Upper Bound on Error Propagation using Triangle Inequality

In experimental physics and engineering, measurements are never perfectly exact. Every measured value carries a degree of uncertainty, usually expressed as experimental error. When these measured values are used to calculate another quantity, the errors in the original measurements propagate into the final result.

A common source of confusion for students preparing for competitive exams such as IIT JEE and NEET is this: why do the absolute errors of two quantities add even when the quantities themselves are subtracted?

This result is not a memorized rule. It follows from a fundamental mathematical result known as the Triangle Inequality. In this lesson, we prove the error propagation rule for subtraction using this inequality.

Notation and Definitions

Before proving the result, we first define the symbols used in the discussion.

Problem: Upper Bound on Error Propagation Using Triangle Inequality

In an experiment, a student measures two independent quantities ##x## and ##y## with maximum absolute errors ##\Delta x## and ##\Delta y## respectively. A derived quantity is defined by the subtraction formula:

###Z = x - y###

Show that the maximum absolute error in ##Z## satisfies:

###\Delta Z \leq \Delta x + \Delta y###

For the usual worst-case error estimate used in school-level physics and competitive exams, we write:

###\Delta Z = \Delta x + \Delta y###

This result will be proved using the triangle inequality:

###|a+b| \leq |a| + |b|###

Worked Solution and Step-by-Step Explanation

To establish the result rigorously, we begin by writing the measured values of ##x## and ##y## in terms of their true values and their actual errors.

Let the measured values be:

###x_m = x + \delta x###

and

###y_m = y + \delta y###

Here, ##\delta x## and ##\delta y## are the actual algebraic errors. They may be positive, negative, or zero. The maximum absolute error limits are written as:

###|\delta x| \leq \Delta x###

and

###|\delta y| \leq \Delta y###

Step 1: Calculate the Measured Value of the Derived Quantity

The derived quantity is calculated from measured values as:

###Z_m = x_m - y_m###

Substituting ##x_m = x + \delta x## and ##y_m = y + \delta y##, we get:

###Z_m = (x + \delta x) - (y + \delta y)###

Expanding the brackets:

###Z_m = x + \delta x - y - \delta y###

Grouping the true-value terms and error terms separately:

###Z_m = (x-y) + (\delta x - \delta y)###

Since the true value of the derived quantity is ##Z = x-y##, we can write:

###Z_m = Z + (\delta x - \delta y)###

Step 2: Isolate the Propagated Error

The actual propagated error in ##Z## is defined as:

###\delta Z = Z_m - Z###

From the equation obtained above,

###Z_m = Z + (\delta x - \delta y)###

we immediately get:

###\delta Z = \delta x - \delta y###

This is the exact algebraic error in the calculated difference.

Step 3: Apply the Triangle Inequality

We are interested in the magnitude of the error, so we take the absolute value:

###|\delta Z| = |\delta x - \delta y|###

Rewrite subtraction as addition of a negative term:

###|\delta Z| = |\delta x + (-\delta y)|###

Now apply the triangle inequality:

###|a+b| \leq |a| + |b|###

Here, let:

###a = \delta x, \qquad b = -\delta y###

Therefore,

###|\delta x + (-\delta y)| \leq |\delta x| + |-\delta y|###

Since ##|-\delta y| = |\delta y|##, this becomes:

###|\delta Z| \leq |\delta x| + |\delta y|###

Step 4: Substitute the Maximum Absolute Error Limits

From the definitions of maximum absolute error:

###|\delta x| \leq \Delta x###

and

###|\delta y| \leq \Delta y###

Substituting these limits into the inequality

###|\delta Z| \leq |\delta x| + |\delta y|###

gives:

###|\delta Z| \leq \Delta x + \Delta y###

If ##\Delta Z## denotes the maximum possible absolute error in ##Z##, then the worst-case estimate is:

###\Delta Z = \Delta x + \Delta y###

Why Errors Add Even During Subtraction

At first, it may seem that if the physical quantities are being subtracted, their errors should also subtract. However, error propagation is not about one particular favorable case. It is about the maximum possible uncertainty.

Suppose ##x## is overestimated and ##y## is underestimated. Then, in the expression ##x-y##, the calculated value becomes larger than the true value from both sides. The error in ##x## increases the result, and the error in ##y## also increases the result because ##y## is being subtracted.

For example, if ##x## is too high by ##0.1\text{ cm}## and ##y## is too low by ##0.1\text{ cm}##, then the error in ##x-y## can become ##0.2\text{ cm}##. This is why the worst-case absolute errors add.

The triangle inequality guarantees that no matter what signs the individual errors have, the magnitude of the final error cannot exceed the sum of their absolute magnitudes.

Comparison of Error Propagation Rules

The way errors combine depends on the mathematical operation used to calculate the derived quantity. The table below summarizes the standard school-level rules used in physics.

Physical Significance and Worst-Case Analysis

In classical error propagation, the goal is usually to find the safest upper limit of uncertainty. This is why we use the worst-case approach. It does not assume that errors will conveniently cancel each other.

When measuring two independent quantities ##x## and ##y##, the actual errors ##\delta x## and ##\delta y## are often unknown. In the best-case scenario, the errors may cancel. But in the worst-case scenario, they reinforce one another.

For the subtraction ##Z = x-y##, the worst case occurs when the error in ##x## and the error in ##y## act in opposite algebraic directions. This makes the final difference deviate as much as possible from the true value.

Therefore, the maximum absolute error is taken as:

###\Delta Z = \Delta x + \Delta y###

Worst-Case Error vs Statistical Error

For school-level physics, board exams, IIT JEE, and NEET, the standard approach is the absolute worst-case method. In advanced laboratory work, where errors are random and statistically independent, another method called quadrature addition or root-sum-square addition is often used.

Final Conclusion

The rule that absolute errors add during subtraction is a direct consequence of the triangle inequality. For the derived quantity

###Z = x-y###

the propagated algebraic error is

###\delta Z = \delta x - \delta y###

Taking the absolute value and applying the triangle inequality gives:

###|\delta Z| \leq |\delta x| + |\delta y|###

Using ##|\delta x| \leq \Delta x## and ##|\delta y| \leq \Delta y##, we get:

###|\delta Z| \leq \Delta x + \Delta y###

Hence, the maximum absolute error is:

###\boxed{\Delta Z = \Delta x + \Delta y}###

This proves that in subtraction, just as in addition, absolute errors are added.