Chapter 1 · Units and Measurements

1.3 Errors in Measurement

No measurement is exact. Least count, systematic and random errors, the mean absolute error, and the rules that carry errors through sums, products and powers: $\dfrac{\Delta Z}{Z}=p\dfrac{\Delta A}{A}+q\dfrac{\Delta B}{B}+r\dfrac{\Delta C}{C}$.

Time one swing of a pendulum with a stopwatch five times and you will almost never get the same number twice: $2.03\ \text{s}$, $1.98\ \text{s}$, $2.01\ \text{s}$… Measure the thickness of a coin with a ruler and you can only say "a little under $2\ \text{mm}$"; a screw gauge says $1.85\ \text{mm}$. Every measured number carries an uncertainty, set by the instrument, the method and the person using it. A result such as $g=9.8\ \text{m/s}^2$ means little until we also say how far it might be wrong: $9.8\pm0.1$ and $9.8\pm2$ are very different claims. Engineers who machine engine parts, doctors reading a blood-pressure gauge and the scientists who fixed the values of the SI constants all live by this rule.

This section teaches you to measure the doubt. First, the smallest step an instrument can show — its least count — and the difference between accuracy and precision. Then the two families of error, systematic and random, and how repeated readings give a best value and a mean absolute error. Finally the rules that tell you how errors in measured quantities combine when you add, subtract, multiply, divide or raise them to powers — the single most frequently tested idea of this chapter in JEE Main. The significant-figure rules of Section 1.1 are a quick form of these rules; here you will see where they come from.

Least count, accuracy and precision

Every instrument has a smallest division it can resolve. A metre scale is marked in millimetres, so a length can be read to the nearest $1\ \text{mm}$ and no better; any reading is uncertain by about one such division. Better instruments push this limit down by clever mechanical magnification: the Vernier callipers slide a second small scale along the main one, and the screw gauge turns a fine screw whose head is divided into many parts.

📘 Definition 1.8 — Least count, accuracy and precision

The least count (LC) of an instrument is the smallest value of a quantity that it can measure — the change in the quantity that moves the reading by one smallest division. Any single reading has an uncertainty (the least count error) of about one least count.

Accuracy describes how close a measurement is to the true value. Precision describes how finely a quantity is resolved and how closely repeated measurements agree with one another. A measurement can be precise without being accurate (all readings agree, but all are off by the same amount) and vice versa.

Vernier callipers. If $n$ divisions of the Vernier scale (VSD) have the same length as $(n-1)$ divisions of the main scale (MSD), then $1\ \text{VSD}=\tfrac{n-1}{n}\ \text{MSD}$ and

$$\text{LC}=1\ \text{MSD}-1\ \text{VSD}=\frac{1\ \text{MSD}}{n}.$$

To read it: take the main-scale reading (MSR) just before the Vernier zero, find the Vernier division $k$ that lines up exactly with a main-scale mark, and add: reading $=\text{MSR}+k\times\text{LC}$. With $1\ \text{MSD}=1\ \text{mm}$ and $n=10$, $\text{LC}=0.1\ \text{mm}=0.01\ \text{cm}$.

Screw gauge. One full turn moves the spindle forward by the pitch $p$; the head carries $N$ circular-scale divisions, so

$$\text{LC}=\frac{\text{pitch}}{N}.$$

A gauge with pitch $0.5\ \text{mm}$ and $50$ divisions has $\text{LC}=0.01\ \text{mm}$. Reading $=$ main (linear) scale reading $+$ (circular-scale division on the reference line) $\times$ LC.

Zero error. If an instrument does not read zero when the jaws are closed, every reading is shifted by the same amount, the zero error $e_0$ (positive if the instrument reads too much, negative if it reads too little). It is removed by subtracting it with its sign:

$$\text{corrected reading}=\text{observed reading}-e_0 .$$
✏️ Example 1 — Reading Vernier callipers with a zero error
On a pair of Vernier callipers the main scale is marked in millimetres and $10$ Vernier divisions match $9$ main-scale divisions. With the jaws closed, the Vernier zero lies just to the right of the main-scale zero and the $4$th Vernier division coincides with a main-scale mark. A rod is then gripped: the main-scale reading is $2.3\ \text{cm}$ and the $7$th Vernier division coincides. Find the least count, the zero error and the length of the rod.
LC $1\ \text{VSD}=\tfrac{9}{10}\ \text{mm}$, so $\text{LC}=1-0.9=0.1\ \text{mm}=0.01\ \text{cm}$.
Zero error The Vernier zero is to the right of the main zero, so the callipers read too much: $e_0=+4\times0.01=+0.04\ \text{cm}$.
Reading Observed $=2.3+7\times0.01=2.37\ \text{cm}$. Corrected $=2.37-(+0.04)=2.33\ \text{cm}$.
Check The result has its last digit in the hundredths of a centimetre — exactly the least count. Writing $2.330\ \text{cm}$ would claim a precision the instrument does not have.

Vernier callipers have a zero error of $-0.03\ \text{cm}$ (with the jaws closed, the Vernier zero lies to the left of the main-scale zero). A rod gives an observed reading of $3.45\ \text{cm}$. Its corrected length is

Systematic and random errors; absolute, relative and percentage errors

The error in a measurement is the difference between the measured value and the true value. We never know the true value exactly, but we can understand where errors come from, remove the ones we can, and estimate the size of the rest.

📘 Definition 1.9 — Systematic and random errors

Systematic errors push every reading in the same direction (always too large or always too small) by roughly the same amount. Their sources are

  • instrumental — a zero error, a badly calibrated scale, a stopwatch that runs slow;
  • imperfect technique or method — ignoring air resistance, heat loss to the surroundings, or measuring a body temperature with the thermometer under the arm;
  • personal — for example, always reading a scale from slightly above (parallax), or always stopping a watch a little late.

They are reduced by better instruments and methods and by applying corrections (such as subtracting the zero error). Repeating the measurement does not reduce them.

Random errors vary irregularly in size and sign from one reading to the next — from small, unpredictable changes in the conditions (vibrations, temperature, the observer's reaction time). They are equally likely to be positive or negative, so they largely cancel when many readings are averaged.

Suppose a quantity is measured $n$ times, giving $a_1,a_2,\ldots,a_n$. With no systematic error, the best estimate of the true value is the arithmetic mean. The scatter of the readings about the mean tells us how large the random errors are.

📘 Definition 1.10 — Absolute, relative and percentage errors

Mean value (best estimate): $\displaystyle\bar a=\frac{a_1+a_2+\dots+a_n}{n}=\frac1n\sum_{i=1}^{n}a_i$.

Absolute error of the $i$-th reading: $\Delta a_i=|a_i-\bar a|$, with the same unit as $a$.

Mean absolute error: $\displaystyle\Delta\bar a=\frac1n\sum_{i=1}^{n}|\Delta a_i|$. The result is reported as $a=\bar a\pm\Delta\bar a$: the true value very probably lies between $\bar a-\Delta\bar a$ and $\bar a+\Delta\bar a$.

Relative (fractional) error: $\dfrac{\Delta\bar a}{\bar a}$ — a pure number. Percentage error: $\delta a=\dfrac{\Delta\bar a}{\bar a}\times100\%$.

Two practical rules. (i) The mean absolute error can never honestly be smaller than the least count: if your readings agree so well that $\Delta\bar a$ comes out below the LC, quote the LC as the error. For a single reading, the error is the least count itself. (ii) An error is itself only an estimate, so round it to one significant figure (two if it begins with 1, when that helps), then round the value to the same decimal place: $9.672\pm0.106$ becomes $9.7\pm0.1$.

The relative error is what tells you how good a measurement is. An error of $1\ \text{mm}$ is excellent on the length of a corridor ($10\ \text{m}$: $0.01\%$) and useless on the diameter of a wire ($0.5\ \text{mm}$: $200\%$).

✏️ Example 2 — Mean absolute error of a wire diameter
A screw gauge of least count $0.01\ \text{mm}$ gives five readings of the diameter of a wire, taken at different places along it: $1.42$, $1.45$, $1.39$, $1.44$ and $1.40\ \text{mm}$. Find the mean diameter, the mean absolute error, the relative error and the percentage error, and write the result properly.
Mean $\bar d=\dfrac{1.42+1.45+1.39+1.44+1.40}{5}=\dfrac{7.10}{5}=1.42\ \text{mm}$.
Deviations $|\Delta d_i|=0.00,\ 0.03,\ 0.03,\ 0.02,\ 0.02\ \text{mm}$; their sum is $0.10\ \text{mm}$, so $\Delta\bar d=\dfrac{0.10}{5}=0.02\ \text{mm}$ (larger than the LC, so it stands).
Relative $\dfrac{\Delta\bar d}{\bar d}=\dfrac{0.02}{1.42}=0.014$; percentage error $1.4\%$.
Result $d=(1.42\pm0.02)\ \text{mm}$, i.e. $1.42\ \text{mm}\pm1.4\%$.
Check Every reading lies within $\pm0.03\ \text{mm}$ of the mean, and the mean absolute error ($0.02$) is a typical deviation — smaller than the largest one, as it must be.

In Interactive 1 keep the scatter at $0.020\ \text{cm}$ and set readings to $20$. Now switch the error type to "random + zero error $+0.04$ cm". What happens to the mean $\bar a$ and to the mean absolute error $\Delta\bar a$?

With the first set of readings you will see mean $\bar a=2.546\ \text{cm}$ and $\Delta\bar a=0.019\ \text{cm}$ for "random only", then $\bar a=2.586\ \text{cm}$ with the same $\Delta\bar a=0.019\ \text{cm}$ when the zero error is switched on (press "New readings" and the numbers change, but the shift is always exactly $+0.040\ \text{cm}$). A systematic error moves every reading by the same amount, so the whole cloud slides sideways: the mean moves, the scatter about the mean does not, and no number of readings removes it — only subtracting the zero error does.
Interactive 1 — Repeated Readings: Random Scatter and a Hidden Offset

A rod whose true length is $2.540\ \text{cm}$ (green dashed line) is measured again and again with Vernier callipers of least count $0.01\ \text{cm}$. Each dot is one reading. The red line is the running mean $\bar a$ and the purple band is $\bar a\pm\Delta\bar a$. Press Play to take readings one at a time. Then switch on the uncorrected zero error: the scatter looks just as tidy, but the whole cloud — and the mean — moves away from the true value, however many readings you take.

0.020 cm5
readings $n$—
mean $\bar a$—
mean abs. error $\Delta\bar a$—
percentage error—

Three readings of a length with a metre scale (least count $0.1\ \text{cm}$) all come out as $25.4\ \text{cm}$. The result should be written as

Combination of errors

Most physical quantities are not measured directly but calculated: a density from a mass and a volume, $g$ from a length and a period, a resistance from a voltage and a current. If each input has an error, how large is the error in the answer? JEE Main uses the maximum (worst-case) error: we assume all the individual errors push the result the same way, so they add. Throughout, $\Delta A$, $\Delta B$, … are small compared with $A$, $B$, …

📐 Rule 1.9 — Error in a sum or a difference

If $A\pm\Delta A$ and $B\pm\Delta B$ are measured and $Z=A+B$ or $Z=A-B$, the maximum absolute error in $Z$ is

$$\Delta Z=\Delta A+\Delta B .$$

Absolute errors add — for a difference as well as for a sum.

Why. For a sum, $Z\pm\Delta Z=(A\pm\Delta A)+(B\pm\Delta B)$; the largest possible deviation is when both errors have the same sign, $\pm(\Delta A+\Delta B)$. For a difference, the extreme values of $A-B$ are $(A+\Delta A)-(B-\Delta B)$ and $(A-\Delta A)-(B+\Delta B)$, i.e. $(A-B)\pm(\Delta A+\Delta B)$. The errors do not cancel just because the quantities are subtracted.

The consequence is important: the difference of two nearly equal quantities has a small value but the full sum of the absolute errors, so its percentage error can be very large (Example 3). Good experiments avoid relying on small differences when they can.

✏️ Example 3 — The extension of a spring
A metre scale (LC $0.1\ \text{cm}$) shows the length of a spring as $58.9\ \text{cm}$ unloaded and $62.4\ \text{cm}$ with a weight hung on it. Find the extension with its error, and compare its percentage error with that of each length.
Value $x=62.4-58.9=3.5\ \text{cm}$.
Error Each reading is uncertain by one LC, $0.1\ \text{cm}$. Rule 1.9: $\Delta x=0.1+0.1=0.2\ \text{cm}$, so $x=(3.5\pm0.2)\ \text{cm}$.
Compare Lengths: $\dfrac{0.1}{62.4}\times100\%=0.16\%$ and $\dfrac{0.1}{58.9}\times100\%=0.17\%$. Extension: $\dfrac{0.2}{3.5}\times100\%=5.7\%$ — about $35$ times worse.
Lesson To measure a small extension precisely, use a finer instrument for the change itself (a Vernier or a travelling microscope), or make the extension larger by hanging a bigger load.

For products and quotients we need one small piece of algebra, the binomial approximation: for a small number $x$ (say $|x|\lt0.05$) and any power $n$,

$$(1+x)^n\approx1+nx .$$

For example, $(1.02)^3=1.0612\approx1.06$ and $\sqrt{1.04}=1.0198\approx1.02$. The neglected terms contain $x^2$, which is tiny when $x$ is a few per cent.

📐 Rule 1.10 — Error in a product or a quotient

If $Z=AB$ or $Z=\dfrac{A}{B}$, the maximum relative error in $Z$ is

$$\frac{\Delta Z}{Z}=\frac{\Delta A}{A}+\frac{\Delta B}{B} .$$

Relative (or percentage) errors add — for a quotient as well as for a product. Valid when the relative errors are small (a few per cent), so that their products can be neglected.

Why. For a product,

$$Z\pm\Delta Z=(A\pm\Delta A)(B\pm\Delta B)$$ $$=AB\pm B\,\Delta A\pm A\,\Delta B\pm\Delta A\,\Delta B .$$

Divide by $Z=AB$:

$$1\pm\frac{\Delta Z}{Z}=1\pm\frac{\Delta A}{A}\pm\frac{\Delta B}{B}\pm\frac{\Delta A}{A}\cdot\frac{\Delta B}{B}.$$

The last term is a product of two small numbers (e.g. $0.02\times0.03=0.0006$) and is dropped; the worst case gives the rule. For a quotient, using the binomial approximation with $n=-1$,

$$\frac{A\pm\Delta A}{B\pm\Delta B}=\frac AB\left(1\pm\frac{\Delta A}{A}\right)\left(1\pm\frac{\Delta B}{B}\right)^{-1}$$ $$\approx\frac AB\left(1\pm\frac{\Delta A}{A}\right)\left(1\mp\frac{\Delta B}{B}\right),$$

and the worst case again adds the two relative errors.

📐 Rule 1.11 — Error in a quantity raised to powers

If $Z=k\,\dfrac{A^pB^q}{C^r}$, where $k$ is an exact constant and $p,q,r$ are positive numbers (whole or fractional), then

$$\frac{\Delta Z}{Z}=p\,\frac{\Delta A}{A}+q\,\frac{\Delta B}{B}+r\,\frac{\Delta C}{C}.$$

Each relative error is multiplied by the magnitude of its power and all the terms are added, whether the quantity is in the numerator or the denominator. Exact constants ($2$, $\pi$, a counted number of oscillations) contribute no error. Valid for small relative errors.

Why. $(A\pm\Delta A)^p=A^p\left(1\pm\dfrac{\Delta A}{A}\right)^p\approx A^p\left(1\pm p\dfrac{\Delta A}{A}\right)$ by the binomial approximation, and the same for $B^q$ and $C^{-r}$; then Rule 1.10 adds the relative errors. A square root halves the relative error ($p=\tfrac12$); a cube triples it. So the quantity with the highest power usually dominates the error, and it is the one to measure most carefully.

🧭 Problem-solving strategy — Finding the error in a calculated quantity
  1. Write the formula in the form $Z=k\,A^pB^qC^{-r}\cdots$. Pull out exact constants; they carry no error.
  2. Check the structure. If the formula contains a sum or difference of measured quantities (such as $l_2-l_1$), first find that combination's absolute error with Rule 1.9 and treat it as one quantity. If the same quantity appears in more than one place in a non-power way (such as $\tfrac{R_1R_2}{R_1+R_2}$), rewrite the formula first (Example 6).
  3. Find each relative error $\dfrac{\Delta A}{A}$ (from the LC or the mean absolute error), multiply by the magnitude of its power, and add — never subtract.
  4. Convert back: $\Delta Z=Z\times\dfrac{\Delta Z}{Z}$; round $\Delta Z$ to one significant figure and $Z$ to the same decimal place.

Calculus shortcut (after Section 2.1): take logarithms, $\ln Z=\ln k+p\ln A+q\ln B-r\ln C$, differentiate, $\dfrac{dZ}{Z}=p\dfrac{dA}{A}+q\dfrac{dB}{B}-r\dfrac{dC}{C}$, and for the maximum error replace every minus sign by a plus.

✏️ Example 4 — Density of a metal cube
The side of a metal cube is measured with Vernier callipers (LC $0.01\ \text{cm}$) as $2.50\ \text{cm}$, and its mass on a balance (LC $0.1\ \text{g}$) as $125.0\ \text{g}$. Find the density with its error, in $\text{g/cm}^3$ and in $\text{kg/m}^3$.
Find the maximum percentage error in the density before you open the solution:
Value $\rho=\dfrac{m}{a^3}=\dfrac{125.0}{(2.50)^3}=\dfrac{125.0}{15.625}=8.00\ \text{g/cm}^3$.
Errors $\dfrac{\Delta m}{m}=\dfrac{0.1}{125.0}=0.08\%$; $\dfrac{\Delta a}{a}=\dfrac{0.01}{2.50}=0.4\%$. Rule 1.11 with $\rho=m\,a^{-3}$: $\dfrac{\Delta\rho}{\rho}=0.08\%+3(0.4\%)=1.28\%$.
Result $\Delta\rho=8.00\times0.0128=0.10\ \text{g/cm}^3$, so $\rho=(8.0\pm0.1)\ \text{g/cm}^3=(8.0\pm0.1)\times10^3\ \text{kg/m}^3$.
Check The side, cubed, gives $1.2\%$ of the $1.28\%$: improving the balance would hardly help, but a screw gauge for the side would.
✏️ Example 5 — Young's modulus of a steel wire (JEE Main level)
In an experiment to find Young's modulus, $Y=\dfrac{MgL}{\pi(d/2)^2\,\ell}=\dfrac{4MgL}{\pi d^2\ell}$, a load $M=4.0\ \text{kg}$ (error negligible) stretches a wire of length $L=2.000\ \text{m}$ (measured to $1\ \text{mm}$) and diameter $d=0.50\ \text{mm}$ (screw gauge, LC $0.01\ \text{mm}$) by $\ell=2.00\ \text{mm}$ (measured to $0.01\ \text{mm}$). Take $g=9.8\ \text{m/s}^2$. Find $Y$ and its maximum percentage error. Which measurement limits the accuracy?
Work out the maximum percentage error in $Y$ yourself first:
Value Area $\tfrac{\pi d^2}{4}=1.963\times10^{-7}\ \text{m}^2$; $Y=\dfrac{MgL}{(\pi d^2/4)\,\ell}=\dfrac{78.4}{(1.963\times10^{-7})(2.00\times10^{-3})}$ $=2.00\times10^{11}\ \text{Pa}$.
Errors $\dfrac{\Delta L}{L}=\dfrac{0.1}{200.0}=0.05\%$; $\dfrac{\Delta d}{d}=\dfrac{0.01}{0.50}=2.0\%$, counted twice because of $d^2$: $4.0\%$; $\dfrac{\Delta\ell}{\ell}=\dfrac{0.01}{2.00}=0.5\%$.
Total $\dfrac{\Delta Y}{Y}=0.05\%+4.0\%+0.5\%=4.55\%\approx4.6\%$; $\Delta Y=0.09\times10^{11}\ \text{Pa}$, so $Y=(2.00\pm0.09)\times10^{11}\ \text{Pa}$.
Verdict The diameter supplies $4.0\%$ of the $4.55\%$: it is small and appears squared. That is why the diameter is measured with a screw gauge at several places and in two perpendicular directions. Interactive 2 (preset "Young's modulus") shows this error budget as bars.

Choose the formula $\rho=6m/(\pi d^3)$ and set both percentage errors — in the mass $m$ and in the diameter $d$ — to $1.00\%$. What total percentage error in $\rho$ will the widget show?

Move the two sliders to $1.00\%$: the bars read $m$: $1.00\%$ and $d$: $3.00\%$, and the total % error readout is $4.00\%$, with the diameter as the largest contribution ($3.00\%$, $75\%$ of the total). The diameter is cubed, so its relative error counts three times, and it is added even though $d$ sits in the denominator. To improve the density, measure the diameter more carefully — a better balance would hardly help.
Interactive 2 — Error Budget: Which Measurement Hurts Most?

Choose a formula and set the percentage error of each measured quantity. Each bar is that quantity's contribution, $|\text{power}|\times(\text{its percentage error})$; the bottom bar stacks them into the total percentage error of the result (Rule 1.11). Quantities in the denominator still add. Press Play to watch the contributions pile up formula by formula; move a slider to explore.

0.10%1.00%0.50%
formula—
total % error—
largest contribution—

$Z=\dfrac{A^2}{B^3}$, where $A$ is measured with a $1\%$ error and $B$ with a $2\%$ error. The maximum percentage error in $Z$ is

Reciprocals, repeated quantities and reporting the result

Rule 1.11 assumes each measured quantity appears once, as a power. Formulas such as the parallel-resistor or lens formula break this: in $R=\dfrac{R_1R_2}{R_1+R_2}$ the resistance $R_1$ appears both on top and underneath. Treating the numerator and denominator as independent would count the error in $R_1$ twice, although an increase in $R_1$ raises both and the effects partly cancel. Instead, work with the reciprocal form. Since $\dfrac{1}{x\pm\Delta x}=\dfrac1x\left(1\pm\dfrac{\Delta x}{x}\right)^{-1}\approx\dfrac1x\mp\dfrac{\Delta x}{x^2}$, the absolute error of $\dfrac1x$ is $\dfrac{\Delta x}{x^2}$. Applying Rule 1.9 to $\dfrac1R=\dfrac1{R_1}+\dfrac1{R_2}$ gives

$$\frac{\Delta R}{R^2}=\frac{\Delta R_1}{R_1^2}+\frac{\Delta R_2}{R_2^2}.$$

The same pattern holds for the lens formula written in magnitudes, $\dfrac1f=\dfrac1u+\dfrac1v$: $\dfrac{\Delta f}{f^2}=\dfrac{\Delta u}{u^2}+\dfrac{\Delta v}{v^2}$. Resistors in series are a plain sum: $\Delta R_s=\Delta R_1+\Delta R_2$.

✏️ Example 6 — Resistors in parallel (JEE Main level)
Two resistors, $R_1=(6.0\pm0.2)\ \Omega$ and $R_2=(3.0\pm0.1)\ \Omega$, are connected (a) in series and (b) in parallel. Find the equivalent resistance and its maximum error in each case. (c) Show why splitting $\tfrac{R_1R_2}{R_1+R_2}$ into numerator and denominator gives the wrong error.
Before you open the solution, find the parallel resistance and its maximum error:
(a) $R_s=9.0\ \Omega$, $\Delta R_s=0.2+0.1=0.3\ \Omega$: $R_s=(9.0\pm0.3)\ \Omega$ ($3.3\%$).
(b) $R_p=\dfrac{6.0\times3.0}{9.0}=2.0\ \Omega$. $\Delta R_p=R_p^2\left(\dfrac{0.2}{36}+\dfrac{0.1}{9}\right)=4.0\times(0.00556+0.01111)=0.067\ \Omega$, so $R_p=(2.00\pm0.07)\ \Omega$, a $3.3\%$ error.
(c) Splitting: numerator $3.3\%+3.3\%=6.7\%$, denominator $\dfrac{0.3}{9.0}=3.3\%$, "total" $10\%$ — three times too big, because $R_1$ and $R_2$ were each counted twice.
Check Worst case directly: $R_1=6.2$, $R_2=3.1$ gives $R_p=\dfrac{19.22}{9.3}=2.067\ \Omega$, exactly $0.067\ \Omega$ above $2.0$ ✓. Here both resistors have the same $3.3\%$ error, and so does $R_p$ — a handy JEE shortcut.

Reporting a result. Write the best value, the error and the unit together, with the error rounded to one significant figure and the value rounded to the same decimal place: $g=(9.7\pm0.1)\ \text{m/s}^2$, $Y=(2.00\pm0.09)\times10^{11}\ \text{Pa}$, never $g=9.6722\pm0.1066\ \text{m/s}^2$. This is the deeper reason behind the significant-figure rules: the digits you keep are the ones the error allows.

⚠️ JEE Main traps

1. Percentage errors for a difference. For $Z=A-B$ the absolute errors add; the percentage error of $Z$ is $\dfrac{\Delta A+\Delta B}{A-B}$, which can be huge (Example 3). Adding the percentage errors of $A$ and $B$ is wrong.

2. Subtracting for the denominator. In $g=\dfrac{4\pi^2l}{T^2}$ the error in $T$ is added: $\dfrac{\Delta g}{g}=\dfrac{\Delta l}{l}+2\dfrac{\Delta T}{T}$. Errors never cancel in the worst case.

3. Forgetting the power — or giving constants an error. Squares double, cubes triple, square roots halve. $\pi$, $\tfrac12$ and a counted number of oscillations are exact; dividing a timed interval by $20$ oscillations divides the absolute error by $20$ but leaves the percentage error unchanged.

4. A quantity that appears twice. Do not split $\dfrac{R_1R_2}{R_1+R_2}$ or $\dfrac{uv}{u+v}$ into separate pieces; use the reciprocal rule $\dfrac{\Delta R}{R^2}=\dfrac{\Delta R_1}{R_1^2}+\dfrac{\Delta R_2}{R_2^2}$ (Example 6).

5. Zero error and least count. Corrected reading $=$ observed $-$ zero error, with its sign (a negative zero error is added). An error can never be quoted smaller than the least count.

Two resistors, each $(10.0\pm0.2)\ \Omega$, are connected in parallel. The percentage error in the equivalent resistance is

Exercises

Use the maximum-error rules of this section. Round each final error to one significant figure and the value to the same decimal place.

I
Least count, accuracy and types of error.
  • 1 A metre scale is divided into millimetres. State its least count and write a length read as $23.4\ \text{cm}$ with its error and percentage error.
  • 2 On a pair of Vernier callipers $1\ \text{MSD}=1\ \text{mm}$ and $20$ Vernier divisions coincide with $19$ main-scale divisions. Find the least count.
  • 3 A screw gauge has pitch $0.5\ \text{mm}$ and $50$ divisions on its circular scale. Find its least count, and the reading when the linear scale shows $3.5\ \text{mm}$ and the $28$th circular division is on the reference line.
  • 4 A screw gauge reads $-0.03\ \text{mm}$ when its jaws are closed. A wire then reads $1.52\ \text{mm}$. Find its true diameter.
  • 5 Classify each error as systematic or random, and name its source: (a) a stopwatch that runs $1\%$ slow; (b) always reading a scale from slightly above; (c) small, unpredictable differences in when you press a stopwatch; (d) a steel tape used in hot sunshine.
  • 6 The true length of a block is $5.40\ \text{cm}$. Student A reads $5.12$, $5.13$, $5.11\ \text{cm}$; student B reads $5.2$, $5.6$, $5.4\ \text{cm}$. Whose set is more precise, and whose mean is more accurate?
  • 7 Explain in two sentences why averaging many readings reduces random error but not systematic error.
1 $\text{LC}=1\ \text{mm}=0.1\ \text{cm}$; $(23.4\pm0.1)\ \text{cm}$, percentage error $\dfrac{0.1}{23.4}\times100\%=0.43\%$.
2 $1\ \text{VSD}=\tfrac{19}{20}\ \text{mm}$; $\text{LC}=\tfrac{1}{20}\ \text{mm}=0.05\ \text{mm}$.
3 $\text{LC}=\dfrac{0.5}{50}=0.01\ \text{mm}$; reading $=3.5+28\times0.01=3.78\ \text{mm}$.
4 Corrected $=1.52-(-0.03)=1.55\ \text{mm}$.
5 (a) systematic, instrumental; (b) systematic, personal (parallax); (c) random; (d) systematic — the tape expands, so every length reads too small (an environmental/instrumental effect that can be corrected).
6 A is more precise (spread $0.02\ \text{cm}$) but its mean $5.12\ \text{cm}$ is far from $5.40$ — a systematic error. B's mean is $5.40\ \text{cm}$: more accurate, less precise.
7 Random errors are equally likely to be positive or negative, so they cancel in the sum of many readings. A systematic error has the same sign every time, so it survives in the mean unchanged.
II
Absolute, relative and percentage errors.
  • 1 The time for $20$ oscillations of a pendulum is measured five times: $39.6$, $40.2$, $39.9$, $40.4$, $39.9\ \text{s}$. Find the mean, the mean absolute error and the percentage error.
  • 2 From part 1, write the period of one oscillation with its error. What is its percentage error?
  • 3 Find the percentage errors in lengths of $4.0\ \text{cm}$ and $40.0\ \text{cm}$, each measured to $\pm0.1\ \text{cm}$.
  • 4 A $250\ \text{g}$ mass is known to $0.4\%$. Find its absolute error.
  • 5 A resistor is marked $470\ \Omega$ with a tolerance of $5\%$. Between which values does its resistance lie?
  • 6 Vernier callipers (LC $0.01\ \text{cm}$) give $3.27$, $3.29$, $3.28$, $3.27$, $3.29\ \text{cm}$. Write the result with its error.
  • 7 A distance of $12.40\ \text{m}$ has a relative error of $0.005$. Write it with its absolute error.
  • 8 Which is the more precise measurement: $(2.5\pm0.1)\ \text{m}$ or $(250\pm2)\ \text{m}$?
1 Mean $=\dfrac{200.0}{5}=40.0\ \text{s}$; deviations $0.4,\ 0.2,\ 0.1,\ 0.4,\ 0.1$, mean absolute error $\dfrac{1.2}{5}=0.24\ \text{s}$; percentage error $0.6\%$. Result $(40.0\pm0.2)\ \text{s}$.
2 $T=\dfrac{40.0}{20}=2.000\ \text{s}$, $\Delta T=\dfrac{0.24}{20}=0.012\ \text{s}$: $T=(2.00\pm0.01)\ \text{s}$. The percentage error is still $0.6\%$ — $20$ is an exact count.
3 $\dfrac{0.1}{4.0}=2.5\%$ and $\dfrac{0.1}{40.0}=0.25\%$: the same absolute error is ten times worse on the shorter length.
4 $\Delta m=250\times0.004=1\ \text{g}$: $(250\pm1)\ \text{g}$.
5 $\pm5\%$ of $470$ is $\pm23.5\ \Omega$: from $446.5\ \Omega$ to $493.5\ \Omega$.
6 Mean $=\dfrac{16.40}{5}=3.28\ \text{cm}$; mean absolute error $\dfrac{0.04}{5}=0.008\ \text{cm}$, below the LC, so quote the LC: $(3.28\pm0.01)\ \text{cm}$.
7 $\Delta=12.40\times0.005=0.062\ \text{m}$: $(12.40\pm0.06)\ \text{m}$.
8 Relative errors: $\dfrac{0.1}{2.5}=4\%$ and $\dfrac{2}{250}=0.8\%$. The second is more precise, although its absolute error is bigger.
III
Combination of errors.
  • 1 $A=(12.5\pm0.2)\ \text{cm}$ and $B=(8.3\pm0.1)\ \text{cm}$. Find $A+B$ and $A-B$ with their errors, and the percentage error of $A-B$.
  • 2 The temperature of water rises from $(20.4\pm0.5)\ ^\circ\text{C}$ to $(45.8\pm0.5)\ ^\circ\text{C}$. Find the rise with its error and percentage error.
  • 3 A rectangle has $l=(5.0\pm0.1)\ \text{cm}$ and $b=(2.0\pm0.1)\ \text{cm}$. Find its area with its error.
  • 4 A voltmeter reads $V=(12.0\pm0.2)\ \text{V}$ and an ammeter $I=(2.50\pm0.05)\ \text{A}$. Find $R=V/I$ with its error.
  • 5 $Z=\dfrac{A^4B^{1/3}}{C\,D^{3/2}}$, and the percentage errors in $A$, $B$, $C$, $D$ are $1\%$, $3\%$, $2\%$, $2\%$. Find the percentage error in $Z$.
  • 6 The mass and the speed of a body are measured with errors of $2\%$ and $3\%$. Find the maximum percentage error in its kinetic energy.
  • 7 The radius of a sphere is $(2.00\pm0.02)\ \text{cm}$. Find its volume with its error.
  • 8 The period of a pendulum is $T=2\pi\sqrt{l/g}$. If $l$ has a $2\%$ error and $g$ is exact, what is the percentage error in $T$?
  • 9 A ball of mass $(0.250\pm0.005)\ \text{kg}$ moves at $(12.0\pm0.3)\ \text{m/s}$. Find its momentum with its error.
Check: enter each quantity's power and percentage error (Rule 1.11) and compare the total with your answer
total % error—
$\Delta Z$—
1 $A+B=(20.8\pm0.3)\ \text{cm}$; $A-B=(4.2\pm0.3)\ \text{cm}$, whose percentage error is $\dfrac{0.3}{4.2}=7.1\%$.
2 Rise $=(25.4\pm1.0)\ ^\circ\text{C}$ (absolute errors add); $\dfrac{1.0}{25.4}=3.9\%$.
3 $lb=10.0\ \text{cm}^2$; $\dfrac{\Delta A}{A}=2\%+5\%=7\%$, so $(10.0\pm0.7)\ \text{cm}^2$.
4 $R=4.80\ \Omega$; $\dfrac{0.2}{12.0}+\dfrac{0.05}{2.50}=1.67\%+2.00\%=3.67\%$; $\Delta R=0.18\ \Omega$, so $R=(4.8\pm0.2)\ \Omega$.
5 $4(1\%)+\tfrac13(3\%)+1(2\%)+\tfrac32(2\%)=4+1+2+3=10\%$.
6 $K=\tfrac12mv^2$: $2\%+2(3\%)=8\%$.
7 $V=\tfrac43\pi r^3=33.5\ \text{cm}^3$; $\dfrac{\Delta V}{V}=3\times1\%=3\%$, $\Delta V=1.0\ \text{cm}^3$, so $V=(34\pm1)\ \text{cm}^3$.
8 $T\propto l^{1/2}$: $\tfrac12\times2\%=1\%$.
9 $p=mv=3.00\ \text{kg m/s}$; $\dfrac{0.005}{0.250}+\dfrac{0.3}{12.0}=2\%+2.5\%=4.5\%$; $\Delta p=0.135\ \text{kg m/s}$, so $p=(3.0\pm0.1)\ \text{kg m/s}$.
IV
Harder problems and applications.
  • 1 A pendulum of length $98.0\ \text{cm}$ (measured to $1\ \text{mm}$) makes $50$ oscillations in $100\ \text{s}$, timed with a watch of resolution $0.5\ \text{s}$. Find $g=\dfrac{4\pi^2l}{T^2}$ with its error.
  • 2 Find the equivalent resistance, with its error, of $R_1=(4.0\pm0.1)\ \Omega$ and $R_2=(12.0\pm0.3)\ \Omega$ in parallel.
  • 3 For a convex lens the object and image distances (magnitudes) are $u=(30.0\pm0.1)\ \text{cm}$ and $v=(60.0\pm0.1)\ \text{cm}$, and $\dfrac1f=\dfrac1u+\dfrac1v$. Find $f$ with its error.
  • 4 In a resonance-tube experiment with a $500\ \text{Hz}$ tuning fork, the first two resonance lengths are $l_1=(16.2\pm0.1)\ \text{cm}$ and $l_2=(50.6\pm0.1)\ \text{cm}$, and the speed of sound is $v=2f(l_2-l_1)$. Find $v$ with its error.
  • 5 The radius of a sphere is measured with a $1.5\%$ error. Find the percentage errors in its surface area and its volume.
  • 6 The area of a square plate is found with a $4\%$ error. What is the percentage error in the measurement of its side?
  • 7 (Modelling) The diameter of a wire is about $0.50\ \text{mm}$. Find the percentage error in the diameter and in the cross-sectional area if it is measured with (a) a metre scale (LC $1\ \text{mm}$), (b) Vernier callipers (LC $0.1\ \text{mm}$), (c) a screw gauge (LC $0.01\ \text{mm}$). Which would you use in a Young's modulus experiment?
1 $T=2.00\ \text{s}$; $g=\dfrac{4\pi^2(0.980)}{4.00}=9.67\ \text{m/s}^2$. $\dfrac{\Delta g}{g}=\dfrac{0.1}{98.0}+2\times\dfrac{0.5}{100}=0.10\%+1.0\%=1.1\%$; $\Delta g=0.11\ \text{m/s}^2$, so $g=(9.7\pm0.1)\ \text{m/s}^2$.
2 $R_p=\dfrac{48}{16}=3.0\ \Omega$; $\Delta R_p=9\left(\dfrac{0.1}{16}+\dfrac{0.3}{144}\right)=0.075\ \Omega$, so $R_p=(3.00\pm0.08)\ \Omega$ — $2.5\%$, the same as each resistor.
3 $\dfrac1f=\dfrac1{30}+\dfrac1{60}=\dfrac{3}{60}$, $f=20.0\ \text{cm}$; $\Delta f=f^2\left(\dfrac{0.1}{900}+\dfrac{0.1}{3600}\right)=400\times1.39\times10^{-4}=0.056\ \text{cm}$, so $f=(20.00\pm0.06)\ \text{cm}$.
4 $l_2-l_1=(34.4\pm0.2)\ \text{cm}$; $v=2\times500\times0.344=344\ \text{m/s}$; $\dfrac{0.2}{34.4}=0.58\%$, $\Delta v=2\ \text{m/s}$: $v=(344\pm2)\ \text{m/s}$.
5 $S\propto r^2$: $3\%$; $V\propto r^3$: $4.5\%$.
6 $A=a^2$, so $\dfrac{\Delta A}{A}=2\dfrac{\Delta a}{a}$: the side was measured with a $2\%$ error.
7 (a) $\dfrac{1}{0.50}=200\%$, area $400\%$ — useless; (b) $20\%$, area $40\%$; (c) $2\%$, area $4\%$. Only the screw gauge is fit for purpose, and even then the area (which enters $Y$) is the largest error in the experiment (Example 5).
🎯 Exam corner JEE Main · MCQJEE Main · Numerical

E1. JEE Main · MCQ A screw gauge has a pitch of $1\ \text{mm}$ and $100$ divisions on its circular scale. With the jaws closed it reads $+0.04\ \text{mm}$. When a wire is placed between the jaws, the linear scale shows $2\ \text{mm}$ and the $63$rd circular division is on the reference line. The diameter of the wire is (A) $2.67\ \text{mm}$ (B) $2.63\ \text{mm}$ (C) $2.59\ \text{mm}$ (D) $2.55\ \text{mm}$

LC $\dfrac{1\ \text{mm}}{100}=0.01\ \text{mm}$.
Reading Observed $=2+63\times0.01=2.63\ \text{mm}$; corrected $=2.63-0.04=2.59\ \text{mm}$.
Answer (C).
⏱ Shortcut A positive zero error means the gauge reads too much — the answer must be below the observed $2.63$. (A) adds the zero error; (B) ignores it.

E2. JEE Main · Numerical A physical quantity is given by $X=\dfrac{p^2\,q^{1/2}}{r^3\,s}$. The percentage errors in the measurements of $p$, $q$, $r$ and $s$ are $1.5\%$, $2\%$, $1\%$ and $2\%$. Find the maximum percentage error in $X$. Give the answer to the nearest integer.

Rule 1.11 $\dfrac{\Delta X}{X}=2(1.5\%)+\tfrac12(2\%)+3(1\%)+1(2\%)=3+1+3+2=9\%$.
Answer $9$.
⏱ Shortcut Write the powers in a row ($2,\ \tfrac12,\ 3,\ 1$) under the errors ($1.5,\ 2,\ 1,\ 2$), multiply column by column and add. A wrong answer of $3$ comes from subtracting the denominator terms — and in Section B a wrong number costs $1$ mark, just like a wrong MCQ option.

E3. JEE Main · MCQ In a simple-pendulum experiment the length is $40.0\ \text{cm}$, measured with a scale of least count $1\ \text{mm}$, and the time for $20$ oscillations is $25.0\ \text{s}$, measured with a watch of resolution $0.1\ \text{s}$. The maximum percentage error in $g=\dfrac{4\pi^2l}{T^2}$ is (A) $0.65\%$ (B) $1.05\%$ (C) $0.80\%$ (D) $0.15\%$

Errors $\dfrac{\Delta l}{l}=\dfrac{0.1}{40.0}=0.25\%$; $\dfrac{\Delta t}{t}=\dfrac{0.1}{25.0}=0.40\%$, and $\dfrac{\Delta T}{T}=\dfrac{\Delta t}{t}$ because $T=t/20$ with $20$ exact.
Combine $\dfrac{\Delta g}{g}=0.25\%+2(0.40\%)=1.05\%$.
Answer (B).
⏱ Shortcut (A) forgets the factor $2$ on $T$; (C) ignores the length; (D) subtracts. Timing many oscillations is what keeps the time error small.

E4. JEE Main · Numerical Two resistors $R_1=(6\pm0.3)\ \Omega$ and $R_2=(12\pm0.24)\ \Omega$ are connected in parallel. Find the maximum percentage error in the equivalent resistance. Give the answer to the nearest integer.

Value $R_p=\dfrac{6\times12}{18}=4\ \Omega$.
Reciprocal rule $\Delta R_p=R_p^2\left(\dfrac{0.3}{36}+\dfrac{0.24}{144}\right)=16\times(0.00833+0.00167)=0.16\ \Omega$.
Answer $\dfrac{0.16}{4}\times100\%=4\%$: enter $4$.
⏱ Shortcut The trap answer $10$ comes from splitting $\tfrac{R_1R_2}{R_1+R_2}$ ($5\%+2\%+3\%$). The parallel error always lies between the two individual percentage errors ($2\%$ and $5\%$).

E5. JEE Main · MCQ Five readings of a length taken with Vernier callipers of least count $0.01\ \text{cm}$ are $2.62$, $2.65$, $2.60$, $2.63$ and $2.65\ \text{cm}$. The percentage error, based on the mean absolute error, is closest to (A) $0.6\%$ (B) $0.4\%$ (C) $1.1\%$ (D) $3.0\%$

Mean $\bar a=\dfrac{13.15}{5}=2.63\ \text{cm}$.
MAE Deviations $0.01,\ 0.02,\ 0.03,\ 0.00,\ 0.02$; $\Delta\bar a=\dfrac{0.08}{5}=0.016\ \text{cm}$ (above the LC).
Answer $\dfrac{0.016}{2.63}\times100\%=0.61\%$: (A).
⏱ Shortcut (B) uses the least count, (C) the largest deviation and (D) the sum of the deviations instead of their mean.

E6. JEE Main · Numerical In an experiment a quantity is calculated as $Z=\dfrac{(a-b)^2}{c}$, where $a=(25.3\pm0.1)\ \text{cm}$, $b=(20.3\pm0.1)\ \text{cm}$ and $c=(4.00\pm0.08)\ \text{cm}$. Find the maximum percentage error in $Z$. Give the answer to the nearest integer.

Difference $x=a-b=(5.0\pm0.2)\ \text{cm}$: a $4\%$ error (Rule 1.9).
Powers $Z=\dfrac{x^2}{c}$: $\dfrac{\Delta Z}{Z}=2(4\%)+\dfrac{0.08}{4.00}=8\%+2\%=10\%$.
Answer $10$.
⏱ Shortcut Collapse every sum or difference into a single quantity first, then apply the power rule. Adding the percentage errors of $a$ and $b$ ($0.4\%+0.5\%$) would give a badly wrong $\approx4\%$ for $Z$.
🔬 Research–Teaching NexusResearch-tutored · optional, about 45 minutes

Research question. How much does your own reaction time limit a hand-timed measurement — and how many oscillations of a pendulum should you time to get the period to $0.5\%$?

Why it matters. Sprint races, swimming events and industrial stop-watch studies all once relied on hand timing; elite sport switched to electronic timing because a human starting and stopping a watch adds both a random scatter and a systematic delay of the order of a tenth of a second. Medical staff counting a pulse for $15$ seconds instead of a full minute face exactly the same trade-off. The error rules of this section tell you how to beat the problem: time a longer interval and divide.

What to do.

  1. Measure your reaction time with the ruler-drop test: a partner drops a $30\ \text{cm}$ ruler between your fingers without warning; from the distance $d$ it falls before you catch it, $t=\sqrt{2d/g}$. Repeat ten times; find the mean and the mean absolute error.
  2. Make a pendulum about $1\ \text{m}$ long. Time $N=1,\ 5,\ 10,\ 20$ and $50$ oscillations, five times for each $N$.
  3. For each $N$ find the mean period $\bar T$ and its mean absolute error $\Delta\bar T$, and plot the percentage error against $N$.
  4. From your graph, find the smallest $N$ that gives the period to $0.5\%$, and the resulting percentage error in $g=4\pi^2l/T^2$.

Tools: A $30\ \text{cm}$ ruler, string, a small heavy bob, a metre scale, a phone stopwatch (optionally a phone slow-motion video to check your hand timing).

Hand in: A table of $\bar T\pm\Delta\bar T$ for each $N$, the graph, your reaction-time result, and a one-paragraph recommendation for a school lab sheet: how many oscillations, and why.

Going further. Film the pendulum in slow motion and find $T$ from the frame count. Is there a systematic difference between your hand-timed and video periods? Which of your errors does timing more oscillations reduce, and which does it leave unchanged?

Instructor note: a typical ruler drop of $d\approx20\ \text{cm}$ gives $t\approx0.20\ \text{s}$, but because the start and stop delays partly cancel, the random scatter in a hand-timed interval is usually only $0.05$–$0.1\ \text{s}$. With $T\approx2\ \text{s}$, timing one oscillation gives errors of several per cent; $20$ oscillations bring it to about $0.2\%$, since the timing error is spread over $N$ periods while the percentage error falls as $1/N$. Students should notice that a systematic delay (always stopping late) is not reduced by repetition, only diluted by a longer interval.

🧪 Self-check quiz

0/5 correct

Q1. On a pair of Vernier callipers, $50$ Vernier divisions equal $49$ main-scale divisions of $0.5\ \text{mm}$ each. The least count is

Q2. If $Z=A-B$, the maximum absolute error in $Z$ is

Q3. The momentum of a body of exactly known mass is measured with a $2\%$ error. The maximum percentage error in its kinetic energy $K=\dfrac{p^2}{2m}$ is

Q4. Which error is reduced by repeating a measurement many times and averaging?

Q5. A quantity $x$ is measured with a $6\%$ error. The percentage error in $\sqrt{x}$ is