, Eq. 6.30):

$$ F_x = F_x(R, U, \rho, \mu, P_\infty, g) \tag{6.30} $$

Exponent matrix, rows $R$, $U$, $\rho$, $\mu$, $P_\infty$, $g$; columns $L$, $M$, $T$ (Notes.pdf, Eq. 6.31):

$$ \begin{pmatrix} 1 & 0 & 0 \\ 1 & 0 & -1 \\ -3 & 1 & 0 \\ -1 & 1 & -1 \\ -1 & 1 & -2 \\ 1 & 0 & -2 \end{pmatrix} \tag{6.31} $$

The rank of a $6\times 3$ matrix is at most 3, and the first three rows give $\det = 1\cdot(0\cdot 0 - (-1)\cdot 1) - 0 + 0 = 1 \neq 0$, so $k = 3$ with $(R, U, \rho)$. Since $[F_x] = MLT^{-2} = [\rho][U]^2[R]^2$, the dimensionless force is (Notes.pdf, Eq. 6.32)

$$ \Pi_0 = \frac{F_x}{\rho U^2 R^2} \tag{6.32} $$

Check the other three: $[\rho][U][R] = ML^{-3}\cdot LT^{-1}\cdot L = ML^{-1}T^{-1} = [\mu]$; $[\rho][U]^2 = ML^{-1}T^{-2} = [P_\infty]$; $[U]^2[R]^{-1} = LT^{-2} = [g]$. So (Notes.pdf, Eq. 6.33)

$$ \Pi_1 = \frac{\mu}{\rho U R}, \qquad \Pi_2 = \frac{P_\infty}{\rho U^2}, \qquad \Pi_3 = \frac{g}{U^2/R} \tag{6.33} $$

and the theorem gives (Notes.pdf, Eq. 6.34)

$$ \frac{F_x}{\rho U^2 R^2} = f\left(\frac{\mu}{\rho U R}, \frac{P_\infty}{\rho U^2}, \frac{g}{U^2/R}\right) \tag{6.34} $$

Six parameters became three, but (6.34) is wrong: $P_\infty$ and $g$ do not influence the drag. With the correct list $F_x = F_x(R, U, \rho, \mu)$ the result is (Notes.pdf, Eq. 6.35)

$$ \frac{F_x}{\rho U^2 R^2} = f\left(\frac{\mu}{\rho U R}\right) \tag{6.35} $$

Reading the parameters from the equations

In the frame of the sphere the liquid satisfies (Notes.pdf, Eq. 6.36)

$$ \nabla\cdot\bar{v} = 0, \quad \rho\,\bar{v}\cdot\nabla\bar{v} = -\nabla p + \mu\nabla^2\bar{v} - \rho g\,\bar{e}_z; \qquad \lvert\bar{x}\rvert\to\infty:\; p = P_\infty - \rho g z,\; \bar{v} = U\bar{e}_x; \qquad \lvert\bar{x}\rvert = R:\; \bar{v} = 0 \tag{6.36} $$

and the force on the sphere is the integral of the pressure and of the viscous stress $\bar{\bar{\tau}}' = \mu(\nabla\bar{v} + \nabla\bar{v}^T)$ over its surface (Notes.pdf, Eqs. 6.37-6.38):

$$ \bar{F} = -\int_{\lvert\bar{x}\rvert = R} p\,\bar{n}\,d\sigma + \int_{\lvert\bar{x}\rvert = R}\mu(\nabla\bar{v} + \nabla\bar{v}^T)\cdot\bar{n}\,d\sigma \tag{6.37} $$
$$ F_x = \bar{F}\cdot\bar{e}_x = -\int_{\lvert\bar{x}\rvert = R} p\,n_x\,d\sigma + \int_{\lvert\bar{x}\rvert = R}\mu\left[(\nabla\bar{v} + \nabla\bar{v}^T)\cdot\bar{n}\right]\cdot\bar{e}_x\,d\sigma \tag{6.38} $$

At first sight $P_\infty$ and $g$ do appear. Two observations remove them:

  1. Gravity is a gradient, $-\nabla p - \rho g\bar{e}_z = -\nabla(p + \rho g z)$, so it can be absorbed in the reduced pressure.
  2. Only pressure differences move the fluid; the level $P_\infty$ does not matter.

Both suggest $P' = p + \rho g z - P_\infty$. In terms of $P'$ (Notes.pdf, Eq. 6.39):

$$ \nabla\cdot\bar{v} = 0, \quad \rho\,\bar{v}\cdot\nabla\bar{v} = -\nabla P' + \mu\nabla^2\bar{v}; \qquad \lvert\bar{x}\rvert\to\infty:\; P' = 0,\; \bar{v} = U\bar{e}_x; \qquad \lvert\bar{x}\rvert = R:\; \bar{v} = 0 \tag{6.39} $$

Neither $P_\infty$ nor $g$ appears, so $P'(\bar{x})$ and $\bar{v}(\bar{x})$ do not depend on them. Substituting $p = P' + P_\infty - \rho g z$ in (6.37), and using $\int P_\infty\bar{n}\,d\sigma = 0$ (a constant pressure on a closed surface gives no force) (Notes.pdf, Eq. 6.40):

$$ \bar{F} = -\int_{\lvert\bar{x}\rvert = R} P'\,\bar{n}\,d\sigma + \int_{\lvert\bar{x}\rvert = R}\mu(\nabla\bar{v} + \nabla\bar{v}^T)\cdot\bar{n}\,d\sigma + \int_{\lvert\bar{x}\rvert = R}\rho g z\,\bar{n}\,d\sigma \tag{6.40} $$

The last term is Archimedes' buoyancy. By the Gauss theorem over the volume $V$ of the sphere ($\bar{n}$ pointing out of the sphere) and $\nabla(\rho g z) = \rho g\,\bar{e}_z$ (Notes.pdf, Eq. 6.41):

$$ \int_{\lvert\bar{x}\rvert = R}\rho g z\,\bar{n}\,d\sigma = \int_V \nabla(\rho g z)\,dV = \rho g\,\frac{4}{3}\pi R^3\,\bar{e}_z \tag{6.41} $$

It is vertical, so it does not contribute to the drag (Notes.pdf, Eq. 6.42):

$$ F_x = \bar{F}\cdot\bar{e}_x = -\int_{\lvert\bar{x}\rvert = R} P'\,n_x\,d\sigma + \int_{\lvert\bar{x}\rvert = R}\mu\left[(\nabla\bar{v} + \nabla\bar{v}^T)\cdot\bar{n}\right]\cdot\bar{e}_x\,d\sigma \tag{6.42} $$

This proves that $F_x = F_x(R, U, \rho, \mu)$ and that (6.35), not (6.34), is the right result.

When pressure level and gravity do matter

  • Gases: the absolute pressure enters through the equation of state $p = \rho R_g T$.
  • Liquids with a free surface that is not flat (waves): the condition $p = p_a$ on the unknown surface $z_s(x, y)$ brings $g$ back. A sphere moving at a depth of order $R$ below the surface feels a drag that depends on $g$ and on the depth, and the reduced pressure no longer removes $g$ (Notes.pdf, p. 66). This is why $g$ appears in the ship, tsunami and wave-maker problems (DA1, DA6, DA11).

The repeating set is not unique

Choosing $(U, R, \mu)$ instead of $(U, R, \rho)$ gives (Notes.pdf, Eq. 6.43)

$$ \frac{F_x}{\mu U R} = f\left(\frac{\rho U R}{\mu}\right) \tag{6.43} $$

Both (6.35) and (6.43) are valid. In both, the dimensionless force depends on the Reynolds number (Notes.pdf, Eq. 6.44):

$$ \mathrm{Re} = \frac{\rho U R}{\mu} \tag{6.44} $$

6. Nondimensionalization of the Navier-Stokes Equations

Scales from the boundary conditions

Instead of the $\Pi$ theorem, the problem itself can be written in dimensionless variables. For the sphere, the boundary conditions give the velocity scale $U$ and the length scale $R$: $\bar{v}^* = \bar{v}/U$, $\bar{x}^* = \bar{x}/R$, hence $\nabla^* = R\nabla$ and $d\sigma^* = R^{-2}d\sigma$ (Notes.pdf, p. 67). The scale $\Delta p$ of $P'$ is not obvious: it depends on the kind of motion.

Order of magnitude of each term

Velocities are of order $U$ and change by $O(U)$ over distances $O(R)$, so $O(\nabla\bar{v}) \sim U/R$ and $O(\nabla^2\bar{v}) \sim U/R^2$ (Notes.pdf, Eq. 6.45):

$$ \underbrace{\rho\,\bar{v}\cdot\nabla\bar{v}}_{\rho U^2/R} = \underbrace{-\nabla P'}_{\Delta p/R} + \underbrace{\mu\nabla^2\bar{v}}_{\mu U/R^2} \tag{6.45} $$

The pressure gradient is almost always essential: without it, (6.39) would have three unknowns (the velocity components) and four equations (continuity and three momentum components). The ratio of the other two terms is (Notes.pdf, Eq. 6.46)

$$ \frac{O(\rho\,\bar{v}\cdot\nabla\bar{v})}{O(\mu\nabla^2\bar{v})} = \frac{\rho U^2/R}{\mu U/R^2} = \frac{\rho U R}{\mu} \tag{6.46} $$

the Reynolds number (6.44). Three regimes (Notes.pdf, p. 67):

RegimeBalance in (6.45)Pressure scale
$\mathrm{Re}\sim 1$all three termseither estimate
$\mathrm{Re}\gg 1$$\rho\,\bar{v}\cdot\nabla\bar{v} = -\nabla P'$ (viscosity negligible)$\Delta p \sim \rho U^2$
$\mathrm{Re}\ll 1$$0 = -\nabla P' + \mu\nabla^2\bar{v}$ (acceleration negligible)$\Delta p \sim \mu U/R$

Scaling for $\mathrm{Re}\gtrsim 1$

With $p^* = (p + \rho g z - P_\infty)/(\rho U^2)$, substitute $\bar{v} = U\bar{v}^*$, $\bar{x} = R\bar{x}^*$, $\nabla = \nabla^*/R$, $P' = \rho U^2 p^*$ in each term of (6.39):

$$ \rho\,\bar{v}\cdot\nabla\bar{v} = \frac{\rho U^2}{R}\,\bar{v}^*\cdot\nabla^*\bar{v}^*, \qquad \nabla P' = \frac{\rho U^2}{R}\nabla^* p^*, \qquad \mu\nabla^2\bar{v} = \frac{\mu U}{R^2}\nabla^{*2}\bar{v}^* $$

Dividing by $\rho U^2/R$, the viscous term carries $\mu U/R^2 \div \rho U^2/R = \mu/(\rho U R) = 1/\mathrm{Re}$. The problem (6.39) becomes (Notes.pdf, Eq. 6.47)

$$ \nabla^*\cdot\bar{v}^* = 0, \quad \bar{v}^*\cdot\nabla^*\bar{v}^* = -\nabla^* p^* + \frac{1}{\mathrm{Re}}\nabla^{*2}\bar{v}^*; \qquad \lvert\bar{x}^*\rvert\to\infty:\; p^* = 0,\; \bar{v}^* = \bar{e}_x; \qquad \lvert\bar{x}^*\rvert = 1:\; \bar{v}^* = 0 \tag{6.47} $$

In the force (6.42), $P'\,d\sigma = \rho U^2 R^2\,p^*\,d\sigma^*$ and the viscous stress is $\mu(U/R)(\nabla^*\bar{v}^* + \nabla^*\bar{v}^{*T})$, which times $d\sigma = R^2 d\sigma^*$ gives $\mu U R\,(\dots)\,d\sigma^*$. Dividing by $\rho U^2 R^2$ (Notes.pdf, Eq. 6.48):

$$ \frac{F_x}{\rho U^2 R^2} = -\int_{\lvert\bar{x}^*\rvert = 1} p^*\,\bar{n}\cdot\bar{e}_x\,d\sigma^* + \frac{1}{\mathrm{Re}}\int_{\lvert\bar{x}^*\rvert = 1}\left[(\nabla^*\bar{v}^* + \nabla^*\bar{v}^{*T})\cdot\bar{n}\right]\cdot\bar{e}_x\,d\sigma^* \tag{6.48} $$

The only parameter left is $\mathrm{Re}$: $F_x/(\rho U^2R^2)$ is a function of $\mathrm{Re}$ alone, exactly (6.35).

Reference in the notes (p. 68)

The notes say that this scaling reduces "(6.40)"; the problem being scaled is the system (6.39) (the field equations and boundary conditions), while the force is (6.42). The same applies to the sentence before (6.49).

Scaling for $\mathrm{Re}\lesssim 1$

With $p^* = (p + \rho g z - P_\infty)/(\mu U/R)$, now $\nabla P' = (\mu U/R^2)\nabla^* p^*$; dividing every term by $\mu U/R^2$, the convective term carries $\rho U^2/R \div \mu U/R^2 = \mathrm{Re}$ (Notes.pdf, Eq. 6.49):

$$ \nabla^*\cdot\bar{v}^* = 0, \quad \mathrm{Re}\,\bar{v}^*\cdot\nabla^*\bar{v}^* = -\nabla^* p^* + \nabla^{*2}\bar{v}^*; \qquad \lvert\bar{x}^*\rvert\to\infty:\; p^* = 0,\; \bar{v}^* = \bar{e}_x; \qquad \lvert\bar{x}^*\rvert = 1:\; \bar{v}^* = 0 \tag{6.49} $$

and the force, divided by $\mu U R$ (Notes.pdf, Eq. 6.50):

$$ \frac{F_x}{\mu U R} = -\int_{\lvert\bar{x}^*\rvert = 1} p^*\,\bar{n}\cdot\bar{e}_x\,d\sigma^* + \int_{\lvert\bar{x}^*\rvert = 1}\left[(\nabla^*\bar{v}^* + \nabla^*\bar{v}^{*T})\cdot\bar{n}\right]\cdot\bar{e}_x\,d\sigma^* \tag{6.50} $$

so $F_x/(\mu U R)$ is a function of $\mathrm{Re}$ only, the result (6.43).

Π theorem or nondimensional equations?

Both give the same groups, but the equations tell you which parameters really enter (no spurious $P_\infty$ or $g$) and which terms are small in each limit. The $\Pi$ theorem alone cannot do either. Problem DA14 compares the two methods on the rotating cavity.

7. Selection of Parameters with Independent Dimensions

The rule

The set $a_1,\dots,a_k$ should contain parameters with a significant influence for the flow considered (Notes.pdf, p. 68). For the sphere, $R$ and $U$ obviously matter. The third one:

  • if viscosity is important, $\mu$, which leads to (6.43);
  • if viscosity is not expected to matter, $\rho$, which leads to (6.35).

Evaluate $\mathrm{Re} = \rho U_c L_c/\mu$ with the characteristic velocity and size: $\mathrm{Re}\lesssim 1$ suggests $(U_c, L_c, \mu)$; $\mathrm{Re}\gtrsim 1$ suggests $(U_c, L_c, \rho)$.

The two extreme limits

For $\mathrm{Re}\gg 1$ the $1/\mathrm{Re}$ term of (6.47)-(6.48) disappears, so the dimensionless force no longer depends on $\mathrm{Re}$ (Notes.pdf, Eq. 6.51):

$$ \lim_{\mathrm{Re}\gg 1}\frac{F_x}{\rho U^2 R^2} = C_1 \tag{6.51} $$

For $\mathrm{Re}\ll 1$ the $\mathrm{Re}$ term of (6.49) disappears (Notes.pdf, Eq. 6.52):

$$ \lim_{\mathrm{Re}\ll 1}\frac{F_x}{\mu U R} = C_2 \tag{6.52} $$

With the "wrong" set, the limit still depends on $\mathrm{Re}$. Since $\frac{F_x}{\rho U^2R^2} = \frac{F_x}{\mu U R}\cdot\frac{\mu U R}{\rho U^2 R^2} = \frac{F_x}{\mu U R}\cdot\frac{1}{\mathrm{Re}}$, (6.52) gives (Notes.pdf, Eq. 6.53)

$$ \lim_{\mathrm{Re}\ll 1}\frac{F_x}{\rho U^2 R^2} = C_2\,\mathrm{Re}^{-1} \tag{6.53} $$

and, since $\frac{F_x}{\mu U R} = \frac{F_x}{\rho U^2R^2}\cdot\mathrm{Re}$, (6.51) gives (Notes.pdf, Eq. 6.54)

$$ \lim_{\mathrm{Re}\gg 1}\frac{F_x}{\mu U R} = C_1\,\mathrm{Re} \tag{6.54} $$
log Re log Fx / (ρU² R²) Re ∼ 1 slope −1: C2 Re⁻¹, Eq. (6.53) C1, Eq. (6.51) Sketch of the two limits only: the constants C₁ and C₂ are not given in the notes
Figure 3. Sketch, on logarithmic axes, of the two limits of Eq. (6.35): for Re ≪ 1 the dimensionless drag falls as C₂/Re, Eq. (6.53); for Re ≫ 1 it tends to the constant C₁, Eq. (6.51). Only the slopes are drawn: the constants and the shape of the curve in between are not in the notes.

In dimensional terms: for $\mathrm{Re}\ll 1$, $F_x = C_2\,\mu U R$ (drag linear in $U$, independent of $\rho$); for $\mathrm{Re}\gg 1$, $F_x = C_1\,\rho U^2 R^2$ (drag quadratic in $U$, independent of $\mu$).

Typing slip in the notes (p. 69)

Before (6.51) the notes write "$F_x/(\rho U R^2)$ becomes independent of $Re$"; the quantity is $F_x/(\rho U^2 R^2)$, as in (6.51).

General form of the limits

When one group of (6.25) takes extreme values, say $\Pi_1\gg 1$, two things can happen (Notes.pdf, Eqs. 6.55-6.56):

$$ \lim_{\Pi_1\gg 1}\Pi_0(\Pi_1,\dots,\Pi_{n-k}) = f(\Pi_2,\dots,\Pi_{n-k}) \tag{6.55} $$
$$ \lim_{\Pi_1\gg 1}\Pi_0(\Pi_1,\dots,\Pi_{n-k}) = \Pi_1^{\lambda}\,f(\Pi_2,\dots,\Pi_{n-k}) \tag{6.56} $$

In (6.55) one parameter disappears; in (6.56) it survives only as a power, with $\lambda$ to be found. A good choice of $a_1,\dots,a_k$ usually gives the simpler form (6.55).

Exam practice

In the problems, "simplify assuming the viscosity has a negligible effect" means: use $\rho$ (not $\mu$) in the repeating set, then drop the group that contains $\mu$. Always finish by checking the assumption with numbers: $\mathrm{Re}\gg 1$ for both the model and the real system (DA1, DA6, DA12).

8. Physical Similarity

Designing an experiment

A full-scale phenomenon often cannot be reproduced in the laboratory. Example of the notes: the drag $F_1$ on a sphere of radius $R_1$ moving at $U_1$ in an oil ($\rho_1$, $\mu_1$), to be measured in a water tunnel ($\rho_2$, $\mu_2$) with the only available spheres, of radius $R_2 \neq R_1$ (Notes.pdf, p. 70).

By (6.35), $F_x/(\rho U^2R^2)$ depends only on $\mathrm{Re}$. If the experiment has the same Reynolds number, $\rho_2U_2R_2/\mu_2 = \rho_1U_1R_1/\mu_1$, solving for $U_2$ (Notes.pdf, Eq. 6.57):

$$ U_2 = \frac{\mu_2}{\mu_1}\,\frac{\rho_1}{\rho_2}\,\frac{R_1}{R_2}\,U_1 \tag{6.57} $$

Then the dimensionless forces are equal (Notes.pdf, Eq. 6.58):

$$ \frac{F_1}{\rho_1U_1^2R_1^2} = \frac{F_2}{\rho_2U_2^2R_2^2} \tag{6.58} $$

and the measured $F_2$ gives the real force (Notes.pdf, Eq. 6.59):

$$ F_1 = \frac{\rho_1U_1^2R_1^2}{\rho_2U_2^2R_2^2}\,F_2 \tag{6.59} $$

The general rule

For $\Pi_0 = F(\Pi_1,\dots,\Pi_{n-k})$, the model does not have to reproduce $a_1,\dots,a_n$. It is physically similar to the real system when every governing group $\Pi_1,\dots,\Pi_{n-k}$ takes the same value in both; then $\Pi_0$ is equal too, and the measured $a_0$ is converted to the real one by equating $\Pi_0$ (Notes.pdf, p. 71).

When not all groups can be matched, similarity is partial: match the groups that dominate the problem and accept different values of the secondary ones (Notes.pdf, p. 71).

Extension: why Reynolds and Froude cannot both be matched with the same fluid

This subsection is an extension used in problems DA1 and DA3. A ship or a stirred free surface depends on $\mathrm{Re} = UL/\nu$ (with $\nu = \mu/\rho$) and on the Froude number $\mathrm{Fr} = U/\sqrt{gL}$. Model $m$, prototype $p$, scale $\lambda = L_m/L_p$, same $g$:

$$ \mathrm{Fr}_m = \mathrm{Fr}_p \;\Longrightarrow\; \frac{U_m}{U_p} = \sqrt{\frac{L_m}{L_p}} = \lambda^{1/2}, \qquad \mathrm{Re}_m = \mathrm{Re}_p \;\Longrightarrow\; \frac{U_m}{U_p} = \frac{\nu_m}{\nu_p}\,\frac{L_p}{L_m} = \frac{\nu_m}{\nu_p}\,\lambda^{-1} $$

Equating both velocity ratios: $\nu_m/\nu_p = \lambda^{3/2}$. With the same fluid ($\nu_m = \nu_p$) this forces $\lambda = 1$: no scaled model can match both. The two ways out are those of the problems:

  • neglect viscosity when $\mathrm{Re}\gg 1$ in model and prototype, and match only Froude (DA1, DA6, DA11);
  • change the fluid so that $\nu_m = \lambda^{3/2}\nu_p$ (DA3, tar replaced by water).

9. Practical Formulary of Topic 6

Dimensions in $[M, L, T, \Theta]$

QuantitySymbolDimensions
Velocity$U$$LT^{-1}$
Acceleration, gravity$g$$LT^{-2}$
Angular velocity, frequency$\Omega$, $f$$T^{-1}$
Volume flux (3-D); per unit width (2-D)$Q$; $q$$L^3T^{-1}$; $L^2T^{-1}$
Density$\rho$$ML^{-3}$
Dynamic viscosity$\mu$$ML^{-1}T^{-1}$
Kinematic viscosity$\nu = \mu/\rho$$L^2T^{-1}$
Pressure, stress$p$, $\tau$$ML^{-1}T^{-2}$
Pressure gradient$P_l = -\partial p/\partial x$$ML^{-2}T^{-2}$
Force; force per unit length$F$; $F/L$$MLT^{-2}$; $MT^{-2}$
Energy, torque$E$, $M$$ML^2T^{-2}$
Power$W$$ML^2T^{-3}$
Speed of sound$c$$LT^{-1}$
Temperature (extension)$T$$\Theta$
Thermal conductivity (extension)$k$$MLT^{-3}\Theta^{-1}$
Specific heat (extension)$c_p$$L^2T^{-2}\Theta^{-1}$

Dimensionless numbers

Only the Reynolds number is defined in Chapter 6; the others are extensions, given here because the problems use them.

NumberDefinitionMeasuresWhere
Reynolds$\mathrm{Re} = \rho U L/\mu = UL/\nu$convection / viscous forcesEq. (6.44), (6.46)
Froude (extension)$\mathrm{Fr} = U/\sqrt{gL}$inertia / gravityDA1, DA3, DA6, DA11, DA13
Mach (extension)$\mathrm{Ma} = U/c$flow speed / speed of soundDA9, DA12
Strouhal (extension)$\mathrm{St} = fL/U$unsteadiness / convectionDA2 (shedding period)
Euler (extension)$\mathrm{Eu} = \Delta p/(\rho U^2)$pressure differences / dynamic pressurescaling of (6.47)
Prandtl (extension)$\mathrm{Pr} = \mu c_p/k$momentum / heat diffusionenergy equation
Eckert (extension)$\mathrm{Ec} = U^2/(c_p\,\Delta T)$kinetic energy / enthalpy differencesenergy equation

Key results

SituationResultSource
Independent dimensionsrank of the exponent matrix $= k$Eqs. (6.10)-(6.12)
Π theorem$a_0 = f(a_1,\dots,a_n) \;\Rightarrow\; \Pi_0 = F(\Pi_1,\dots,\Pi_{n-k})$Eqs. (6.13)-(6.25)
$k = n$ (no group left)$\Pi_0 = C$, constantEqs. (6.6)-(6.7), (6.28)
Pipe flow$P_l D^5/(\rho Q^2) = F\left(\mu D/(\rho Q)\right)$Eq. (6.24)
Strong explosion$r_o = C_1E^{1/5}\rho_a^{-1/5}t^{2/5}$, $p_o = C_2E^{2/5}\rho_a^{3/5}t^{-6/5}$Eq. (6.29)
Reduced pressure removes $g$ and $P_\infty$$P' = p + \rho g z - P_\infty$ (not with a free surface)Eq. (6.39)
Sphere drag$F_x/(\rho U^2R^2) = f(\mathrm{Re})$, or $F_x/(\mu UR) = f(\mathrm{Re})$Eqs. (6.35), (6.43)
$\mathrm{Re}\gg 1$$F_x = C_1\rho U^2R^2$; pressure scale $\rho U^2$Eqs. (6.47), (6.51)
$\mathrm{Re}\ll 1$$F_x = C_2\mu UR$; pressure scale $\mu U/R$Eqs. (6.49), (6.52)
Similarityequal $\Pi_1,\dots,\Pi_{n-k}$ $\Rightarrow$ equal $\Pi_0$Eqs. (6.57)-(6.59)
Re and Fr with the same $g$ (extension)needs $\nu_m/\nu_p = (L_m/L_p)^{3/2}$Section 8

Exam recipe

  1. List the parameters. Write the unknown $a_0$ and everything it depends on. In doubt, write the equations and boundary conditions: geometry, velocities or rotation rates imposed at the boundaries, fluid properties that appear ($\rho$, $\mu$; $c$ for a gas at high speed; $g$ only with a free surface or a weight). Count $n$.
  2. Build the exponent matrix (rows: parameters; columns: $M$, $L$, $T$) and find $k$ with a non-zero determinant.
  3. Choose the repeating set of $k$ parameters: those that clearly matter. Use $\rho$ if viscosity is expected to be negligible ($\mathrm{Re}\gg 1$), $\mu$ if inertia is ($\mathrm{Re}\ll 1$). Never put the unknown in the set.
  4. Form each Π by solving the system of exponents, one equation per fundamental unit, and check that each group is dimensionless.
  5. Simplify with the limits stated in the problem: drop the group that contains the negligible parameter (6.55), or keep it as a power (6.56) when the set was not the natural one.
  6. Similarity. Equate every remaining governing group between model and prototype to get the test conditions (velocity, rotation rate, fluid, mass). Equate $\Pi_0$ to convert the measurement. If two groups cannot both be matched, match the dominant one and check numerically that the neglected group is extreme in both systems.
  7. Tables of measurements. Each row of a model table maps to a prototype row through the same scale factors (one for the independent variable, one for the unknown).
  8. Check the dimensions of the final formula and its limits.