Differentials & Error Estimation — Differentials and error estimation (multivariable) - Multivariable Calculus

SECTION 1

Historical Context & Motivation

Imagine you're building a metal box and each measurement—length, width, and height—is slightly off. How much does the final volume change because of those small errors? This is exactly the kind of question that drove mathematicians to develop differentials for functions of several variables. Long before computers could simulate error propagation, scientists needed a quick, reliable pencil-and-paper method to estimate how measurement uncertainties ripple through formulas.

The story of differentials begins with the invention of calculus itself. As physics and engineering grew more complex, so did the need to handle functions that depend on two, three, or even more variables simultaneously. The idea of a total differential emerged naturally: extend the single-variable differential to account for changes in every input at once.

1680s • Leibniz Introduces Differentials

Gottfried Wilhelm Leibniz develops the notation dy = f′(x) dx, giving calculus a powerful symbolic language for infinitesimal changes in a single variable.

1740s • Euler Extends to Several Variables

Leonhard Euler formalizes partial derivatives, allowing mathematicians to measure how a function changes with respect to one variable while holding others constant.

1800s • Total Differential in Physics

Thermodynamics and fluid dynamics require functions of pressure, temperature, and volume. Scientists routinely use total differentials to estimate how small changes in conditions affect energy and entropy.

1900s • Error Propagation Becomes Standard

Experimental scientists adopt differentials as the backbone of error analysis. The technique becomes a core topic in physics, chemistry, and engineering lab courses worldwide.

The central question this lesson addresses is: If each input variable has a small error, how large is the resulting error in the output? Differentials give us a clean, linear approximation that answers this question without having to recompute the entire function from scratch.

SECTION 2

Core Principles & Definitions

Before diving into calculations, you need to understand the foundational ideas behind multivariable differentials. In single-variable calculus, you learned that dy = f′(x) dx approximates the change in y when x changes by a tiny amount dx. The multivariable version follows the same logic but adds up contributions from every variable.

Partial Derivative

The rate of change of a function with respect to one variable while all other variables are held constant. Written as ∂f/∂x or ∂f/∂y.

Total Differential (df)

The sum of each partial derivative multiplied by the corresponding small change: df = (∂f/∂x) dx + (∂f/∂y) dy. It captures how all variables contribute to the overall change.

Absolute Error (Δ vs. d)

Δf is the true change in f; df is the linear approximation. When the changes dx and dy are small, df ≈ Δf. The differential df serves as an estimate of the actual error.

Relative & Percentage Error

Relative error is |df/f|, and percentage error is 100 × |df/f|%. These measures tell you how significant the error is compared to the value itself, which is often more meaningful than the raw number.

✦

KEY TAKEAWAY

Think of the total differential like adjusting a recipe that uses multiple ingredients. If you add a little too much flour and a little too much sugar, the overall taste change depends on how sensitive the recipe is to each ingredient. The total differential adds up each ingredient's individual effect, weighted by how much the recipe 'cares' about that ingredient (the partial derivative).

SECTION 3

Visual Explanation — The Tangent Plane Approximation

In single-variable calculus, the differential corresponds to moving along the tangent line instead of along the actual curve. In multivariable calculus, the differential corresponds to moving along the tangent plane instead of along the actual surface. The diagram below shows a surface z = f(x, y) with a tangent plane at a point. The true change Δf follows the surface, while the differential df follows the flat tangent plane.

The pink dot marks the point P where the tangent plane (green, dashed) touches the surface z = f(x, y) (purple curve). The gold segment labeled df shows the change predicted by the differential (along the plane), while the red dashed segment Δf shows the actual change along the curved surface. When dx and dy are small, these two are nearly equal.

The key geometric insight is that the tangent plane is the best linear approximation to the surface near the point of tangency. Just as a tangent line in 2-D hugs a curve locally, a tangent plane in 3-D hugs a surface. The total differential df captures the height change on this flat plane, making it a fast and easy estimate of the true change Δf whenever the inputs shift by small amounts.

SECTION 4

Mathematical Framework

Let's build the formulas step by step. Suppose you have a function f that depends on two variables, x and y. When both variables change by small amounts dx and dy, the total differential gives us the approximate change in f.

TOTAL DIFFERENTIAL (TWO VARIABLES)

df = (∂f/∂x) dx + (∂f/∂y) dy

∂f/∂x = partial derivative of f with respect to x (y held constant); ∂f/∂y = partial derivative of f with respect to y (x held constant); dx, dy = small changes in x and y.

This formula extends naturally to three or more variables. If f depends on x, y, and z, you simply add another term.

TOTAL DIFFERENTIAL (THREE VARIABLES)

df = (∂f/∂x) dx + (∂f/∂y) dy + (∂f/∂z) dz

Each term accounts for the contribution of one variable's change to the overall change in f. The pattern continues: for n variables, there are n terms.

For error estimation, we interpret dx, dy, and dz as the maximum possible measurement errors (often written as Δx, Δy, Δz). Because errors can add up in the worst case, we use absolute values to get the maximum absolute error:

MAXIMUM ABSOLUTE ERROR

|df| ≤ |∂f/∂x| |dx| + |∂f/∂y| |dy| + |∂f/∂z| |dz|

Using absolute values ensures we capture the worst-case scenario, where all errors push the result in the same direction.

RELATIVE ERROR

df/f = (∂f/∂x)(dx/f) + (∂f/∂y)(dy/f) + …

Divide the total differential by f to express the error as a fraction of the function value. Multiply by 100 for percentage error.

SECTION 5

Detailed Breakdown — How Errors Combine

Not all variables contribute equally to the total error. The partial derivative acts like a sensitivity multiplier: a large partial derivative means the function is very sensitive to that variable, so even a tiny measurement error in that variable will cause a big change in f. The diagram below visualizes how two error contributions combine.

For the function f = x²y evaluated at (3, 4), the x-variable contributes 64% of the total error while y contributes 36%. The bar chart makes it clear that the variable with the larger sensitivity (larger partial derivative times its error) dominates the total.

💡

Practical Tip

When trying to reduce total error in a real experiment, focus your effort on improving the measurement of the variable with the largest contribution to df. Buying a more accurate ruler for the less sensitive variable won't help much!

Variable	Partial Derivative	Error (dx or dy)	Contribution to df
x	∂f/∂x = 2xy = 24	dx = 0.02	0.48
y	∂f/∂y = x² = 9	dy = 0.03	0.27
Total	—	—	df = 0.75

Error breakdown for f = x²y at (3, 4)

SECTION 6

Worked Example — Volume of a Cylinder

A cylindrical tank has a measured radius of r = 5 cm with a possible error of dr = ±0.1 cm, and a measured height of h = 12 cm with a possible error of dh = ±0.2 cm. Use differentials to estimate the maximum error and the percentage error in the calculated volume V = πr²h.

Estimating Volume Error

1
Step 1 — Write Down the Formula

The volume of a cylinder is V = πr²h. This is a function of two variables: r and h.
2
Step 2 — Compute Partial Derivatives

∂V/∂r = 2πrh. This tells us how sensitive V is to changes in the radius. ∂V/∂h = πr². This tells us how sensitive V is to changes in the height.

∂V/∂r = 2π(5)(12) = 120π ≈ 376.99; ∂V/∂h = π(25) = 25π ≈ 78.54
3
Step 3 — Write the Total Differential

dV = (∂V/∂r) dr + (∂V/∂h) dh = 2πrh · dr + πr² · dh.
4
Step 4 — Substitute Known Values

Plug in r = 5, h = 12, dr = 0.1, dh = 0.2: dV = 120π(0.1) + 25π(0.2) = 12π + 5π = 17π.

dV = 17π ≈ 53.41 cm³
5
Step 5 — Compute the Percentage Error

The calculated volume is V = π(25)(12) = 300π ≈ 942.48 cm³. The relative error is |dV/V| = 17π / 300π = 17/300 ≈ 0.0567. Multiply by 100 to get the percentage error.

Percentage error ≈ 5.67%

ℹ

Check Your Intuition

Notice that the radius error (dr = 0.1) contributes 12π ≈ 37.70 cm³ while the height error (dh = 0.2) contributes only 5π ≈ 15.71 cm³, even though dh is twice as large as dr. Why? Because V depends on r², so V is more sensitive to radius changes. This is the power of partial derivatives—they reveal which measurements matter most.

SECTION 7

Strengths & Limitations of the Differential Approximation

Aspect	Strength	Limitation
Computation Speed	Only requires partial derivatives evaluated at one point—no need to recompute the entire function.	Only valid when dx, dy, dz are small relative to x, y, z.
Accuracy	Excellent for small errors; the linear approximation is very close to the true change.	Ignores higher-order terms (curvature), so it can underestimate or overestimate for larger errors.
Error Type	Gives a quick maximum bound for the propagated error.	Assumes worst-case (all errors in the same direction). Real errors may partially cancel.
Applicability	Works for any differentiable function, regardless of the number of variables.	The function must be differentiable at the point of interest; doesn't work at sharp corners or discontinuities.

Strengths vs. Limitations of Differential Error Estimation

✦

KEY TAKEAWAY

The differential approximation is like using a flat map of a small neighborhood instead of a globe. For a few city blocks, the flat map is practically perfect. But if you try to use it for an entire continent, the curvature of the Earth makes the map wildly inaccurate. Similarly, df approximates Δf beautifully for small errors but breaks down when errors are large.

SECTION 8

Connection to Advanced Theory

The total differential is actually the first term in a larger expansion called the multivariable Taylor series. As you progress in mathematics, you'll see that adding second-order and higher-order terms gives increasingly accurate approximations. In experimental science, a more refined version called propagation of uncertainty uses the sum of squares of the individual error terms (root-sum-square method), which accounts for the fact that random errors tend to partially cancel rather than all stack in the worst direction.

Feature	Differential (This Lesson)	Advanced Methods
Order of Approximation	First-order (linear)	Second-order or higher (Taylor expansion)
Error Combination	Sum of absolute values (worst case)	Root-sum-square (statistical, more realistic)
When to Use	Quick estimate; small errors; exam problems	Lab reports; engineering tolerances; research
Prerequisites	Partial derivatives	Statistics, second-order partial derivatives

Current method vs. advanced error analysis

Understanding the total differential now gives you a strong foundation for these advanced topics. The concepts of sensitivity analysis and linear approximation appear throughout engineering, economics, machine learning, and data science—any field where you need to predict how outputs respond to input changes.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

Explain in your own words why the total differential df uses partial derivatives rather than ordinary derivatives. Why can't you just take one derivative and multiply by the total change?

Show answer

When a function depends on multiple variables, each variable contributes independently to the overall change. An ordinary derivative only captures how the function changes with respect to a single variable. Partial derivatives isolate the effect of each variable one at a time, and the total differential adds these individual effects together. This additive structure reflects the fact that each variable can change by a different amount, so you need a separate 'sensitivity' factor (partial derivative) for each one.

Question 2: BASIC CALCULATION

Let f(x, y) = 3x + 2y². Compute the total differential df and evaluate it at (1, 2) with dx = 0.05 and dy = −0.03.

Show answer

First, find the partial derivatives: ∂f/∂x = 3 and ∂f/∂y = 4y. The total differential is df = 3 dx + 4y dy. At (1, 2) with dx = 0.05 and dy = −0.03: df = 3(0.05) + 4(2)(−0.03) = 0.15 − 0.24 = −0.09. This means the function decreases by approximately 0.09 due to the combined changes in x and y.

Question 3: INTERMEDIATE

The area of a triangle is A = ½ · b · h. The base is measured as b = 10 cm with error db = ±0.3 cm, and the height is h = 6 cm with error dh = ±0.2 cm. Find the maximum absolute error and the maximum percentage error in A.

Show answer

Compute ∂A/∂b = h/2 = 3 and ∂A/∂h = b/2 = 5. The total differential is dA = 3(0.3) + 5(0.2) = 0.9 + 1.0 = 1.9 cm². The calculated area is A = ½(10)(6) = 30 cm². The maximum percentage error is (1.9/30) × 100 ≈ 6.33%. Notice that even though the base has a larger absolute error (0.3 vs. 0.2), the height error contributes more to the total because ∂A/∂h = 5 is larger than ∂A/∂b = 3.

Question 4: APPLIED

An electrical engineer uses the formula for power P = V²/R, where voltage V = 120 V with error dV = ±2 V, and resistance R = 30 Ω with error dR = ±0.5 Ω. Estimate the maximum absolute error and the percentage error in the calculated power.

Show answer

First, compute the partial derivatives: ∂P/∂V = 2V/R = 2(120)/30 = 8, and ∂P/∂R = −V²/R² = −(14400)/900 = −16. The maximum absolute error is |dP| ≤ |8|(2) + |−16|(0.5) = 16 + 8 = 24 watts. The nominal power is P = 120²/30 = 480 W. The percentage error is (24/480) × 100 = 5%. The voltage contributes 16 W and the resistance contributes 8 W, so improving the voltmeter's precision would be more effective at reducing overall error.

Question 5: CRITICAL THINKING

For the function f(x, y, z) = xyz, show that the relative error df/f simplifies to dx/x + dy/y + dz/z. Then explain why this result means the percentage errors of individual variables add directly when the function is a product.

Show answer

The partial derivatives are ∂f/∂x = yz, ∂f/∂y = xz, and ∂f/∂z = xy. So df = yz·dx + xz·dy + xy·dz. Dividing both sides by f = xyz: df/f = (yz·dx)/(xyz) + (xz·dy)/(xyz) + (xy·dz)/(xyz) = dx/x + dy/y + dz/z. This shows that for a pure product of variables, the relative error of the output is simply the sum of the relative errors of the inputs. This is a remarkably clean result: if x has 1% error, y has 2% error, and z has 0.5% error, the maximum relative error in f is about 3.5%. This principle extends to any product-form function and is extremely useful in physics, where many formulas (like V = lwh or F = ma) are products.

SUMMARY

Lesson Summary

Summary

The total differential df = (∂f/∂x) dx + (∂f/∂y) dy + … provides a linear approximation of how a multivariable function changes when its inputs shift by small amounts. Each term pairs a partial derivative (the sensitivity) with a small change (the measurement error), and the sum of all terms gives the estimated total change. Geometrically, this corresponds to sliding along the tangent plane rather than the actual curved surface.

For error estimation, use absolute values of each term and add them to find the maximum absolute error. Divide by the function value for the relative (percentage) error. The variable whose term |∂f/∂variable| × |d(variable)| is largest dominates the total error, so improving that measurement gives the best payoff. This technique is valid when the errors are small and the function is differentiable—conditions that are met in most real-world measurement scenarios.