Estimation and Approximation

Choose the purpose before the precision

Estimating the cost of a small purchase may require only a nearby whole-dollar total. Designing a component requires a specified tolerance. Estimating the number of packages needed usually requires enough capacity, so an underestimate may be unacceptable.

State what the estimate is meant to establish. “About 50” might be sufficient to reject a calculator result of 5000, but it is not sufficient to distinguish 49.8 from 50.2.

Use $=$ for equality and $\approx$ for approximation. If an estimate is also a verified lower or upper bound, use the corresponding inequality. These symbols communicate different levels of information.

Round to convenient numbers

For $198\times49$, nearby values give

The exact result is 9702. The estimate is high by 298, about 3.07% of the exact value. That may be adequate for a magnitude check but not for a payment that must be exact.

Here both positive factors were rounded upward, so the estimate is an upper bound. If one factor is rounded upward and the other downward, their effects can offset or reinforce in ways that require checking.

For a quotient, compatible numbers are often more useful than mechanically rounding every number to one significant figure:

The exact value is 49.75. Keeping the denominator 8 made the mental calculation easy and retained useful information.

Compensation can recover an exact result

A nearby calculation can also be the starting point for exact arithmetic. Write the changes explicitly:

The final $+2$ restores the product of the two subtracted adjustments. Omitting it would count that small portion incorrectly.

Compensation is not guessing. It uses the distributive property to turn an awkward multiplication into a convenient one and then correct it. It connects estimation to simplifying algebraic expressions.

Use upper and lower bounds

Suppose positive length readings round to 20 cm and 10 cm, each to the nearest centimetre using ties away from zero. Their pre-rounded values satisfy

Because both quantities are positive, multiplying the lower endpoints gives a lower area bound, and multiplying the upper endpoints gives an excluded upper bound:

The area calculated from the rounded readings is $20\times10=200\text{ cm}^2$. Its possible difference from the pre-rounded product is not simply half a square centimetre. Multiplication changes how the original rounding intervals combine.

These are rounding bounds on the input readings. They are not a complete uncertainty statement about the physical rectangle. For actual measurement uncertainty, see Error, Accuracy and Precision.

Division needs additional care

If $0<c\le y\le d$ and $0<a\le x\le b$, then

The smallest quotient uses the smaller numerator and larger denominator. This simple formula relies on positive bounds and a denominator bounded away from zero. An interval containing zero in the denominator needs a different analysis.

Locate irrational values between known values

Since $7^2=49$ and $8^2=64$, the positive square root of 50 lies between 7 and 8. A closer calculation gives

Therefore $7.07<\sqrt{50}<7.08$. Squaring is increasing for non-negative inputs, which justifies moving from the squared comparison to the square-root comparison.

This bracket shows that the error in using 7.07 is less than 0.01. It does not, by itself, establish every later decimal digit. To round to a particular place, compare against the relevant midpoint or obtain a sufficiently narrow interval.

Rounding is not always the right practical action

Suppose 1.8 L must be placed into containers that each hold at most 0.25 L. The quotient is

Seven containers are insufficient. The practical requirement needs eight containers, assuming they can be filled up to that stated capacity and no additional allowance is required. Rounding 7.2 to the nearest whole number would answer the wrong question.

By contrast, if a budget permits 7.2 items at a fixed price and only whole items may be purchased, the maximum is seven. The context decides whether to round up, round down, or report a nearest approximation.

Avoid losing small differences

Subtracting nearby rounded values can erase the difference you are trying to measure. For example,

If both readings are rounded to one decimal place first, the calculation becomes $10.0-10.0=0.0$. The original small difference disappears. Retain sufficient intermediate digits and decide what the measurements justify before interpreting that difference.

Double rounding can also change a reported answer. Under the tie-away rule, 1.449 rounded directly to one decimal place is 1.4, but rounding through 1.45 can give 1.5. Significant Figures explains the role of guard digits.

Build a transparent rough model

A rough quantity estimate should expose its assumptions. Suppose a hypothetical hall appears to have about 24 rows with about 18 seats in each row. The model gives $24\times18=432$ seats. A quicker benchmark, $20\times20=400$, is in the same range.

Neither number is an observed count. Aisles, partial rows or different seat arrangements could change the total. Writing the factors makes it possible to improve the estimate by checking the uncertain assumption rather than merely adding digits to the answer.

Use this method for everyday quantities such as storage capacity or travel time, but do not treat an informal estimate as a safety limit or a verified capacity.

Check scale, units and error direction

Before accepting a result, compare it with a benchmark and check its units. A length converted from metres to centimetres should have a larger numerical value; an area requires a squared conversion factor. A positive fraction of a quantity should not exceed the whole unless the fraction exceeds one.

Then consider whether the estimate is above or below the target and whether that direction matters. A bound, a rounded value and an informal guess are different tools. Reporting which tool was used makes the answer more useful than a bare approximate number.

Related topics

Dimensional Analysis checks the structure of an applied calculation. Error, Accuracy and Precision distinguishes rounding error from other limits on measurement quality.