Physics · Article

Approximation methods and perturbative thinking

Build a useful estimate, compare a correction with a stated reference scale, and check what is lost when you leave out a term.

Previous article: Hypotheses, laws, principles, and theories

Back to Physics

After reading, you can

Explain what an approximation keeps and what it neglects.
Use powers of ten to describe scale while retaining units.
Distinguish an absolute change from a relative change.
Build a baseline plus a first correction using a square-area model.
Compare an omitted term with a stated tolerance and explain the model's limits.

Estimate with a purpose

An approximation replaces a value or calculation with something simpler for a stated purpose. You might round a length, use a simple shape for an irregular object, or neglect an effect that appears small. A useful estimate combines relevant information with a reasoned model; it is not a number chosen at random.

First decide what you need to know. Finding whether a stack is roughly a metre high needs less detail than deciding whether it fits beneath a shelf with a narrow clearance. Write ≈ for approximately equal and = for an equality within the stated model. Say which details you leave out.

Use powers of ten to identify the scale

Scientific notation separates a number's scale from its multiplier. For example, 400 m = 4 × 10^2 m and 4000 m = 4 × 10^3 m: 10^2 = 100 and 10^3 = 1000. The second length is ten times the first. A factor of ten corresponds to one order of magnitude; a factor of one hundred corresponds to two.

An order-of-magnitude estimate asks for a rough power-of-ten scale, rather than every digit. Here, saying hundreds of metres describes the scale of 400 m; it does not replace 400 m with an exact 100 m. State the unit and how rough the estimate is. A scale check can expose a large arithmetic or unit mistake.

Small compared with what?

An absolute change keeps its unit: adding 1 mm adds the same length to any object. A relative change divides that change by a stated nonzero reference. With an original length L and added length d, use r = d/L. Both lengths must use the same unit. Their units cancel, so r is dimensionless: a number with no unit.

Adding 1 mm to 2 mm gives r = 1/2 = 0.5, or 50%. Adding 1 mm to 1000 mm gives r = 1/1000 = 0.001, or 0.1%. To express a fraction as a percentage, multiply it by 100%. The same absolute change can have very different relative sizes. These are stipulated comparisons, not measurement results.

A small numerical value alone does not justify neglect. The reference scale, the predicted quantity and the required accuracy all matter. Changing from metres to millimetres changes the numbers but does not change the relative size of the same two lengths.

Start with a baseline and add a correction

Perturbative thinking starts with a simpler case and asks how a specified change modifies it. A parameter describes that change; a dimensionless ratio such as r can compare it with the baseline. Formal perturbation methods develop this approach further. This lesson uses one finite algebraic example, without requiring calculus or an infinite series.

Take an ideal square with original side L greater than zero. Increase each side by the same length d. Its new area is (L + d)^2, where squaring means multiplying the quantity by itself. Multiply the two factors: (L + d)(L + d) = L^2 + Ld + dL + d^2 = L^2 + 2Ld + d^2. This identity is exact for the square model. Every term has units of area, such as m².

The baseline area is L². Keeping the correction 2Ld gives A_first = L^2 + 2Ld, with d² omitted. It is called first order because the correction contains one power of d; the omitted term is second order because it contains d multiplied by itself. Dividing by L² gives the equivalent exact relationship A/L^2 = 1 + 2r + r^2.

Three levels for the stipulated square model
Calculation	Area expression	What it omits
Baseline	L²	2Ld and d²
First correction	L² + 2Ld	d²
Full square formula	L² + 2Ld + d²	No algebraic term from this square formula

Check the term you left out

For this square formula, the difference between the exact model area and A_first is exactly d². With L = 1.00 m and d = 0.02 m, the baseline is 1.0000 m², the first correction is 0.0400 m², and the omitted term is 0.0004 m². Thus A_first is 1.0400 m² and the exact model area is 1.0404 m².

Compare that difference with the criterion for the task. If the allowed absolute area difference is 0.001 m², the omitted 0.0004 m² is within it. If the allowed difference is 0.0001 m², it is too large. The same approximation can serve one task and fail another.

To report a relative approximation difference here, divide its magnitude by the stated reference, the exact model area: 0.0004/1.0404 ≈ 0.0003845, or about 0.03845%. This differs from the 2% change in side length. A relative input change and a relative output error are different comparisons.

All these dimensions are supplied exactly for an invented teaching case. The extra digits make the omitted term visible; they do not establish the precision of a real length or area measurement.

Keep the validity conditions attached

For the area model, the sides must remain equal and positive, so L is positive and L + d is positive. The first correction is useful for a requested absolute tolerance only when the omitted d² is within that tolerance. If the criterion is relative, calculate the corresponding relative difference. There is no universal percentage below which every effect may be ignored.

Keeping d² restores the exact square formula in this particular example. That does not show that adding terms to every approximation always improves it. Other problems need their own error checks; a small parameter alone does not establish validity over every location or time interval.

An algebraic check also depends on the physical model. If a real panel becomes a rectangle or curves, the square formula may no longer describe it. Check the neglected term inside the model, then check whether the model and input information suit the object. A careful approximation states both kinds of limitation.

A baseline, a correction and a check

1. Baseline: L²
Start with the original square area: 1.0000 m².
2. First correction: 2Ld
Add 0.0400 m² to obtain A_first = 1.0400 m².
3. Omitted term: d²
The omitted 0.0004 m² gives an exact square-model area of 1.0404 m² when restored.
4. Check the requested tolerance
A difference of 0.0004 m² meets a 0.001 m² allowance, but exceeds a 0.0001 m² allowance. The same approximation serves one criterion and fails another.

For the invented ideal square, L = 1.00 m and d = 0.02 m are supplied exactly. The extra digits display the omitted term; they do not establish real measurement precision.

Worked examples

Estimate a stack before adding detail

In an invented teaching case, 120 spacers are stacked with no gaps or compression. Use an estimated thickness of 8 mm per spacer. If each thickness is stipulated to lie between 7 mm and 9 mm, what estimate and bounds follow for the stack height?

Multiply the estimated thickness by the count: 120 × 8 mm = 960 mm.
Use 1000 mm = 1 m, so the estimate is 0.960 m, roughly 1 m.
The stipulated lower and upper totals are 120 × 7 mm = 840 mm and 120 × 9 mm = 1080 mm, or 0.840–1.080 m.

The estimate is roughly 1 m, with model bounds of 0.840–1.080 m under the supplied assumptions. This does not settle a tight shelf-clearance question. Real gaps, compression or different thickness information would require revisiting the model.

Compare an absolute change with two references

Two stipulated lengths are 2 mm and 1000 mm. Add 1 mm to each. Is the change equally small relative to their original lengths?

The absolute change is 1 mm in both cases.
For 2 mm, the relative change is 1 mm / 2 mm = 0.5 = 50%.
For 1000 mm, it is 1 mm / 1000 mm = 0.001 = 0.1%.

No. Relative to the original lengths, the changes are 50% and 0.1%. Whether either can be neglected still depends on what you are calculating and the accuracy required.

Check a first correction against an exact model result

An ideal square has supplied side L = 1.00 m. Increase each side by d = 0.02 m. Find the baseline area, first-correction area, exact square-model area and omitted term. Does the first-correction result meet an allowed absolute difference of 0.001 m²?

The baseline is L^2 = 1.0000 m².
The first correction is 2Ld = 2 × 1.00 m × 0.02 m = 0.0400 m², giving 1.0400 m².
The omitted term is d^2 = (0.02 m)^2 = 0.0004 m². Add it to obtain the exact model area 1.0404 m².
Compare 0.0004 m² with the stated 0.001 m² tolerance.

Yes: the difference from the exact square-model result is 0.0004 m², within 0.001 m². This checks algebraic truncation for the supplied square; it does not establish the accuracy of a real panel measurement.

Try it yourself

Supplied lengths are 400 m and 4000 m. What is their ratio, and how many orders of magnitude separate them?
Show the answer and reasoning
The second is ten times the first, a difference of one order of magnitude.
Divide quantities in the same unit: 4000 m / 400 m = 10. Their scientific-notation powers are 10² and 10³. The multiplier 4 is the same.
A supplied length increases from 100 mm to 101 mm. What is its relative change using the original length? Does that percentage alone show that the change may be ignored?
Show the answer and reasoning
1%. No; the task and required accuracy are still needed.
The added length is 1 mm, so 1 mm / 100 mm = 0.01 = 1%. This describes the input change, not an error bound for every quantity computed from it.
For an ideal square, let L = 1.00 m and d = 0.50 m. Find A_first and the exact area. Does A_first meet an allowed absolute difference of 0.01 m²?
Show the answer and reasoning
A_first = 2.00 m²; exact area = 2.25 m². No: the difference is 0.25 m².
The baseline is 1.00 m² and 2Ld = 1.00 m². The omitted d² = 0.25 m² is greater than the stated 0.01 m² tolerance. A calculation called first order is not automatically adequate.
For the worked small-square case, d/L is 2%, while the first-correction area differs from the exact model area by about 0.03845%. Why are these percentages different, and would retaining d² prove that a real panel's area is known exactly?
Show the answer and reasoning
They compare different quantities with different references. Retaining d² gives the exact result only within the supplied square model.
The 2% ratio compares added side length with original side length. The 0.03845% ratio compares omitted area with exact model area. Real dimensions, shape and measurement limitations still require evaluation.

What to carry forward

Choose the purpose and required accuracy before simplifying.
Use powers of ten for scale estimates and keep the units.
Judge a change relative to an explicit reference, not its number alone.
Build a baseline plus a correction and identify the term omitted.
Check that term against the task's criterion and keep the physical model's limits visible.

References

Sources for the concepts and teaching approaches in this article.