Uncertainty and Error Propagation

When Converting Between Units

Article By Industries Needs

Every physical measurement is an approximation. Whether measuring length with a wooden ruler, determining mass using an analytical balance, or recording a time interval with a digital stopwatch, the

recorded value inherently carries a degree of uncertainty. In scientific and engineering contexts, reported values are incomplete without a clear statement of this experimental error or uncertainty.

When converting physical quantities from one unit system to another—such as changing miles per hour to meters per second, or converting Fahrenheit to Kelvin—the associated uncertainties do not disappear. Instead, they transform according to mathematical rules. Understanding how uncertainty and error propagate through unit conversions is essential for maintaining data integrity in physics, chemistry, engineering, and metrology.

1. Fundamentals of Measurement, Error, and Uncertainty

Absolute vs. Relative Error

To quantify uncertainty in physical measurements, two primary definitions are used:

• Absolute Uncertainty ($\Delta x$): The margin of uncertainty associated with a measurement $x$, expressed in the same units as $x$ itself. A measurement expressed as $(10.0 \pm 0.2)\text{ cm}$ has an absolute uncertainty of $0.2\text{ cm}$.
  
  
• Relative Uncertainty ($\delta x$): The ratio of the absolute uncertainty to the measured quantity, often expressed as a percentage or fraction:
  
  

$$\delta x = \frac{\Delta x}{\vert{}x\vert{}}$$

For the measurement $(10.0 \pm 0.2)\text{ cm}$, the relative uncertainty is $\frac{0.2}{10.0} = 0.02$, or $2\%$.

Systematic Errors vs. Random Errors

• Systematic Errors: Flaws in equipment or experimental design that bias results consistently in one direction (e.g., an uncalibrated balance reading $0.5\text{ g}$ high).
  
  
• Random Errors: Fluctuations in measurement caused by unpredictable environmental noise, observer variability, or precision limitations of instruments.
  
  

Unit conversions do not remove systematic or random errors; they simply express them in the scale and dimension of the new unit system.

2. Linear Unit Conversions (Exact Conversion Factors)

The most common unit conversions rely on exact, defined multipliers. Examples include converting meters to kilometers ($1000\text{ m} = 1\text{ km}$) or inches to centimeters ($1\text{ in} = 2.54\text{ cm}$).

Mathematical Principle

When a variable $x$ with absolute uncertainty $\Delta x$ is multiplied by an exact conversion factor $C$, the new variable $y$ and its absolute uncertainty $\Delta y$ are:

$$y = C \cdot x$$

$$\Delta y = \vert{}C\vert{} \cdot \Delta x$$

Because $C$ is an exact defined integer or mathematical constant, it carries zero uncertainty ($\Delta C = 0$). Consequently, the relative uncertainty remains completely invariant under linear exact unit conversions.

$$\frac{\Delta y}{\vert{}y\vert{}} = \frac{\vert{}C\vert{} \cdot \Delta x}{\vert{}C \cdot x\vert{}} = \frac{\Delta x}{\vert{}x\vert{}}$$

Step-by-Step Example: Length Conversion

Suppose a distance is measured as $x = (5.24 \pm 0.03)\text{ inches}$. Convert this measurement and its uncertainty into centimeters ($1\text{ in} = 2.54\text{ cm}$ exactly).

1. Calculate the central value:
   
   
   $$y = 5.24 \times 2.54 = 13.3096\text{ cm}$$
2. Calculate the absolute uncertainty:
   
   
   $$\Delta y = 0.03 \times 2.54 = 0.0762\text{ cm}$$
3. Format according to significant figures:
   Since the original uncertainty $0.03$ has one significant figure, round the new absolute uncertainty to $0.08\text{ cm}$ and match the precision of the converted value:
   
   
   $$y = (13.31 \pm 0.08)\text{ cm}$$
4. Verify relative uncertainty invariance:
   
   
   $$\text{Original Relative Uncertainty} = \frac{0.03}{5.24} \approx 0.5725\%$$
   $$\text{Converted Relative Uncertainty} = \frac{0.08}{13.31} \approx 0.6011\%\quad (\text{differs slightly due to rounding})$$
   $$\text{Exact Converted Relative Uncertainty} = \frac{0.0762}{13.3096} \approx 0.5725\%$$

3. Conversions Involving Inexact (Empirical) Conversion Factors

In certain domain-specific situations, conversion factors are determined empirically and are not exact mathematical definitions. Examples include standard density conversions or empirical rate factors.

The General Propagation of Uncertainty Formula

When a quantity $y$ depends on independent variables $u, v, w, \dots$, where each variable has an uncertainty $\sigma_u, \sigma_v, \sigma_w, \dots$, the standard variance formula (derived from a first-order Taylor series expansion) is:

$$\sigma_y^2 = \left( \frac{\partial y}{\partial u} \right)^2 \sigma_u^2 + \left( \frac{\partial y}{\partial v} \right)^2 \sigma_v^2 + \left( \frac{\partial y}{\partial w} \right)^2 \sigma_w^2 + \dots$$

When converting $x$ using an inexact factor $C \pm \Delta C$, where $y = C \cdot x$, both $C$ and $x$ contribute to the final error.

Using the propagation formula:

$$\Delta y = \sqrt{\left(\frac{\partial y}{\partial x}\right)^2 (\Delta x)^2 + \left(\frac{\partial y}{\partial C}\right)^2 (\Delta C)^2}$$

Taking the partial derivatives $\frac{\partial y}{\partial x} = C$ and $\frac{\partial y}{\partial C} = x$:

$$\Delta y = \sqrt{C^2 (\Delta x)^2 + x^2 (\Delta C)^2}$$

Dividing both sides by $y = C \cdot x$ gives the fractional uncertainty rule for multiplication:

$$\frac{\Delta y}{\vert{}y\vert{}} = \sqrt{\left(\frac{\Delta x}{x}\right)^2 + \left(\frac{\Delta C}{C}\right)^2}$$

Step-by-Step Example: Mass to Volume Conversion via Empirical Density

Consider converting a mass measurement $m = (250.0 \pm 0.5)\text{ g}$ into volume $V$ using an estimated liquid density $\rho = (1.05 \pm 0.02)\text{ g/cm}^3$.

1. Calculate the central value:
   
   
   $$V = \frac{m}{\rho} = \frac{250.0}{1.05} \approx 238.095\text{ cm}^3$$
2. Calculate relative uncertainties:
   
   
   $$\frac{\Delta m}{m} = \frac{0.5}{250.0} = 0.002\quad (0.2\%)$$
   $$\frac{\Delta \rho}{\rho} = \frac{0.02}{1.05} \approx 0.01905\quad (1.905\%)$$
3. Combine relative uncertainties in quadrature:
   
   
   $$\frac{\Delta V}{V} = \sqrt{(0.002)^2 + (0.01905)^2} = \sqrt{0.000004 + 0.0003629} \approx 0.01915\quad (1.915\%)$$
4. Calculate final absolute uncertainty:
   
   
   $$\Delta V = 238.095 \times 0.01915 \approx 4.56\text{ cm}^3$$
5. Final Result:
   
   
   $$V = (238 \pm 5)\text{ cm}^3$$

Notice how the larger relative uncertainty in the empirical conversion factor dominated the final error.

4. Non-Linear and Affine Unit Conversions

Not all unit conversions are purely multiplicative. Temperature conversions, logarithmic scales (such as decibels or pH), and non-linear physical scales require different treatment.

Affine Conversions: Temperature ($^\circ\text{F}$ to $^\circ\text{C}$ and $\text{K}$)

Affine conversions take the general form $y = C \cdot x + b$, where $b$ is a constant offset.

Example: Fahrenheit to Celsius

The conversion formula is:

$$T_C = \frac{5}{9}(T_F - 32)$$

Taking the partial derivative with respect to $T_F$:

$$\frac{d T_C}{d T_F} = \frac{5}{9}$$

Therefore, the absolute uncertainty converts without the offset:

$$\Delta T_C = \frac{5}{9} \Delta T_F$$

Crucial Insight: While absolute uncertainty scales linearly by $\frac{5}{9}$, relative uncertainty changes because of the offset $b = -32$.

Suppose $T_F = (98.6 \pm 0.9)^\circ\text{F}$:

• Absolute Uncertainty in Celsius:
  
  
  $$\Delta T_C = \frac{5}{9} \times 0.9 = 0.5^\circ\text{C}$$
• Central Value in Celsius:
  
  
  $$T_C = \frac{5}{9}(98.6 - 32) = 37.0^\circ\text{C}$$
• Relative Uncertainty Comparison:
  
  
  $$\text{Relative } T_F = \frac{0.9}{98.6} \approx 0.91\%$$
  $$\text{Relative } T_C = \frac{0.5}{37.0} \approx 1.35\%$$

The relative uncertainty changed significantly because temperature zero-points are arbitrary in non-absolute scales ($^\circ\text{F}$ and $^\circ\text{C}$). To compute physically meaningful relative temperature uncertainties, temperatures must first be converted to an absolute scale like Kelvin ($\text{K}$).

Non-Linear Conversions: Logarithmic and Power-Law Conversions

For non-linear function conversions $y = f(x)$, derivative-based propagation must be applied:

$$\Delta y \approx \left\vert{} \frac{df}{dx} \right\vert{} \Delta x$$

Example: Converting Hydrogen Ion Concentration $[H^+]$ to pH

$$pH = -\log_{10}[H^+] = -\frac{\ln[H^+]}{\ln(10)}$$

Differentiating with respect to $[H^+]$:

$$\frac{d(pH)}{d[H^+]} = -\frac{1}{[H^+] \ln(10)}$$

Applying absolute uncertainty propagation:

$$\Delta pH = \left\vert{} -\frac{1}{[H^+] \ln(10)} \right\vert{} \Delta [H^+] = \frac{1}{\ln(10)} \left( \frac{\Delta [H^+]}{[H^+]} \right)$$

This yields a useful metrological property: The absolute uncertainty in a logarithmic unit (like pH) is directly proportional to the relative uncertainty of the linear measurement ($[H^+]$).

5. Summary Matrix of Conversion Types

Conversion Type	Transformation Function y=f(x)	Absolute Uncertainty Propagation Δy	Relative Uncertainty Relationship
Linear (Exact Factor)	$y = C \cdot x$	$\Delta y = \Vert{}C\Vert{} \cdot \Delta x$	Unchanged ($\frac{\Delta y}{y} = \frac{\Delta x}{x}$)
Linear (Inexact Factor)	$y = C \cdot x$	$\Delta y = \Vert{}y\Vert{} \sqrt{\left(\frac{\Delta x}{x}\right)^2 + \left(\frac{\Delta C}{C}\right)^2}$	Increases ($\frac{\Delta y}{y} > \frac{\Delta x}{x}$)
Affine (Offset Scale)	$y = C \cdot x + b$	$\Delta y = \Vert{}C\Vert{} \cdot \Delta x$	Changes ($\frac{\Delta y}{y} \neq \frac{\Delta x}{x}$)
Power Law	$y = A \cdot x^n$	$\Delta y = \Vert{}n \cdot A \cdot x^{n-1}\Vert{} \cdot \Delta x$	Scales by power ($\frac{\Delta y}{y} = \Vert{}n\Vert{} \frac{\Delta x}{x}$)
Logarithmic	$y = K \cdot \ln(x)$	$\Delta y = \Vert{}K\Vert{} \cdot \frac{\Delta x}{x}$	Absolute $\Delta y$ proportional to relative $\frac{\Delta x}{x}$

6. Significant Figures and Rounding Guidelines

A common practical challenge when performing unit conversions is managing decimal places and significant figures to avoid artificially adding or losing precision.

Key Rules for Formatting Converted Uncertainty

1. Uncertainty Precision: In standard scientific practice, reported absolute uncertainties should generally be rounded to one or two significant figures (e.g., $0.08\text{ cm}$ or $0.076\text{ cm}$, depending on institutional standards).
   
   
2. Matching Value Precision: The central value must be rounded so its last reported decimal place matches the position of the last significant digit of the absolute uncertainty.
   
   
3. Avoid Intermediate Rounding: Always retain full double-precision floating-point numbers in intermediate calculation steps, applying rounding rules only to final published results.
   
   

7. Real-World Applications and Best Practices

Practical Engineering Example: Compound Units

Consider converting a velocity measurement from imperial to SI units:

$$v = (65.0 \pm 1.5)\text{ miles per hour (mph)}$$

Convert $v$ to meters per second ($\text{m/s}$):

• $1\text{ mile} = 1609.344\text{ meters}$ (exact definition)
  
  
• $1\text{ hour} = 3600\text{ seconds}$ (exact definition)
  
  

The combined conversion factor $C$ is:

$$C = \frac{1609.344}{3600} \approx 0.44704\quad \text{(exact ratio)}$$

1. Convert the Central Value:
   
   
   $$v_{\text{SI}} = 65.0 \times 0.44704 = 29.0576\text{ m/s}$$
2. Convert the Absolute Uncertainty:
   
   
   $$\Delta v_{\text{SI}} = 1.5 \times 0.44704 = 0.67056\text{ m/s}$$
3. Format the Result:
   Rounding uncertainty to one significant figure gives $\Delta v_{\text{SI}} \approx 0.7\text{ m/s}$. Match the central value precision accordingly:
   
   
   $$v = (29.1 \pm 0.7)\text{ m/s}$$

Summary of Best Practices

• Identify Factor Uncertainty: Always verify whether conversion factors are exact defined standards or empirical approximations.
  
  
• Keep Absolute vs. Relative Rules Clear: Remember that linear exact conversions preserve relative uncertainty, while non-linear or offset conversions alter relative uncertainty.
  
  
• Use Partial Derivatives for Complex Formulas: Whenever conversions involve non-linear mathematical operations, use the standard differential formula $\Delta y = \vert{}\frac{df}{dx}\vert{} \Delta x$ rather than simple multiplicative scaling.

Digital Tools vs. Manual Calculation:

Significant Figures and Unit Conversion: