Uncertainty, gradients and model testing
Introducido| English | Significado |
|---|---|
| systematic error/ˌsɪstəˈmætɪk ˈerə/ | error sistemático |
| random error/ˈrændəm ˈerə/ | Unpredictable variation between repeated measurements. |
| zero error/ˈzɪərəʊ ˈerə/ | An instrument reading that is nonzero when the true quantity is zero. |
| uncertainty/ʌnˈsɜːtənti/ | incertidumbre |
| absolute uncertainty/ˈæbsəluːt ʌnˈsɜːtənti/ | An uncertainty quoted in the same unit as the measured quantity, such as ±1 mm for a length. |
| percentage uncertainty/pəˈsentɪdʒ ʌnˈsɜːtənti/ | Absolute uncertainty divided by the magnitude of the measured value, multiplied by one hundred. |
| error bar/ˈerə bɑː/ | A line through a plotted point showing the uncertainty interval of the measurement. |
| line of best fit/laɪn ɒv best fɪt/ | A line chosen to summarise the overall relationship between two variables in a scatter plot. |
| gradient/ˈɡreɪdɪənt/ | The change in the vertical coordinate divided by the change in the horizontal coordinate. |
| model/ˈmɒdl/ | A simplified representation of a system used to explain, analyse or predict its behaviour. |
Faster than light? OPERA, 2011
- In September 2011 the OPERA experiment reported neutrinos arriving from CERN in Italy 60 ns earlier than light would. The team asked other scientists to find the mistake.
- In February 2012 they found it: a loose fibre-optic cable in the timing system. It was a systematic error 系统误差, and a claim that broke physics turned out to be a measurement problem.
Random and systematic errors
- A random error 随机误差 scatters readings both ways — reaction time on a stopwatch, a fluctuating meter. Repeating and taking a mean reduces its effect.
- A systematic error shifts every reading the same way — a balance that reads 0.2 g with nothing on it (a zero error 零点误差), or a loose cable. Repeats do not remove it; you must find and correct the cause.
A balance reads 0.3 g with nothing on it. What kind of error is this?
Dificultad 1/5 · 100 monedas + un pequeño bono en el primer intento de hoy
It shifts every reading by the same 0.3 g. Subtract it from each reading or re-zero the balance.
Stating uncertainty
- Absolute uncertainty 绝对不确定度 is in the units of the reading: 25.0 ± 0.1 g. From repeats, use half the range: 15.2, 15.6 and 15.4 s give $15.4 \pm 0.2$ s.
- Percentage uncertainty 百分比不确定度 $= \dfrac{\text{absolute}}{\text{value}} \times 100\,\%$. When values are added or subtracted, add absolute uncertainties; when multiplied or divided, add percentages; a square doubles the percentage.
Three readings are 4.2 s, 4.6 s and 4.4 s. What is the absolute uncertainty in the mean, in s?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
The range is 4.6 − 4.2 = 0.4 s, and half the range is 0.2 s.
A length is $2.00 \pm 0.05$ m. What is the percentage uncertainty?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
$\dfrac{0.05}{2.00} \times 100 = 2.5\,\%$.
Worked example: the density of a stone
A stone has mass $25.0 \pm 0.1$ g and volume $10.0 \pm 0.5\ \text{cm}^3$. Find its density with its uncertainty.
- Value: $\rho = \dfrac{m}{V} = \dfrac{25.0}{10.0} = 2.50\ \dfrac{\text{g}}{\text{cm}^3}$.
- Percentages: mass $\dfrac{0.1}{25.0} \times 100 = 0.4\,\%$; volume $\dfrac{0.5}{10.0} \times 100 = 5\,\%$. A quotient adds them: $5.4\,\%$.
- Result: $5.4\,\%$ of 2.50 is 0.135, so $\rho = 2.50 \pm 0.14\ \dfrac{\text{g}}{\text{cm}^3}$. The volume dominates: a better measuring cylinder would help most.
Measurement ranges and the quotient estimate
50 ± 1 mm has a 2% uncertainty. Compare it with a 100 ± 3 mm reading: the summed quotient estimate is 5%, while exact endpoint calculations reveal a slightly asymmetric range.
The side of a square is $5.0 \pm 0.1$ cm. What is the percentage uncertainty in its area?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
The side has $\dfrac{0.1}{5.0} \times 100 = 2\,\%$. Area is side × side, so the percentages add: 4 %.
Graphs and error bars
- Plot each mean with an error bar 误差棒 showing its uncertainty, then draw one straight line of best fit 最佳拟合线 with points spread evenly on both sides.
- Find the gradient 斜率 from a large triangle: $\dfrac{\Delta y}{\Delta x}$. In the figure, the line goes through (0, 0) and (4, 8), so the gradient is $\dfrac{8}{4} = 2.0$.
A line of best fit passes through (0.5, 1.0) and (4.0, 8.0). What is its gradient?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
$\dfrac{8.0 - 1.0}{4.0 - 0.5} = \dfrac{7.0}{3.5} = 2.0$.
Uncertainty in a gradient
- Draw the steepest and the shallowest lines that still pass through every error bar. If they have gradients 2.3 and 1.8, the gradient is $2.0 \pm 0.25$, about $2.0 \pm 0.3$.
- The percentage uncertainty of the gradient is then $\dfrac{0.25}{2.0} \times 100 \approx 13\,\%$, and it carries into any quantity calculated from it.
The steepest line through the error bars has gradient 3.6 and the shallowest has gradient 3.0. What is the uncertainty in the gradient?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
Half the difference between the two lines: $\dfrac{3.6 - 3.0}{2} = 0.3$.
Testing a model
- A model 模型 such as $y \propto x$ predicts a straight line through the origin. In the figure every error bar touches the line, so the data are consistent with the model.
- "Consistent" does not mean proved. If the best line misses the origin by more than the uncertainty, look for a systematic error or a wrong model, as the OPERA team did.
Which of these reduce the effect of random error?
Dificultad 2/5 · 140 monedas + un pequeño bono en el primer intento de hoy
Averaging and measuring a larger quantity shrink random scatter. The last two correct systematic errors.
If a straight line through the origin passes through every error bar, the data prove that $y \propto x$.
Dificultad 1/5 · 100 monedas + un pequeño bono en el primer intento de hoy
The data are consistent with the model. Better measurements could still show a curve or an intercept.
"More repeats remove all error" is wrong. Repeats average out random error only. A zero error or a slow stopwatch shifts every repeat the same way, so the mean is just as wrong; you must recalibrate or correct the readings.
Every measurement carries an uncertainty: half the range for repeats, and percentages added for products and quotients. On a graph, error bars and steepest/shallowest lines give the gradient's uncertainty, and a model is supported only if the line fits within the error bars.
A student finds $g$ from a gradient: $9.6 \pm 0.4\ \dfrac{\text{m}}{\text{s}^2}$. The accepted value is $9.81\ \dfrac{\text{m}}{\text{s}^2}$. Is the result consistent, and how could the uncertainty be reduced?
Compare with the accepted value through the uncertainty, not just the difference.