Differentials and Error Propagation

Core Definition

For a differentiable function $z = f(x, y)$ , let $dx = \Delta x$ and $dy = \Delta y$ represent small independent changes (differentials) in the input variables. The total differential $dz$ (or $df$ ) is defined as:

Total Differential Equation: $dz = \frac{\partial f}{\partial x} dx + \frac{\partial f}{\partial y} dy = f_x(x, y) dx + f_y(x, y) dy$
Increment Approximation: $\Delta z = f(x + \Delta x, y + \Delta y) - f(x, y) \approx dz$
Maximum Absolute Error Bound: $|\Delta z| \lesssim |dz|_{\max} = \left|\frac{\partial f}{\partial x}\right| |dx| + \left|\frac{\partial f}{\partial y}\right| |dy|$
Relative and Percentage Error: $\text{Relative Error} = \frac{dz}{z}, \quad \text{Percentage Error} = \frac{dz}{z} \times 100\%$

Concept Explanation

Geometric Distinction: $\Delta z$ versus $dz$

Understanding the fundamental difference between the true increment $\Delta z$ and the differential $dz$ is essential in engineering error analysis:

$\Delta z$ (True Increment): The actual vertical change in height measured directly along the curved surface $z = f(x, y)$ as the input moves from $(x_0, y_0)$ to $(x_0 + \Delta x, y_0 + \Delta y)$ .
$dz$ (Total Differential): The vertical change in height measured along the flat tangent plane at $(x_0, y_0)$ .

Because the tangent plane is the best linear approximation to the surface, $dz$ closely tracks $\Delta z$ for small variations in $dx$ and $dy$ . As $dx, dy \to 0$ , the discrepancy between $\Delta z$ and $dz$ shrinks faster than the distance $\sqrt{(\Delta x)^2 + (\Delta y)^2}$ .

             ^ z
             |                     . True Surface point (z + \Delta z)
             |                    /
             |                   /|  \Delta z (True change)
             |      dz          / |
             |   +-------------+  |
             |  /|             |  |
             | / | dz          |  . Tangent plane point (z + dz)
             |/  |             | /
             +---|-------------+/--------------> y
             |   (x_0, y_0)    (x_0 + dx, y_0 + dy)
             v x

Physical Sensitivity and Sensitivity Coefficients

In engineering systems, output performance $z$ depends on several independently measured inputs $(x, y)$ . The partial derivatives in the total differential formula act as sensitivity coefficients:

dz = \underbrace{\left(\frac{\partial f}{\partial x}\right)}_{\text{Sensitivity to } x} dx + \underbrace{\left(\frac{\partial f}{\partial y}\right)}_{\text{Sensitivity to } y} dy

A large magnitude for $|\frac{\partial f}{\partial x}|$ indicates that the system is hypersensitive to small perturbations or measurement errors in $x$ . Conversely, a small value indicates robustness against fluctuations in that variable. Engineers use this sensitivity weighting to decide where to invest in expensive high-precision instrumentation versus where cheaper sensors suffice.

Worst-Case Error Bounds (The Triangle Inequality)

When instruments measure physical parameters $x$ and $y$ with known manufacturing tolerances or measurement uncertainties $|dx| \le \delta_x$ and $|dy| \le \delta_y$ , the errors can have either positive or negative signs. In safety-critical engineering, we calculate the worst-case maximum absolute error by applying the Triangle Inequality, assuming the errors reinforce each other constructively:

|dz| = \left|f_x dx + f_y dy\right| \le |f_x| |dx| + |f_y| |dy| \le |f_x| \delta_x + |f_y| \delta_y

Extension to Three or More Variables

For a system governed by $n$ variables $w = f(x_1, x_2, \dots, x_n)$ :

dw = \sum_{i=1}^n \frac{\partial f}{\partial x_i} dx_i = \frac{\partial f}{\partial x_1} dx_1 + \frac{\partial f}{\partial x_2} dx_2 + \dots + \frac{\partial f}{\partial x_n} dx_n

|dw|_{\max} \le \sum_{i=1}^n \left|\frac{\partial f}{\partial x_i}\right| \delta_{x_i}

Engineering Applications

Manufacturing Quality Assurance (QA/QC): Establishing dimensional tolerances on machined shafts, bolt diameters, and housing bores so that assembled clearances remain within safe operational limits.
Electrical Circuit Tolerance Analysis: Computing the total uncertainty in electrical power dissipation $P = I^2 R$ or resonant frequency $\omega = \frac{1}{\sqrt{LC}}$ resulting from component manufacturing tolerances ( $\pm 5\%$ resistors and capacitors).
Civil & Geotechnical Surveying: Propagating measurement inaccuracies from laser total stations (angle and distance errors) into computed boundary land areas and building foundation elevations.
Fluid Mechanics & Pipeline Metering: Calculating the propagated uncertainty in volumetric flow rate $Q = \frac{\pi D^4 \Delta P}{128 \mu L}$ (Hagen-Poiseuille equation) caused by pipe diameter manufacturing variations.

Example Problems

Example 1: Cylindrical Storage Tank Error Propagation

The volume of a cylindrical fluid storage tank is given by $V(r, h) = \pi r^2 h$ . The radius is measured as $r = 2.0\text{ m}$ with an uncertainty of $\pm 0.02\text{ m}$ , and height is measured as $h = 5.0\text{ m}$ with an uncertainty of $\pm 0.05\text{ m}$ .

Compute the total differential $dV$ .
Estimate the maximum absolute error in the calculated volume.
Determine the relative error $\frac{dV}{V}$ and percentage error.

Solution

Compute Partial Derivatives: $\frac{\partial V}{\partial r} = 2\pi r h, \quad \frac{\partial V}{\partial h} = \pi r^2$ State the Total Differential Equation: $dV = \frac{\partial V}{\partial r} dr + \frac{\partial V}{\partial h} dh = 2\pi r h \, dr + \pi r^2 \, dh$
Estimate Maximum Absolute Error at $(r, h) = (2.0, 5.0)$ : $\frac{\partial V}{\partial r}(2, 5) = 2\pi(2.0)(5.0) = 20\pi$ $\frac{\partial V}{\partial h}(2, 5) = \pi(2.0)^2 = 4\pi$ Applying worst-case tolerances $|dr| \le 0.02\text{ m}$ and $|dh| \le 0.05\text{ m}$ : $|dV|_{\max} \le |20\pi|(0.02) + |4\pi|(0.05) = 0.40\pi + 0.20\pi = 0.60\pi \approx 1.885\text{ m}^3$
Compute Relative and Percentage Error: Nominal volume: $V(2, 5) = \pi(2.0)^2(5.0) = 20\pi\text{ m}^3 \approx 62.83\text{ m}^3$ . $\text{Relative Error} = \frac{dV}{V} = \frac{2\pi r h \, dr + \pi r^2 \, dh}{\pi r^2 h} = 2\frac{dr}{r} + \frac{dh}{h}$ $\frac{dV}{V} = 2\left(\frac{0.02}{2.0}\right) + \left(\frac{0.05}{5.0}\right) = 2(0.01) + 0.01 = 0.02 + 0.01 = 0.030$ $\text{Percentage Error} = 0.030 \times 100\% = 3.0\%$ Engineering Takeaway: Notice that radius contributes twice as much relative error as height ( $2 \frac{dr}{r}$ vs $\frac{dh}{h}$ ) because radius is squared in the volume equation.

Example 2: Kinetic Energy Sensitivity in Mechanical Vehicles

The kinetic energy of a moving drone is $K(m, v) = \frac{1}{2} m v^2$ . A drone's mass is measured as $m = 4.0\text{ kg}$ with an uncertainty of $\pm 0.1\text{ kg}$ , and its velocity is measured as $v = 10.0\text{ m/s}$ with an uncertainty of $\pm 0.2\text{ m/s}$ . Estimate the percentage uncertainty in kinetic energy.