Skip to main content
CBE Nexus · Global University Physics

Unit 6: Experimental Physics I

University experimental physics: measurement systems, uncertainty analysis, calibration, data acquisition, graphing, error propagation and reproducible laboratory practice.

All 36 Physics UnitsPrevious unitNext unit

Unit overview

Experimental physics connects mathematical models to observable quantities. A reliable experiment does more than produce a number: it defines the measurand, establishes a traceable measurement procedure, quantifies uncertainty, tests assumptions and reports enough information for another investigator to reproduce the result.

Level: Undergraduate university physics. Prerequisites: introductory mechanics, algebra, trigonometry, calculus and basic laboratory or mathematical reasoning appropriate to the topic. Study approach: work through each concept, reproduce the derivations, attempt the numerical examples independently, then complete the examination questions.

Learning outcomes

Measurandphysical quantitySensortransductionReadoutindicationResult± u
Figure 6.1. Original teaching schematic for Experimental Physics I. The drawing is conceptual and not to scale; use the accompanying definitions and equations for quantitative work.

Core textbook content

Experimental physics connects mathematical models to observable quantities. A reliable experiment does more than produce a number: it defines the measurand, establishes a traceable measurement procedure, quantifies uncertainty, tests assumptions and reports enough information for another investigator to reproduce the result. The sections below develop the main concepts from definitions to applications. For each topic, identify the system, list known quantities, choose a model, state approximations, and test the result against units and physical limits.

1. Measurement, measurands and SI units

A measurand is the quantity intended to be measured, stated with enough detail to distinguish it from related quantities. Before using an instrument, define the physical quantity, operating conditions, range and measurement model. SI base units provide a coherent foundation: metre, kilogram, second, ampere, kelvin, mole and candela. Derived units follow algebraically, such as the newton, joule, pascal and volt. A result without a unit is incomplete, while a unit attached to an ill-defined measurand does not make the result meaningful. Distinguish resolution, sensitivity, range, repeatability, reproducibility and accuracy. Resolution is the smallest displayed or detectable increment; it is not automatically the uncertainty. Sensitivity is the change in indication per change in input. Repeatability concerns measurements under the same conditions, whereas reproducibility includes changed operators, instruments or laboratories.

Worked example / application. A digital balance reads 125.34 g with a display increment of 0.01 g. The resolution is 0.01 g, but uncertainty must also consider calibration, repeatability, drift and sample handling. Reporting 125.34 g alone would imply more certainty than the display can justify.

2. Instrument characteristics and selection

Choose an instrument by matching its range, resolution, accuracy class, response time, loading effect and environmental tolerance to the measurand. A voltmeter with finite input resistance draws current and may alter a high-resistance circuit. An ammeter has a small but non-zero series resistance and can change the current being measured. A thermometer has thermal capacity and requires time to approach equilibrium. Sensor calibration converts indication into a physical value through a documented relationship. Check zero offset, linearity, hysteresis, drift, dead band and repeatability. A calibration certificate states the reference, conditions, corrections and uncertainty; it does not guarantee that the instrument remains calibrated indefinitely.

Worked example / application. A 0–10 V voltmeter with input resistance 1 MΩ is connected across a 100 kΩ resistor in series with 10 V and 100 kΩ. The meter loads the measured branch. The parallel resistance is (100 kΩ × 1 MΩ)/(1.1 MΩ) = 90.9 kΩ, so the measured voltage is 10 × 90.9/(100 + 90.9) = 4.76 V rather than the unloaded 5.00 V.

3. Random, systematic and gross effects

Random effects cause scatter and are treated statistically through repeated observations. Systematic effects shift results in a consistent direction and require calibration, correction or a revised model. Gross errors include recording the wrong range, transposing digits or using an incorrect unit; they should be prevented by procedures and checked against raw records rather than silently removed. Repetition reduces the standard error of a mean when independent random effects dominate, but repetition does not remove a common zero offset. A good laboratory report separates observed scatter from known corrections and residual limitations.

Worked example / application. Five independent time measurements are 2.01, 1.99, 2.03, 2.00 and 1.97 s. Their mean is 2.00 s. A stopwatch that runs 0.08 s fast still biases every reading; averaging cannot correct that systematic offset.

4. Describing data: mean, spread and distributions

For observations x_i, the arithmetic mean is x̄ = Σx_i/n. The sample variance s² = Σ(x_i−x̄)²/(n−1) estimates population variance when observations are independent and identically distributed. The sample standard deviation s describes scatter of individual observations; the standard uncertainty of the mean is s/√n under suitable assumptions. A histogram reveals shape, outliers and possible multiple populations. Median and interquartile range can be useful for skewed distributions. Never report only the mean when spread or distribution shape is scientifically relevant. Preserve raw data, units, significant digits and the conditions under which observations were collected.

Worked example / application. For 4.8, 5.0, 5.1, 4.9 and 5.2 cm, x̄ = 5.00 cm. Deviations are −0.2, 0, 0.1, −0.1, 0.2 cm; sum of squares = 0.10 cm². Thus s = √(0.10/4) = 0.158 cm and standard uncertainty of the mean = 0.071 cm.

5. Uncertainty and significant reporting

Measurement uncertainty quantifies the dispersion of values reasonably attributable to the measurand. Type A evaluation uses statistical analysis of observations; Type B uses other information such as calibration certificates, manufacturer specifications, resolution or prior knowledge. Standard uncertainty is expressed like a standard deviation. Expanded uncertainty U = k u_c uses a coverage factor k, often approximately 2 when a roughly 95% coverage interpretation is justified. State the coverage factor and basis rather than treating k = 2 as universally exact. Round uncertainty to one or two significant figures and round the measured value to the same decimal place.

Worked example / application. A length is 24.63 cm with combined standard uncertainty 0.08 cm. With k = 2, U = 0.16 cm. Report 24.63 ± 0.16 cm (k = 2), subject to the stated coverage assumptions.

6. Propagation of uncertainty

If y = f(x_1,…,x_n), and input uncertainties are small, the first-order law of propagation uses sensitivity coefficients: u_c²(y) = Σ(c_i²u²(x_i)) + covariance terms, where c_i = ∂f/∂x_i. For independent inputs the covariance terms vanish. For y = x^a, relative standard uncertainty is approximately |a|u(x)/x. For products and quotients, independent relative uncertainties combine in quadrature. For sums and differences, absolute uncertainties combine in quadrature. These approximations rely on a locally linear model and suitable probability assumptions; nonlinear or bounded problems may need Monte Carlo propagation.

Worked example / application. A resistance R = V/I with V = 12.0 ± 0.1 V and I = 2.00 ± 0.03 A. R = 6.00 Ω. Relative uncertainty = √[(0.1/12)²+(0.03/2)²] = 0.0176, so u(R) ≈ 0.106 Ω. Report approximately 6.00 ± 0.11 Ω as standard uncertainty.

7. Calibration curves and regression

A calibration curve relates instrument indication to a reference value. For a linear model y = a + bx, least-squares fitting estimates slope and intercept by minimising the sum of squared vertical residuals, when the assumptions of the model are appropriate. Inspect residuals for curvature, changing variance, drift and outliers. A high coefficient of determination alone does not prove a valid model. If both axes have significant uncertainty, ordinary least squares may be inappropriate; consider orthogonal-distance methods or a physically justified model. Do not extrapolate beyond the calibrated range without evidence.

Worked example / application. A sensor gives output 0.42, 0.81 and 1.19 V at reference temperatures 10, 20 and 30 °C. The approximate slope is (1.19−0.42)/(30−10)=0.0385 V °C⁻¹. A full calibration should fit all points, estimate uncertainty and inspect residuals.

8. Graphing and linearisation

A graph should communicate the tested physical relationship. Put the independent variable on the horizontal axis and dependent variable on the vertical axis unless the model dictates otherwise. Label axes with quantity and unit, select readable scales, show uncertainty bars when meaningful, and distinguish data from fitted models. Linearisation can reveal parameters: for exponential decay y = y₀e^(−t/τ), plotting ln y against t gives slope −1/τ. Transformations also transform uncertainty and may alter error distributions, so use them thoughtfully. A straight line is not proof of causation or of a correct physical model.

Worked example / application. For ln A versus time, a fitted gradient of −0.25 s⁻¹ implies τ = −1/slope = 4.0 s. If the gradient uncertainty is 0.02 s⁻¹, first-order propagation gives u(τ) ≈ 0.32 s.

9. Experimental design and control variables

A defensible experiment begins with a question, a testable model and a defined dependent variable. Identify independent variables, control variables, confounders and a feasible measurement range. Use repeats to estimate variability and choose enough measurement points to reveal departures from a model. Randomise order where drift or environmental changes could bias the sequence. Include blanks, controls or reference standards when appropriate. Design for sensitivity: the expected signal should be distinguishable from noise and instrument resolution. A pilot experiment can expose practical limitations before the full data collection.

Worked example / application. To test pendulum period against length, vary length over a broad safe range, measure several oscillations per timing run, repeat each length, keep amplitude small and approximately constant, and record temperature and timing method. Measuring one swing only magnifies reaction-time error.

10. Data acquisition, sampling and aliasing

A digital acquisition system samples a continuous signal at discrete times. The sampling frequency must be high enough to represent the highest relevant signal frequency; the Nyquist criterion requires sampling above twice the highest frequency for ideal band-limited signals, while practical systems need anti-alias filtering and margin. Quantisation maps continuous amplitude to finite digital steps. Sensor bandwidth, ADC resolution, trigger settings, grounding and electromagnetic interference all affect data quality. Record sampling rate, bit depth, filter settings, sensor range and time base so that the measurement can be interpreted later.

Worked example / application. A vibration signal contains frequencies up to 800 Hz. Sampling at 1 kHz is inadequate because components above 500 Hz can alias. A rate above 1.6 kHz is the theoretical minimum for a perfectly band-limited signal; a higher practical rate and suitable anti-alias filter are preferable.

11. Laboratory safety, ethics and reproducibility

Risk assessment must precede work involving electrical supplies, lasers, hot surfaces, vacuum systems, moving apparatus, chemicals or radiation. Isolate power before rewiring, use appropriate eye protection, never look into a laser beam, secure rotating parts and allow hot equipment to cool. Follow local rules for waste and hazardous materials. Scientific integrity requires retaining original data, documenting exclusions, distinguishing measured from calculated values and reporting inconvenient results. Reproducibility depends on apparatus diagrams, instrument models, calibration status, procedure, environmental conditions, analysis choices and uncertainty statements.

Worked example / application. If one reading appears anomalous, retain it in the raw data, investigate the apparatus and procedure, and document any justified exclusion using a pre-stated criterion. Removing it solely because it weakens the preferred conclusion is not acceptable.

12. Integrated experiment: determine g with a pendulum

For a simple pendulum at small angular amplitude, T = 2π√(L/g), so T² = (4π²/g)L. Plot T² against L; the gradient m = 4π²/g and g = 4π²/m. Measure length from pivot to the bob's centre of mass, not to its top. Time multiple oscillations, repeat measurements, keep amplitude small, and use several lengths. Evaluate uncertainty in length and timing, inspect linearity, and discuss air resistance, pivot friction, finite amplitude and human reaction time. The experiment tests a model and estimates a physical constant; it does not merely demonstrate a formula.

Worked example / application. If the fitted gradient is 4.03 s² m⁻¹, g = 4π²/4.03 = 9.80 m s⁻². If u(m)=0.05 s² m⁻¹, then u(g)/g ≈ u(m)/m = 0.0124, giving u(g)≈0.12 m s⁻².

Summary equations and principles

Model → assumptions → governing equation → boundary / initial data → solution → dimensional and physical checks.

Use the equations introduced in each section as a connected framework rather than as isolated formulae. Define every symbol, retain units through calculations and identify whether quantities are vectors, scalars, fields, rates or integrated totals. Check special cases and conservation statements before accepting a numerical answer.

University examination and problem set

Attempt all questions. Show derivations, diagrams, units and intermediate steps. Answers should distinguish exact results from approximations and should state assumptions explicitly.

  1. 1. Define the central quantities and state the assumptions used in experimental physics i.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  2. 2. Derive the principal governing relation for one of the systems discussed in this unit.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  3. 3. Explain how the choice of model or coordinate system affects the solution.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  4. 4. A measured or calculated quantity changes by 8%. Determine the effect on a dependent quantity using the relevant sensitivity relation.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  5. 5. Compare two limiting cases and explain the physical meaning of the difference.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  6. 6. Draw and label a diagram showing the geometry, directions and relevant variables for a representative problem.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  7. 7. Identify three sources of systematic limitation and explain how each could be reduced.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  8. 8. Construct a graph that would test a predicted linear relationship. State the expected gradient and intercept.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  9. 9. Solve a multi-step numerical problem using consistent SI units and show intermediate working.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  10. 10. Explain where the approximation used in the unit would fail and propose a more complete model.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  11. 11. Relate a mathematical result to an observable experimental consequence.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  12. 12. Evaluate a proposed experimental method, including controls, uncertainty and safety where relevant.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  13. 13. Show how a conservation law or boundary condition constrains the solution.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  14. 14. Interpret a non-ideal result and distinguish measurement effects from failure of the theoretical model.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

  15. 15. Write a concise scientific conclusion that includes the result, uncertainty or limitations, and physical interpretation.

    Answer guidance: Define symbols and assumptions; show the governing equation; substitute values with units; check dimensions and limiting behaviour; state the physical interpretation. For explanatory items, support the argument with a labelled sketch or a relevant equation.

Independent study and further reading

Use a recognised university-level physics textbook and laboratory manual alongside these notes. Compare notation and sign conventions across sources, but do not combine formulae until their conventions and assumptions have been reconciled. For deeper study, locate peer-reviewed papers or university lecture material on the specific physical model, then reproduce at least one derivation or numerical result independently.

For revision, prepare a one-page concept map linking definitions, governing equations, experimental observables, limiting cases and common failure modes. Rework the examples with altered numerical values and explain why the qualitative outcome does or does not change.

Measurement models and the definition of the measurand

A measurement is an experimentally supported estimate of a specified quantity, not simply a number copied from a display. The quantity intended to be measured is the measurand. It must be defined with sufficient precision to distinguish it from nearby or related quantities. “The length of a rod” may be ambiguous unless the endpoints, temperature, orientation, applied force and state of the rod are specified. A metal rod expands when heated; a flexible specimen may change length under tension; an irregular object may have several possible dimensions. A complete definition therefore includes the physical object, the property of interest and the conditions under which the property is defined.

Represent a measurement process by a model y=f(x₁,x₂,…,xₙ;θ), where y is the output estimate, xᵢ are input quantities and θ represents model parameters or environmental variables. The model may be a simple equation such as density ρ=m/V or a more complicated calibration function. Every input must have a defined unit and a traceable origin. A measurement result is incomplete if the method used to obtain an input, the corrections applied or the assumptions in the model are hidden.

The measurement chain includes the physical system, the interaction between the system and sensor, transduction, signal conditioning, digitisation, recording and data analysis. Each stage can introduce an effect. A thermometer exchanges heat with the object; a voltmeter draws current; a balance is influenced by air currents; a photodetector has a wavelength-dependent response. Instrument loading occurs when the measuring device changes the state of the system. A good experimental design estimates this effect and either makes it negligible, corrects it or includes it in the uncertainty budget.

Distinguish a true value, a reference value and a measured value. A true value is the ideal value of the defined quantity, but it is rarely known exactly. A reference value is accepted for a stated purpose and has its own uncertainty. A measured value is the estimate obtained by a particular procedure. Agreement with a reference does not prove that the method is free from bias: two errors can compensate. Conversely, a small difference does not establish that the physical model is correct.

Worked example — specify the measurand.

A student reports the diameter of a copper wire as 0.82 mm. Improve the definition and name two conditions to control.

Solution. Define the measurand as the mean external diameter of a stated straight section of the wire, measured perpendicular to its axis at a specified temperature using a calibrated micrometer and a controlled contact force. Temperature affects thermal expansion; excessive contact force can deform the wire. Several positions and orientations should be measured to test uniformity and ovality.

SI units, dimensional analysis and measurement traceability

The SI system is based on seven base units: second (s), metre (m), kilogram (kg), ampere (A), kelvin (K), mole (mol) and candela (cd). Derived units are products of powers of base units. A newton is kg m s⁻², a joule is kg m² s⁻², a pascal is kg m⁻¹ s⁻² and a volt is kg m² s⁻³ A⁻¹. Unit symbols are case-sensitive: m is metre, M is the prefix mega when attached to a unit, and kg is kilogram. A number without a unit is often physically incomplete.

Dimensional analysis checks whether an equation is internally consistent. If terms are added, they must have the same dimensions. If a quantity is a length, its expression must reduce to L; if it is an acceleration, it must reduce to LT⁻². Dimensional consistency is necessary but not sufficient. Both 2π√(L/g) and 3√(L/g) have dimensions of time, but only the first has the correct coefficient for the ideal small-angle pendulum. Physical reasoning, boundary conditions or experiment must establish the dimensionless constants.

Traceability is a documented, unbroken chain of calibrations linking a measurement result to an appropriate reference, with uncertainty stated at each stage. A traceable result is not automatically accurate: the measurand may be poorly defined, the instrument may be misused, the sample may be unrepresentative or the model may be inadequate. A laboratory should record the instrument identity, calibration date, certificate number, range, correction values, environmental conditions and any restrictions on use.

Calibration compares indications with reference values under stated conditions. Verification determines whether an instrument meets specified requirements. Adjustment changes the instrument response. These activities are related but are not interchangeable. A calibration certificate may identify an offset without authorising the operator to alter the instrument. Calibration points should span the range of intended use and should include enough points to reveal nonlinearity or hysteresis.

Worked example — dimensional check.

Show that v²/r has the dimensions of acceleration.

Solution. Velocity has dimension LT⁻¹ and radius has dimension L. Therefore [v²/r]=L²T⁻²/L=LT⁻², which is the dimension of acceleration. The expression is dimensionally suitable for centripetal acceleration, but dimensional analysis does not determine its vector direction.

Instrument performance: range, resolution, sensitivity and response

Range is the interval over which an instrument is intended to operate; span is the difference between its upper and lower limits. Resolution is the smallest change in input that produces a discernible change in indication. Sensitivity is the ratio of output change to input change. Accuracy describes closeness to a reference value in qualitative terms, while precision describes closeness among repeated results. A digital meter displaying 1 mV increments has a resolution of 1 mV, but its accuracy may be much worse.

Repeatability concerns measurements made by the same method, operator, instrument and location over a short interval. Reproducibility concerns agreement when specified conditions change, such as operator, laboratory or instrument. Hysteresis occurs when an instrument gives different indications at the same input depending on the direction from which the input was approached. Drift is a gradual change in indication with time. Dead band is an input interval over which the output does not respond. Response time describes how quickly an instrument follows a changing input.

Linearity describes the closeness of the input-output relationship to a straight line over a stated range. A sensor may be linear over one range and nonlinear outside it. A calibration curve may be represented by I=a+bx, where a is the offset and b is the sensitivity, or by a nonlinear function if the physical response requires it. Do not force a straight line through data simply because the instrument display is digital or the graph looks nearly linear.

Electrical instruments have finite input characteristics. An ideal voltmeter has infinite resistance and draws no current; a real voltmeter has finite input resistance. An ideal ammeter has zero resistance; a real ammeter introduces a small voltage drop. The input impedance of an oscilloscope probe can load a circuit, especially at high frequency. Similar loading occurs when a thermometer, force sensor or pressure transducer alters the system it measures.

Worked example — instrument specification.

A meter has a limit of error ±(0.5% of reading + 2 counts), with a least significant digit of 0.01 V. It indicates 8.42 V. Calculate the stated limit.

Solution. The percentage term is 0.005×8.42=0.0421 V. The count term is 2×0.01=0.02 V. Total limit=0.0621 V, approximately ±0.06 V. The displayed digits do not imply that the result is accurate to 0.01 V.

Random effects, systematic effects and gross mistakes

Random effects cause observations to vary in an unpredictable way under nominally identical conditions. Sources include electrical noise, vibration, reaction time, small temperature fluctuations and unstable contacts. Repeated observations help estimate the dispersion caused by these effects. Systematic effects shift results in a consistent direction or according to a repeatable pattern. Examples include a zero offset, incorrect scale factor, calibration bias, parallax and uncorrected temperature dependence. Repetition does not remove a constant systematic effect.

Gross mistakes include incorrect transcription, wrong range selection, miscounted oscillations, reversed units and accidental disconnection. Such events should be prevented by procedure and detected by checks. They should not be casually described as random errors. An observation should only be excluded when there is a documented and defensible reason, such as a known malfunction or recording error. Removing a point merely because it disagrees with the expected theory is scientifically unacceptable.

Absolute error is expressed in the units of the quantity. Relative error is the absolute difference divided by a reference magnitude, while percentage error is relative error multiplied by 100%. Uncertainty is not the same as error: error is a difference from a reference, while uncertainty describes the dispersion of values reasonably attributable to the measurand. When the reference itself is uncertain, that uncertainty must be included in a comparison.

Accuracy and precision are independent ideas. A tight cluster displaced from the reference is precise but biased. A broad cluster centred on the reference may have a mean close to the reference but poor repeatability. Calibration corrections can address known systematic effects; repeated readings, shielding, stable environmental conditions and improved apparatus can reduce random variation. A reliable experiment needs both.

Worked example — compare two measurement sets.

A reference resistor is 100.0 Ω. Group A obtains 104.1, 104.0 and 104.2 Ω. Group B obtains 99.0, 101.0 and 100.0 Ω. Compare the results.

Solution. Group A has a narrow spread and mean 104.1 Ω, so it is repeatable but biased high by about 4.1%. Group B has mean 100.0 Ω but a wider spread. It agrees with the reference on average but is less repeatable. Group A should investigate calibration and connections; Group B should investigate noise, contact stability and reading procedure.

Type A evaluation: mean, standard deviation and confidence intervals

For n independent observations x₁,…,xₙ, the arithmetic mean is x̄=Σxᵢ/n. The sample standard deviation is s=√[Σ(xᵢ−x̄)²/(n−1)]. It describes the spread of individual observations. The denominator n−1 accounts for the fact that the mean is estimated from the same sample. The standard uncertainty of the mean is s/√n when the observations are independent and the process is stable.

The standard uncertainty decreases in proportion to 1/√n. Doubling the number of observations does not halve the uncertainty; approximately four times as many independent observations are needed to halve it. This improvement is not unlimited. A common calibration offset, drift or model error is not reduced by averaging. Before calculating statistics, plot observations against measurement order to detect trends, clusters, periodic interference or abrupt changes.

For small samples from an approximately normal population, a confidence interval for the population mean uses the Student t distribution: x̄ ± t(1−α/2,n−1)s/√n. The multiplier depends on the selected confidence level and degrees of freedom. A 95% confidence procedure means that, under repeated sampling and the model assumptions, 95% of intervals constructed in this way would contain the fixed population mean. It is not a statement that the fixed mean has a 95% probability of lying in an already calculated interval.

Correlated observations require special care. If readings are taken rapidly from a slowly drifting instrument, neighbouring values may be correlated. The usual s/√n formula can then underestimate uncertainty. Record the sequence and consider a time-series or block-averaging analysis. Do not retain only the final average; preserve raw readings so that patterns can be investigated later.

Worked example — standard uncertainty of a mean.

Five times are 2.41, 2.38, 2.43, 2.40 and 2.38 s. Calculate the mean and standard uncertainty.

Solution. The mean is 2.400 s. The deviations are 0.010, −0.020, 0.030, 0 and −0.020 s; their squared sum is 0.0018 s². Thus s=√(0.0018/4)=0.0212 s and u(x̄)=0.0212/√5=0.00949 s. Report approximately (2.400 ± 0.009) s as a standard-uncertainty result.

Type B evaluation: resolution, certificates and specifications

Type B evaluation uses information other than the statistical analysis of repeated observations. Sources include calibration certificates, manufacturer specifications, resolution, reference tables, previous measurements and physical limits. Type A and Type B classify methods of evaluating uncertainty; they are not simply synonyms for random and systematic effects. A systematic contribution can be evaluated statistically in a suitable experiment, and a random contribution can be estimated from a specification.

If a digital display rounds to the nearest increment q and every rounding error within ±q/2 is considered equally plausible, the distribution is rectangular. Its variance is u²=(1/q)∫ from −q/2 to q/2 e²de=q²/12, so its standard uncertainty is q/√12. This model is justified only when the rectangular assumption is reasonable. An accuracy limit from a manufacturer may require another probability distribution.

If a calibration certificate gives expanded uncertainty U with coverage factor k, the standard uncertainty is u=U/k, provided the certificate's stated model applies. A tolerance should not automatically be divided by √3 or √12 unless the distributional assumption justifies it. Document the source, limit, distribution, divisor, sensitivity coefficient and resulting contribution in an uncertainty budget.

Worked example — digital resolution.

A balance displays to 0.01 g. Estimate the standard uncertainty due only to rounding.

Solution. q=0.01 g and the rounding interval is ±0.005 g. Thus u=q/√12=0.00289 g. Calibration, repeatability, buoyancy, drift and handling effects may add further contributions.

Combining uncertainty contributions and coverage intervals

For independent standard uncertainty contributions u₁,…,uₘ affecting the same measurand, the combined standard uncertainty is u_c=√(u₁²+…+uₘ²). This follows because the variance of a sum of independent quantities equals the sum of their variances. It is not correct to add standard uncertainties arithmetically unless a conservative bound or another model specifically requires it.

Independence must be assessed. Two instruments calibrated against the same reference may share a correlated contribution. For two quantities a and b, Var(a+b)=u²(a)+u²(b)+2cov(a,b), while Var(a−b)=u²(a)+u²(b)−2cov(a,b). Positive correlation can increase uncertainty in a sum and reduce uncertainty in a difference. Omitting covariance can therefore either overestimate or underestimate the result.

Expanded uncertainty is U=k u_c, where k is a coverage factor chosen to give an interval with a stated coverage interpretation under the adopted model. A factor near 2 is often used for approximately 95% coverage when the distribution is close to normal and effective degrees of freedom are sufficiently large. It is not a universal rule. Small samples, non-normal distributions or strongly nonlinear models may require a different factor or Monte Carlo interval.

Worked example — uncertainty budget.

Independent standard contributions are 0.20 mm from repeatability, 0.10 mm from calibration and 0.06 mm from resolution. Calculate u_c and U for k=2.

Solution. u_c=√(0.20²+0.10²+0.06²)=√0.0536=0.2315 mm. U=2u_c=0.463 mm. A result of 125.42 mm could be reported as (125.42 ± 0.46) mm, k=2, with the coverage interpretation stated.

Derivation of uncertainty propagation for calculated quantities

Let y=f(x₁,…,xₙ). Expand the function in a Taylor series about the measured input values. For small input deviations, retain only first-order terms: δy≈Σ(∂f/∂xᵢ)δxᵢ. The partial derivatives are sensitivity coefficients. They show how strongly the output responds to small changes in each input.

Taking the variance gives u²(y)=ΣᵢΣⱼ(∂f/∂xᵢ)(∂f/∂xⱼ)cov(xᵢ,xⱼ). If the inputs are independent, off-diagonal covariance terms vanish and u²(y)=Σᵢ[(∂f/∂xᵢ)u(xᵢ)]². This is the first-order law of propagation of uncertainty. It assumes the model is sufficiently linear over the uncertainty region. Large uncertainties, strong curvature or asymmetric distributions may require numerical propagation.

For a product y=ab, dy=b da+a db; dividing by ab gives dy/y=da/a+db/b. For independent inputs, [u(y)/y]²=[u(a)/a]²+[u(b)/b]². For a quotient, the differential has a minus sign but independent variance contributions still add. For y=aⁿ, dy/y=n da/a, so the relative uncertainty contribution is |n|u(a)/|a|. For a sum or difference, use absolute uncertainties because relative uncertainty can become misleading when the result is close to zero.

Worked example — resistivity.

R=4.00 ± 0.04 Ω, L=2.00 ± 0.01 m and d=0.500 ± 0.005 mm. Use ρ=Rπd²/(4L).

Solution. Convert d=5.00×10⁻⁴ m. Then ρ=3.927×10⁻⁷ Ω m. Relative uncertainty is √[(0.04/4.00)²+(2×0.005/0.500)²+(0.01/2.00)²]=0.0229, or 2.29%. Thus u(ρ)=0.090×10⁻⁷ Ω m and ρ≈(3.93 ± 0.09)×10⁻⁷ Ω m as a standard-uncertainty result.

MeasuranddefinitionSensor andinstrumentSignalconditioningSampling andrecordingModel andreported resultCalibration andresolutionNoise andenvironmentSampling andmodel limitsUncertainty can enter at every stage of the measurement chain.
Figure: measurement chain and common uncertainty sources.

Calibration curves, zero correction, hysteresis and drift

A calibration experiment records instrument indication I for known reference input x. A linear model I=a+bx contains an offset a and sensitivity b. If the response is ideal in the selected units, a=0 and b=1. A correction is applied algebraically using a stated sign convention. If an instrument reads +0.12 mm at zero, subtract 0.12 mm from subsequent indications. If it reads −0.08 mm at zero, add 0.08 mm. Preserve the raw indication and record the correction separately.

Calibration should cover the intended range and relevant operating conditions. A sensor calibrated at room temperature may respond differently when hot. A force sensor may have cross-axis sensitivity, and a pressure sensor may exhibit hysteresis. Measure reference points while increasing and decreasing the input to test for hysteresis. A gradual change in response over time is drift; periodic reference checks can reveal it. If a correction is interpolated between checks, justify the assumed drift model.

For a linear calibration, plot indication against reference input and inspect the intercept, slope and residuals. A nonlinear response may require a physical response equation, polynomial or lookup table. Avoid high-order polynomials without justification because they can oscillate between calibration points. Calibration uncertainty includes reference uncertainty, repeatability, fitted parameter uncertainty and interpolation effects. A result outside the calibrated range is an extrapolation and should be treated cautiously.

Reference input, xInstrument indication, IIdeal responseMeasured response
Figure: an offset and sensitivity difference separate the measured response from the ideal line.
Worked example — calibration equation.

A sensor follows I=0.15+0.982x, with both quantities in millimetres. It indicates 9.97 mm. Estimate x.

Solution. x=(I−0.15)/0.982=(9.97−0.15)/0.982≈10.00 mm. The calibration equation is valid only over its calibrated range. Uncertainty in the offset, sensitivity and reference points must be included in a full uncertainty budget.

Experimental graphs and the derivation of least-squares regression

A graph should show the independent variable horizontally and dependent variable vertically unless the physical model suggests another arrangement. Label each axis with quantity and unit. Use a scale that occupies the plotting area without implying unsupported precision. Plot observations and uncertainty bars rather than drawing a smooth curve through every point. A gradient has units of vertical quantity divided by horizontal quantity and often represents a physical parameter.

For y=mx+c, define residual rᵢ=yᵢ−mxᵢ−c and minimise S=Σrᵢ². Differentiating with respect to c gives −2Σ(yᵢ−mxᵢ−c)=0, hence c=ȳ−m x̄. Differentiating with respect to m and substituting the intercept gives m=Σ(xᵢ−x̄)(yᵢ−ȳ)/Σ(xᵢ−x̄)². The line therefore passes through the centroid (x̄,ȳ) when the intercept is free.

Ordinary least squares assumes x is known with negligible uncertainty and y errors are independent with common variance. If each point has a different standard uncertainty σᵢ, weighted least squares minimises χ²=Σ[(yᵢ−f(xᵢ;θ))/σᵢ]². If x uncertainty is significant, use an errors-in-variables approach. If residuals are correlated, their covariance matrix must be included. A high coefficient of determination does not prove that the model is physically correct.

Residuals are observed values minus model predictions. Random scatter around zero supports the model more than a curved pattern, funnel shape or time trend. Reduced chi-square χ²ᵣ=χ²/ν, where ν is degrees of freedom, is often near one when the model and uncertainty estimates are reasonable. A much larger value can indicate underestimated uncertainties or an inadequate model; a much smaller value can indicate overestimated uncertainties or correlation. These are diagnostic clues, not automatic conclusions.

Worked example — least-squares line.

For (1,3), (2,5) and (3,7), calculate the fitted line.

Solution. x̄=2 and ȳ=5. The numerator is (−1)(−2)+0+(1)(2)=4; the denominator is 1+0+1=2. Thus m=2 and c=ȳ−m x̄=1. The fitted line is y=2x+1. The points lie exactly on it, so the residual sum of squares is zero.

Experimental design, replication, randomisation and sampling

Begin with a question that can be answered using observable evidence. State a hypothesis or quantitative model and identify the independent variable, dependent variable, controlled variables and nuisance variables. A controlled variable is deliberately held fixed. A nuisance variable may influence the result but cannot be perfectly controlled; it should be monitored or included in the analysis. Choose a measurement range that is safe and lies within the calibrated range of the apparatus.

Replication means repeating an experimental condition. Repeated readings at one condition estimate short-term repeatability; independent repetitions of the full procedure test a broader range of variability. Randomising the order of measurements reduces confounding between measurement order and slow environmental changes. For example, measuring all low temperatures first and all high temperatures later can confuse temperature dependence with instrument drift. Alternating or randomising settings helps separate these effects.

Sampling frequency must be adequate for the phenomenon. A band-limited signal with maximum frequency f_max must be sampled above 2f_max under the ideal Nyquist condition, and practical reconstruction requires a higher rate and anti-alias filtering. A transient may be missed by slow sampling even if the average is recorded accurately. Report the sampling interval, trigger, sensor response time, filter and acquisition duration.

Choose measurement points to distinguish competing models. If two models differ mainly at high values, a cluster near zero provides little information. A broad range can improve slope precision, provided the physical model remains valid. Pilot measurements help estimate signal size, noise and practical constraints. Blinding can reduce observer bias when subjective judgement is involved, for example in visual threshold measurements.

Worked example — identify a confounder.

A resistance measurement increases as current is stepped upward in ascending order. Give two explanations and an improved design.

Solution. The increase may be genuine temperature dependence from Joule heating or instrument drift correlated with time. Randomise or alternate current settings, monitor temperature, allow thermal equilibrium where appropriate and repeat selected settings at the start and end. Consider a four-wire method to reduce lead resistance.

Practical instruments: vernier calipers, micrometers, balances and timers

A vernier caliper measures external dimensions, internal dimensions and depth. Read the main scale and identify the vernier division that aligns with a main-scale mark. Check zero before measurement. A positive zero error is subtracted from the indicated reading; a negative zero error is corrected by adding its magnitude. Keep the jaws aligned with the dimension and avoid excessive force. Dirt on the jaws and parallax can produce errors larger than the nominal least count.

A micrometer screw gauge converts rotation into linear displacement. If the screw pitch is p and there are N circular divisions, least count is p/N. With a 0.5 mm pitch and 50 divisions, least count is 0.01 mm. Use the ratchet to apply consistent force. Check for zero error and backlash. Measure at multiple positions and orientations if the specimen may be non-uniform or oval.

An electronic balance should be level, stable and protected from drafts and vibration. Allow warm-up where required, check zero, use suitable reference masses and tare the container correctly. Static charge, buoyancy, temperature gradients and magnetic forces can affect small mass measurements. A stopwatch is limited by human reaction time. Timing N cycles and calculating T=t/N reduces the effect of start-stop timing uncertainty on the period, provided the cycle count is correct and the process remains stable.

Worked example — micrometer.

A micrometer has a 0.5 mm pitch and 50 divisions. Sleeve reading is 7.5 mm and thimble division 23 aligns. Find the indication before zero correction.

Solution. Least count=0.5/50=0.01 mm. Thimble contribution=23×0.01=0.23 mm. Indication=7.50+0.23=7.73 mm. Apply the zero correction and assess specimen variation by repeating at suitable positions.

Electrical measurement: multimeters, oscilloscopes and transducers

A voltmeter is connected in parallel across a component; an ammeter is connected in series with the branch. Select the correct function, terminal and range before connecting. Begin with a safe range and observe polarity for DC. Never place an ordinary ammeter directly across a voltage source. Resistance should generally be measured with the circuit de-energised and isolated from parallel paths because the ohmmeter supplies its own test current. Capacitors may retain charge after power is removed.

An oscilloscope displays voltage against time. The vertical scale is volts per division and the horizontal scale is seconds per division. The period is measured between equivalent phase points and frequency is f=1/T. Peak-to-peak voltage is the difference between maximum and minimum. For a sinusoid centred on zero, V_rms=V_peak/√2; this formula does not apply to every waveform. Probe attenuation, bandwidth, input impedance, grounding and sample rate affect the measurement.

Conventional bench oscilloscope ground clips may be connected to protective earth. Connecting the clip to a point that is not at earth potential can create a short circuit. Use a suitable differential probe or isolated measurement arrangement when required. Never defeat protective grounding. A transducer converts a physical quantity into a signal: thermocouples produce a voltage related to temperature difference, strain gauges change resistance with deformation, photodiodes produce current in response to light and Hall sensors respond to magnetic field.

Laboratory experiment: measuring gravitational acceleration with a pendulum

For a simple pendulum of length L and bob mass m, the restoring torque at small angular displacement θ is approximately −mgLθ. The moment of inertia about the pivot is mL². Rotational dynamics gives mL²θ¨=−mgLθ, or θ¨+(g/L)θ=0. This is simple harmonic motion with angular frequency ω=√(g/L), so T=2π√(L/g). Squaring gives T²=(4π²/g)L. A graph of T² against L has gradient 4π²/g and therefore g=4π²/gradient.

Measure length from the pivot to the centre of mass of the bob. Displace the bob by a small angle and release without pushing. Time a large number N of complete oscillations and calculate T=t/N. Repeat the timing and measure several lengths. Keep the amplitude small and approximately consistent. Record raw time, cycle count, length, instrument resolution and environmental observations.

Limitations include finite amplitude, air resistance, pivot friction, string mass and stretch, uncertainty in the pivot location and reaction time. The small-angle approximation uses sinθ≈θ in radians. At larger amplitudes, the period increases slightly. The bob should be compact and dense so that its moment of inertia and air resistance are small compared with the ideal point-mass model.

Bob, mass mStringLθSupportEquilibrium line
Figure: pendulum geometry; effective length is measured to the bob's centre of mass.
Worked example — pendulum gradient.

A T² against L graph has gradient 4.02 s² m⁻¹. Calculate g.

Solution. g=4π²/4.02=9.82 m s⁻². If gradient uncertainty is 0.08 s² m⁻¹, then u(g)/g=u(m)/m=0.08/4.02=0.0199, giving u(g)=0.20 m s⁻². Report approximately (9.82 ± 0.20) m s⁻² as a standard-uncertainty result.

Laboratory experiment: Ohm's law and resistance measurement

For a conductor under constant physical conditions, especially constant temperature, Ohm's law states V=IR. Connect a DC supply, variable resistor, ammeter in series and voltmeter across the test resistor. Vary the voltage in controlled steps and record both voltage and current. A graph of V against I has gradient R; a graph of I against V has gradient 1/R. State the axes before interpreting the gradient.

The resistor dissipates power P=VI=I²R=V²/R. Heating can change resistance. A filament lamp becomes hotter as current increases and is not ohmic over a wide range. A diode has a strongly nonlinear current-voltage characteristic. Curvature may therefore be a real physical effect rather than poor measurement. Allow thermal equilibrium where appropriate and monitor temperature.

Two-wire resistance measurements include lead and contact resistance. A four-wire Kelvin arrangement separates current-carrying leads from voltage-sensing leads and reduces the influence of lead resistance. Contact pressure, oxidation and thermal electromotive forces can still matter. For high resistance, insulation leakage and instrument input resistance become important.

ARVDC supplyResistorVoltmeter branch
Figure: ammeter in series and voltmeter in parallel across the resistor.
Worked example — resistance from gradient.

A V-I graph has gradient 47.8 V A⁻¹ with standard uncertainty 0.6 V A⁻¹. Find R.

Solution. R=47.8 Ω and u(R)=0.6 Ω. The relative uncertainty is 0.6/47.8=1.26%. Report R=(47.8 ± 0.6) Ω, and inspect residuals and temperature stability before concluding that the conductor is ohmic.

Laboratory experiment: Newton's law of cooling

Newton's law of cooling approximates the rate of temperature change as proportional to the temperature difference from the surroundings. Let θ=T−T_s. The model is dθ/dt=−kθ. Separating variables gives dθ/θ=−k dt. Integration yields ln θ=−kt+C. Applying θ(0)=θ₀ gives θ=θ₀e^(−kt), or T−T_s=(T₀−T_s)e^(−kt). A graph of ln(T−T_s) against time has gradient −k.

The time constant is τ=1/k; after one time constant the excess temperature falls to 1/e of its initial value. The half-life is t₁/₂=ln2/k. The model assumes a reasonably constant surrounding temperature and heat-transfer coefficient, with the body close to uniform temperature. Radiation, evaporation and changing convection can produce deviations. Use a sensor with suitable response time, monitor surroundings and avoid drafts and direct sunlight.

Worked example — cooling constant.

The temperature difference falls from 60 K to 30 K in 240 s. Determine k and τ.

Solution. 30/60=e^(−240k), so k=ln2/240=2.89×10⁻³ s⁻¹. The time constant τ=1/k≈346 s. The result assumes the surroundings and heat-transfer conditions remain sufficiently constant.

Data acquisition, digitisation and signal conditioning

A data-acquisition system converts an analogue signal into numerical samples. A sensor produces a voltage, current or resistance; signal conditioning may amplify, filter, isolate or linearise it; an analogue-to-digital converter samples and quantises the signal. The result depends on sensor bandwidth, amplifier noise, filter response, sampling rate, converter resolution and reference stability.

An n-bit converter has 2ⁿ levels. For a full-scale range V_FS, the ideal quantisation step is approximately q=V_FS/2ⁿ. A 12-bit converter over 0–10 V has q≈10/4096=2.44 mV. This is nominal resolution, not total accuracy. Offset, gain error, nonlinearity, noise and reference uncertainty may dominate. Oversampling and averaging can improve effective resolution under suitable noise conditions, but cannot correct systematic calibration bias.

The Nyquist condition requires sampling above twice the highest frequency of a band-limited signal. Practical reconstruction requires a higher rate and an analogue anti-alias filter. Filtering can suppress noise but may also alter transients and phase. Report the filter type, cut-off, sampling rate, acquisition duration, trigger and software processing. Preserve unfiltered raw data wherever possible.

Worked example — converter.

A 10-bit converter spans 0–5.00 V and the signal contains frequencies up to 400 Hz. Find the step and ideal minimum sampling rate.

Solution. q=5.00/1024=4.88 mV. The ideal Nyquist lower limit is greater than 800 samples s⁻¹. A practical system should sample faster and apply an anti-alias filter with a suitable transition band.

Laboratory records, reproducibility and safety

A laboratory notebook is a primary scientific record. Record date, aim, apparatus identifiers, calibration status, procedure, environmental conditions, raw observations, units, anomalies, calculations and decisions. Write observations when they occur. Preserve raw data separately from corrected or processed values. If a transcription error is found, retain a traceable record of the original and document the correction. Electronic files should include column headings, units, timestamps, instrument settings and software versions.

Reproducibility requires enough detail for another competent investigator to repeat the work. State sensor placement, sampling rate, filtering, calibration procedure, number of repeats, exclusion rules and fitting method. Do not fabricate or selectively omit measurements to make a graph appear linear. Document exclusions and distinguish planned analysis from exploratory analysis. Research integrity includes accurate attribution and transparent limitations.

Before work begins, identify hazards and select controls. Hazards include electrical energy, hot surfaces, moving machinery, lasers, glassware, pressure systems, chemicals and radiation. Use the hierarchy of controls: eliminate, substitute, engineer, administer and then use personal protective equipment. De-energise circuits before changing connections unless an authorised live measurement is essential. Capacitors may retain charge. Never defeat protective grounding.

Laser work requires wavelength-appropriate protection, beam stops, controlled beam height and avoidance of specular reflections. Hot objects should be handled with suitable tools. Inspect glassware, secure apparatus and use rated pressure or vacuum equipment. Stop work if there is unexpected heating, noise, smell, leakage or instability. Know emergency isolation and first-aid arrangements and follow the institution's specific procedures.

Advanced uncertainty: Monte Carlo propagation and comparison of results

First-order analytical propagation is convenient when the model is smooth and uncertainties are small. For a strongly nonlinear model, asymmetric output or constrained inputs, Monte Carlo propagation may be more suitable. Assign justified probability distributions to inputs, generate many input sets, evaluate the model for each set and inspect the resulting output distribution. Summarise it using a mean or median, standard deviation and selected quantiles. The method does not rescue a poor model or unjustified input distributions.

For correlated inputs, generate samples from a joint distribution with the required covariance. Check that the covariance matrix is valid and that sampled values respect physical constraints. A diameter cannot be negative; a probability must lie between zero and one. Increase the number of trials and check convergence of the reported statistics. Record the random seed if exact reproducibility is required. Compare the simulation with analytical propagation for simple cases.

To compare independent measurements x₁ and x₂, calculate Δ=x₁−x₂ and u(Δ)=√(u₁²+u₂²). The normalised difference z=Δ/u(Δ) expresses the discrepancy in combined standard uncertainty units. For correlated estimates, u²(Δ)=u₁²+u₂²−2cov(x₁,x₂). Confirm that the two results refer to the same measurand and conditions before judging agreement.

Worked example — compare independent results.

Two measurements are 9.72 ± 0.10 and 9.91 ± 0.12, with standard uncertainties. Calculate z.

Solution. Δ=−0.19. u(Δ)=√(0.10²+0.12²)=0.156. Thus z=−1.22. The difference is about 1.2 combined standard uncertainties. This is not unusually large under a normal independent-error model, but it does not rule out shared systematic effects.

Scientific reporting, laboratory report structure and examination questions

A report should make the reasoning auditable. The abstract summarises aim, method, result with uncertainty and conclusion. The introduction defines the physical model and derives the equations. The apparatus and method describe instruments, calibration, connections, sampling, controls and safety. Results present raw and processed data with units, tables, graphs, fit parameters and representative calculations. The discussion interprets residuals, uncertainty contributions, model limitations and agreement with reference values. The conclusion answers the aim without overstating what the data show.

Tables need headings and units; graphs need labelled axes, sensible scales, captions and uncertainty bars where justified. State whether uncertainties are standard or expanded and identify the coverage factor. Preserve original data. A report should explain the dominant uncertainty sources rather than list generic phrases such as “human error”. References should identify textbooks, standards, instrument manuals and other sources used.

University examination questions

  1. Define a measurand and explain why operational definition matters.

    A measurand is the quantity intended to be measured. Its definition specifies object, conditions, location and relevant state. It makes results comparable and reproducible; two laboratories may otherwise measure different quantities under the same name.

  2. Distinguish resolution, sensitivity, accuracy and precision.

    Resolution is the smallest discernible input change; sensitivity is output change per input change; accuracy is closeness to a reference; precision is closeness among repeated results. High resolution does not guarantee accuracy.

  3. Derive the standard uncertainty of the mean.

    For independent observations with variance σ², Var(x̄)=Var[(ΣXᵢ)/n]=nσ²/n²=σ²/n. Therefore the standard deviation of the mean is σ/√n; estimate σ by sample standard deviation s to obtain s/√n.

  4. Why does repetition not remove a constant zero error?

    A constant offset shifts every observation by the same amount. Averaging reduces random scatter but preserves the offset. Check zero, apply a correction or recalibrate and include the correction uncertainty.

  5. Derive the rectangular resolution contribution q/√12.

    For a uniform rounding error between −q/2 and +q/2, variance=(1/q)∫e²de=q²/12. The standard uncertainty is q/√12. The result relies on a rectangular distribution.

  6. State the conditions for combining uncertainties by root-sum-of-squares.

    Contributions must be standard uncertainties and independent, or covariance must be negligible. Correlated quantities require covariance terms; expanded uncertainties must be converted to standard uncertainties before combination.

  7. Derive first-order propagation for y=f(x).

    Taylor expansion gives δy≈(df/dx)δx. Thus u(y)=|df/dx|u(x). For multiple independent inputs, square sensitivity-weighted contributions and add variances. The approximation requires small uncertainties and a locally smooth model.

  8. For y=a²/b, derive relative uncertainty.

    Logarithmic differentiation gives dy/y=2da/a−db/b. For independent inputs, [u(y)/y]²=4[u(a)/a]²+[u(b)/b]².

  9. Explain traceability and why it does not guarantee accuracy.

    Traceability is a documented chain of calibrations to a reference, with uncertainty stated at each stage. It does not ensure a correct measurand, suitable use, representative sample or valid model.

  10. Derive the least-squares gradient for a straight line.

    Minimise Σ(yᵢ−mxᵢ−c)². The c derivative gives c=ȳ−m x̄. Substitution into the m derivative gives m=Σ(xᵢ−x̄)(yᵢ−ȳ)/Σ(xᵢ−x̄)².

  11. What does reduced chi-square much greater than one suggest?

    Residual scatter exceeds that expected from stated uncertainties. Investigate underestimated uncertainty, model inadequacy, correlation, outliers and environmental variation using residual plots and the procedure record.

  12. Explain aliasing and prevention.

    Aliasing makes a sampled signal appear at an incorrect lower frequency when sampling is insufficient. Sample above twice the highest frequency and use an analogue anti-alias filter; practical rates should exceed the ideal minimum.

  13. Derive the small-angle pendulum period.

    Torque is approximately −mgLθ and moment of inertia is mL². Thus θ¨+(g/L)θ=0, ω=√(g/L) and T=2π√(L/g). The approximation requires small angles in radians.

  14. Discuss limitations of Newton's law of cooling.

    The heat-transfer coefficient may vary, radiation is nonlinear in absolute temperature, evaporation may contribute and the body may not be isothermal. Test the model over the measured interval using residuals.

  15. Describe defensible treatment of an anomalous observation.

    Preserve the raw value, inspect apparatus and record, identify a specific cause and document any exclusion. Never delete a point solely because it disagrees with theory; report the effect of the decision where relevant.

  16. Why is a voltmeter parallel and an ammeter series?

    A voltmeter measures potential difference across a component and has high input resistance. An ammeter measures branch current and has low resistance. Real instruments load the circuit and their input characteristics matter.

  17. Calculate the ideal step of a 0–10 V 12-bit converter.

    There are 4096 levels, so q≈10/4096=2.44 mV. Reference uncertainty, offset, gain error, nonlinearity and noise are additional contributions.

  18. Distinguish a confidence interval from expanded uncertainty.

    A confidence interval is a statistical procedure with long-run coverage for a parameter. Expanded uncertainty is combined standard uncertainty multiplied by a coverage factor under a measurement uncertainty model. They may be similar numerically but are not automatically interchangeable.

  19. Design an experiment to determine wire resistivity.

    Measure length and diameter at multiple positions, determine resistance with four-wire connections where possible, monitor temperature and plot resistance against length. For uniform cross-section, gradient is ρ/A. Propagate uncertainties in gradient and area and assess contact, heating and non-uniformity.

  20. Why do uncertainty bars not replace residual analysis?

    Bars describe estimated measurement dispersion; residuals reveal systematic model departures such as curvature, changing variance or time trends. Both are required to assess fit quality.

  21. Explain the role of a laboratory notebook.

    It preserves contemporaneous raw data, settings, methods, conditions, anomalies and decisions. It enables audit and reproduction. Raw data remain unchanged and corrections or exclusions are traceable.

Revision summary, key equations and laboratory checklist

Measurement begins with a defined measurand and a documented model. SI units and dimensional analysis maintain consistency. Instrument range, resolution, sensitivity, hysteresis, drift and input impedance determine suitability. Calibration establishes the relationship between indications and references; traceability documents the calibration chain. Repeated observations estimate random variation, while Type B information captures specifications, certificates and resolution.

Independent standard uncertainties combine in quadrature; correlated quantities require covariance. First-order propagation uses partial derivatives as sensitivity coefficients. For products and powers, relative uncertainty is often convenient; for sums and differences, absolute uncertainty is usually more useful. Nonlinear models may require Monte Carlo propagation. Graphs, least-squares fitting, weighted fitting, chi-square and residual plots help analyse data, but physical reasoning remains essential.

Key equations
x̄=Σxᵢ/n
s=√[Σ(xᵢ−x̄)²/(n−1)]
u(x̄)=s/√n
u_resolution=q/√12
u_c=√Σuᵢ² (independent inputs)
u²(y)=Σ[(∂f/∂xᵢ)u(xᵢ)]² (independent inputs)
U=k u_c
m=Σ(xᵢ−x̄)(yᵢ−ȳ)/Σ(xᵢ−x̄)²
χ²ᵣ=χ²/ν
T_pendulum=2π√(L/g)
V=IR
T−T_s=(T₀−T_s)e^(−kt)
q_ADC≈V_FS/2ⁿ

CBE Nexus · Undergraduate Physics · Unit 6 of 36 · Educational resource. Check the assumptions and conventions stated in each derivation before applying equations to a different physical system.

Official Education References

Use these official organisations for curriculum guidance, examinations and further learning information.