University Physics / Undergraduate Physics / Unit 16
Learning outcomes and scope
Students should design defensible experiments, distinguish calibration from adjustment and verification, establish traceability, evaluate Type A and Type B uncertainty, propagate uncertainty including covariance, fit and validate physical models, and select suitable methods for electrical, optical, thermal, mechanical, vacuum and radiation measurements. They should build complete uncertainty budgets, preserve raw data and communicate results in a reproducible scientific report.
Core constants: k_B=1.380649×10⁻²³ J K⁻¹; q=1.602176634×10⁻¹⁹ C; c=299 792 458 m s⁻¹; h=6.62607015×10⁻³⁴ J s. The numerical values of these defining SI constants are exact; experimental realisations and practical measurements still carry uncertainty.
Contents
- The philosophy of experimental physics
- Measurement standards and metrological traceability
- Calibration, adjustment and verification
- Random and systematic effects
- Type A and Type B uncertainty evaluation
- Propagation of uncertainty and covariance
- Regression, fitting and residual analysis
- Electrical measurement methods
- Impedance and frequency-domain measurements
- Oscilloscope and time-domain metrology
- Optical measurement and photometry
- Interferometry and precision displacement
- Thermal measurements and calorimetry
- Mechanical measurements and dynamic experiments
- Vacuum, pressure and low-temperature measurement
- Photon counting and radiation detection
- Data acquisition and sampling design
- Experimental design, repeatability and reproducibility
- Laboratory quality systems and uncertainty budgets
- Scientific reporting, ethics and data integrity
- Advanced experimental methods and model validation
- Planning an advanced research experiment
- Worked numerical examples
- Laboratory investigations
- Examination questions
1. The philosophy of experimental physics
Experimental physics connects mathematical models with observations of nature. A measurement is not simply reading a number from an instrument: it is an operational procedure that defines a measurand, establishes a measurement model, realises a scale, controls influences and assigns uncertainty. The experimenter must decide what physical quantity is intended, under what conditions it is defined, and how the instrument's response is related to it. A clear measurement procedure makes results reproducible by another laboratory.
An experiment tests a hypothesis or estimates a parameter. A hypothesis should lead to observable consequences and a method capable of distinguishing competing explanations. Good experimental design identifies independent, dependent and controlled variables; defines a range and resolution; estimates expected signal size; and considers confounding effects before apparatus is assembled. Pilot measurements can reveal saturation, drift, poor signal-to-noise ratio or an unsuitable sampling rate.
The result should be reported with units, uncertainty, conditions, method and traceability information. Significant figures communicate numerical resolution but do not replace uncertainty analysis. A result such as (2.43 ± 0.05) m is incomplete unless the uncertainty type, coverage convention and measurement conditions are stated. A scientifically useful report also discusses model limitations and whether the data support the original hypothesis.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
2. Measurement standards and metrological traceability
Metrology is the science of measurement and its application. The International System of Units provides coherent definitions for base and derived units. Since the 2019 SI revision, all seven base units are defined by fixing exact values of defining constants. Practical realisations use physical experiments and calibrated instruments. A unit definition, a realisation of the unit and a working standard are distinct concepts.
Traceability is a property of a measurement result whereby it can be related to a reference through a documented, unbroken chain of calibrations, each contributing to measurement uncertainty. Traceability is not guaranteed merely because an instrument carries a calibration sticker. The calibration certificate, reference standards, methods, environmental conditions, uncertainty statements and validity interval must be relevant to the intended measurement.
A calibration hierarchy may run from an international or national metrology institute through accredited calibration laboratories to a working instrument. Intermediate checks, reference materials and control charts help detect changes between formal calibrations. A laboratory should define acceptance criteria, identify critical instruments and prevent equipment outside calibration from being used for measurements that require traceability.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
3. Calibration, adjustment and verification
Calibration establishes the relationship between indications and reference values under specified conditions. It may produce a correction table, calibration curve or mathematical model. Adjustment changes an instrument so that its indications better match the reference; verification provides objective evidence that specified requirements are fulfilled. These activities are related but not interchangeable. A calibration may show that an instrument is unsuitable without any adjustment being performed.
A calibration experiment should cover the operating range and include enough points to identify nonlinearity. Reference values should have uncertainty substantially smaller than the intended tolerance. Measurements may be repeated at each point, and both increasing and decreasing sequences can reveal hysteresis. A fitted line y=a+bx should be examined through residuals, not accepted merely because the correlation coefficient is close to one.
After adjustment, repeat the calibration and retain both as-found and as-left results. Document instrument identity, serial number, firmware, range, resolution, reference standard, environmental conditions, operator, date, method, corrections and uncertainty. A calibration interval should be justified by stability history, use intensity, environment, consequences of error and intermediate check results rather than selected arbitrarily.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
4. Random and systematic effects
Random effects cause repeated indications to vary unpredictably under nominally identical conditions. They may arise from thermal fluctuations, electronic noise, finite counting statistics, vibration or observer variability. Repetition helps estimate their dispersion, but the sample standard deviation is not automatically the uncertainty of the reported mean. For n independent observations, the standard uncertainty of the mean is s/√n, subject to the assumptions behind the sampling model.
Systematic effects shift results in a consistent direction or according to a reproducible pattern. Examples include zero offset, scale-factor error, misalignment, calibration bias, thermal expansion, parallax and an incorrect physical model. Repeating a measurement does not eliminate a fixed systematic effect. Corrections should be applied when known, while uncertainty remains to represent incomplete knowledge of the correction.
An experiment should separate observed scatter from possible bias. Randomisation, reversal, substitution, null methods, independent reference measurements and control experiments can reveal systematic effects. A residual trend with time or temperature indicates that the assumed constant conditions may not hold. Reporting only a standard deviation while ignoring a known calibration bias gives a misleading impression of accuracy.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
5. Type A and Type B uncertainty evaluation
Type A evaluation uses statistical analysis of a series of observations. For n independent measurements x_i, the mean is x̄=Σx_i/n and sample standard deviation is s=[Σ(x_i−x̄)²/(n−1)]¹ᐟ². The standard uncertainty of the mean is u_A=s/√n. The degrees of freedom and assumptions about independence, stationarity and distribution should be considered, especially for small samples or correlated time series.
Type B evaluation uses information other than repeated statistical observations, such as calibration certificates, manufacturer specifications, resolution, prior data, physical bounds and expert knowledge. If a digital display rounds to the nearest increment d, a rectangular model over ±d/2 gives standard uncertainty d/√12. A certificate may state expanded uncertainty U with coverage factor k; under the stated assumptions, standard uncertainty is u=U/k.
Type A and Type B describe methods of evaluation, not categories of error. Both produce standard uncertainties that can be combined when the measurement model and dependence assumptions permit. The uncertainty budget should identify each input quantity, estimate, standard uncertainty, probability distribution, sensitivity coefficient and degrees of freedom where relevant.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
6. Propagation of uncertainty and covariance
If the result y depends on input quantities x_i through y=f(x_1,…,x_n), first-order propagation for uncorrelated inputs gives u_c²(y)=Σ[c_i u(x_i)]², where c_i=∂f/∂x_i is the sensitivity coefficient evaluated at the estimates. This is a local linear approximation. It is appropriate when uncertainties are small relative to scales over which the function changes significantly.
For correlated inputs, covariance terms must be included: u_c²(y)=Σ_iΣ_j c_i c_j u(x_i,x_j). Positive correlation can increase or decrease the result uncertainty depending on the signs of sensitivity coefficients. Ignoring correlation is not automatically conservative. Shared calibration standards, common temperature corrections and common timing references often create correlations between measured quantities.
For strongly nonlinear models, asymmetric distributions or large relative uncertainties, Monte Carlo propagation may be more appropriate. Sample input values from their probability distributions, evaluate the model repeatedly and analyse the output distribution. Report the assumptions, number of trials, random seed or reproducibility method, and how coverage intervals are obtained. Numerical propagation does not repair a poor measurement model.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
7. Regression, fitting and residual analysis
Many experiments infer physical parameters by fitting a model to measured data. Ordinary least squares minimises the sum of squared vertical residuals when uncertainties in the independent variable are negligible and response errors are independent with equal variance. Weighted least squares uses weights commonly proportional to 1/u_i² when point uncertainties differ. If both axes have significant uncertainty, an errors-in-variables method may be required.
A fitted parameter has uncertainty and parameters can be correlated. The covariance matrix from a fit describes local parameter uncertainty under the model assumptions. Residuals r_i=y_i−f(x_i) should be plotted against independent variable, time and fitted value. Curvature, periodic structure, changing variance or long runs of one sign may indicate model inadequacy, drift or underestimated uncertainty.
A high coefficient of determination does not prove that a model is physically correct. Compare candidate models using residuals, parameter plausibility, uncertainty, independent validation and physical constraints. Avoid fitting high-order polynomials merely to pass through data. Report the fitted equation, parameter uncertainties, units, weighting method, number of points and fit diagnostics.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
8. Electrical measurement methods
Electrical experiments measure voltage, current, resistance, impedance, charge, frequency and power. A voltmeter should have sufficiently high input impedance to minimise loading; an ammeter should have sufficiently low burden voltage and be inserted in series. An ohmmeter supplies its own test current and must not be connected to an energised circuit. Instrument range, bandwidth, common-mode limits, isolation and input protection matter.
Four-wire Kelvin measurement separates current-carrying leads from voltage-sensing leads, reducing the influence of lead and contact resistance. Bridge methods compare an unknown with known standards and can provide high sensitivity near balance. Null methods often reduce dependence on instrument scale calibration because the measurement is made near zero deflection, although reference values and detector sensitivity remain important.
For AC quantities, specify waveform, frequency, amplitude convention and impedance. True-RMS instruments can measure non-sinusoidal waveforms within their bandwidth and crest-factor limits; average-responding meters calibrated for sine waves may be inaccurate on distorted signals. Power in AC circuits depends on phase and waveform; apparent, real and reactive power must not be confused.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
9. Impedance and frequency-domain measurements
Impedance is the complex ratio Z(ω)=V(ω)/I(ω) under sinusoidal steady-state conditions. It contains resistive and reactive components and depends on frequency. A network analyser or impedance analyser measures magnitude and phase across a frequency range. The instrument and fixture must be calibrated or compensated for open, short and load conditions, especially when parasitic capacitance and inductance are comparable to the device under test.
A frequency response experiment applies a known sinusoidal input and measures the output amplitude and phase. The transfer function H(jω)=V_out/V_in can be displayed as Bode magnitude and phase plots. Sweep resolution, averaging, source impedance, load impedance, cable delay and dynamic range influence the result. Measurements near resonance may require finer frequency spacing and careful control of drive amplitude.
Resonance can be characterised by resonant frequency, bandwidth and quality factor Q=f₀/Δf for a lightly damped resonator under suitable definitions. The half-power bandwidth method assumes a clear isolated resonance and a response whose power relation is understood. For coupled or overlapping modes, fit an appropriate physical model rather than reading a single peak and treating it as an exact resonance.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
10. Oscilloscope and time-domain metrology
An oscilloscope measures voltage as a function of time. Its finite analogue bandwidth, sample rate, vertical resolution, timebase accuracy, trigger jitter, input impedance and probe characteristics limit measurement quality. The displayed waveform is a processed representation of sampled data, not a perfect view of the physical signal. Probe attenuation and compensation must be set correctly, and the ground connection must be safe.
Rise time is affected by both the signal and instrument. For approximately Gaussian responses, 1/t_r,system²≈1/t_r,signal²+1/t_r,instrument². This relation helps estimate the instrument contribution when the measured rise time is not much slower than the scope response. A probe with excessive capacitance can slow a fast circuit and alter the phenomenon being measured.
Time interval measurements require attention to trigger level, noise, interpolation, clock accuracy and jitter. Averaging reduces uncorrelated noise but may conceal non-repetitive events. Peak detection can reveal narrow glitches but may increase apparent noise. Record bandwidth limits, sample rate, acquisition mode, probe type and measurement statistics so that another researcher can reproduce the measurement.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
11. Optical measurement and photometry
Optical experiments measure intensity, wavelength, polarisation, phase, reflectance, transmittance and image position. A photodetector converts incident optical power into an electrical signal; its responsivity depends on wavelength. A spectrometer separates wavelengths using a dispersive or interferometric element and detects the spectrum. Wavelength calibration uses known spectral lines, while intensity calibration requires a suitable radiometric reference and correction for system response.
Optical alignment affects throughput and systematic uncertainty. Apertures, lenses, fibres and detector active area define acceptance angles and spatial response. Stray light, dark current, detector nonlinearity, saturation, polarisation sensitivity and ambient light can bias results. Measure a dark signal and appropriate reference signal, and document slit width, integration time, grating, detector temperature and optical geometry.
In photometry, distinguish radiometric quantities such as watts from photometric quantities weighted by human visual response. In spectroscopy, distinguish wavelength resolution from wavelength accuracy. A narrow observed line may be limited by the instrument line-spread function rather than the source. Deconvolution can be unstable and should be supported by a measured instrument response.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
12. Interferometry and precision displacement
Interferometers use superposition of coherent waves to measure optical path differences. In a Michelson interferometer, moving one mirror by distance Δx changes round-trip optical path by 2Δx, producing approximately 2Δx/λ fringe cycles. Thus a count of N fringes corresponds to displacement Δx=Nλ/2 for normal incidence and ideal geometry. The wavelength in the medium, alignment and fringe-counting method must be specified.
Interferometry can achieve very high displacement sensitivity, but the result is sensitive to air refractive index, temperature, pressure, humidity, vibration, wavefront quality and alignment. A vacuum wavelength differs from wavelength in air. For precision work, correct the refractive index or control environmental conditions. Cosine error arises when the measurement axis is misaligned with the displacement direction.
Fringe visibility provides information about coherence, polarisation matching and beam overlap. Phase-shifting methods estimate fractional fringe phase and can improve resolution beyond simple fringe counting, but require stable phase steps and careful treatment of nonlinearity. Report the optical configuration, reference wavelength, environmental corrections, phase algorithm and uncertainty.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
13. Thermal measurements and calorimetry
Thermal experiments measure temperature, heat capacity, thermal conductivity, diffusivity, expansion and phase transitions. A thermometer has finite response time, calibration uncertainty, self-heating or stem-conduction effects. Contact sensors require thermal equilibrium with the sample; infrared methods depend on emissivity, reflected background radiation, viewing angle and spectral response.
Calorimetry estimates heat exchange from temperature changes and a model of the calorimeter. A simple energy balance may include sample heat, container heat, heater input and heat lost to surroundings. The calorimeter constant must be measured or calibrated. For phase transitions, latent heat is inferred from energy supplied during a temperature plateau, with corrections for heat loss and sensor lag.
Thermal conductivity measurements are sensitive to geometry, contact resistance, heat losses and boundary conditions. Guarded hot-plate and comparative methods reduce some errors by controlling heat flow. Report sample dimensions, density, moisture state, temperature range, contact method and steady-state criterion. A result should not be presented as an intrinsic material constant unless the sample condition and measurement model justify that interpretation.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
14. Mechanical measurements and dynamic experiments
Mechanical experiments measure length, displacement, mass, force, torque, pressure, acceleration, strain and vibration. Calibration and alignment are crucial for micrometers, load cells, accelerometers and pressure sensors. A load cell's output may depend on temperature, loading history, mounting and transverse force. Dead-weight calibration provides a traceable force reference when local gravity and mass uncertainty are considered.
Dynamic measurements introduce bandwidth, phase lag, damping and sampling requirements. An accelerometer mounted with poor mechanical coupling may report fixture resonance rather than the object's motion. Strain gauges require bridge completion, temperature compensation and controlled adhesive installation. Force and displacement channels must be time-synchronised when calculating work, power or hysteresis loops.
A dynamic system can be characterised by its frequency response, impulse response, damping ratio and natural frequencies. Excitation amplitude should remain within the linear range if linear models are used. Windowing, leakage and spectral resolution matter in Fourier analysis. Report mounting, sensor orientation, sampling rate, anti-alias filter and any signal processing applied.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
15. Vacuum, pressure and low-temperature measurement
Vacuum systems are characterised by pressure, leak rate, pumping speed and residual gas composition. Pressure gauges operate on different physical principles and have different ranges: mechanical gauges, capacitance manometers, Pirani gauges and ionisation gauges are not interchangeable. Gas composition, contamination, outgassing and gauge location influence readings. A gauge should be calibrated for the gas and pressure range of interest.
Leak detection can use pressure-rise tests, tracer gases or residual-gas analysis. A pressure-rise method estimates leak and outgassing contributions together unless the system is properly isolated and background effects are characterised. Pumping speed at the pump inlet differs from effective speed at the chamber because conductance of connecting lines limits gas flow.
Low-temperature experiments require calibrated thermometers, thermal anchoring, controlled heat leaks and attention to sensor self-heating. Temperature gradients and thermal contraction can dominate uncertainty. Sensor calibration may be nonlinear and range-specific. Cryogenic systems introduce hazards from cold surfaces, pressure buildup and oxygen displacement; use appropriate ventilation, pressure relief and training.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
16. Photon counting and radiation detection
Photon and particle detectors convert individual interactions into electrical pulses or charge. A photomultiplier tube uses photoemission and electron multiplication; a semiconductor detector creates electron–hole pairs; scintillators convert deposited energy into light. Detection efficiency depends on energy, geometry, material, threshold and dead time. Background counts must be measured under comparable conditions.
For independent Poisson counts N accumulated over time t, the standard deviation is approximately √N and the rate is R=N/t with standard uncertainty √N/t. Background subtraction combines uncertainties from sample and background counts. If dead time is significant, the observed count rate underestimates the true rate and a suitable detector model is required.
Energy spectroscopy requires calibration of channel number against known energies, resolution analysis and peak identification. Efficiency calibration must account for source geometry, self-absorption and energy dependence. Radiation experiments require appropriate authorisation, shielding, monitoring and ALARA principles. Never improvise a source or detector arrangement outside institutional radiation-safety controls.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
17. Data acquisition and sampling design
A data-acquisition system includes sensors, analogue conditioning, ADC, clock, computer interface, software and storage. Sampling frequency should be chosen from the signal bandwidth and experiment dynamics, not simply from the maximum rate advertised by the converter. Anti-alias filtering must precede sampling. Channel-to-channel timing skew matters when measuring phase or transient relationships.
ADC resolution determines code spacing but not the complete measurement uncertainty. Reference voltage accuracy, gain and offset error, integral nonlinearity, differential nonlinearity, noise, missing codes, aperture jitter and temperature drift contribute. Multiplexed systems may require settling time after switching channels. The analogue input range should be used efficiently without clipping expected peaks or transients.
Software should retain raw data, metadata, timestamps, instrument configuration, calibration coefficients and processing history. File formats should be open or documented, with checksums and version control for critical datasets. Automated scripts should be tested with known inputs and boundary cases. A reproducible workflow separates raw observations from processed results and records every transformation.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
18. Experimental design, repeatability and reproducibility
Repeatability concerns agreement under the same method, operator, equipment, location and short time interval. Reproducibility concerns agreement under changed conditions such as laboratory, operator or instrument. Both are important but answer different questions. A method may be repeatable while systematically biased, so repeatability alone does not establish accuracy.
Experimental design should include sufficient replication, randomised run order and blocking where nuisance variables are known. Factorial designs can identify interactions between variables more efficiently than changing one factor at a time. Control samples and reference measurements reveal drift. Predefine exclusion rules and acceptance criteria to avoid selectively removing inconvenient observations.
A reproducible experiment documents apparatus, calibration, sample preparation, environmental conditions, software versions, analysis code and deviations from protocol. Share data in a form that preserves units, metadata and provenance. Independent replication is especially important when results depend on subtle corrections or complex fitting. Report failed runs and limitations when they materially affect interpretation.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
19. Laboratory quality systems and uncertainty budgets
A laboratory quality system defines document control, equipment management, staff competence, method validation, sample handling, data integrity, corrective action and internal audits. ISO/IEC 17025 is a widely used standard for competence of testing and calibration laboratories. Accreditation applies to a defined scope and does not mean every possible measurement by the organisation is accredited.
An uncertainty budget is a structured accounting of contributions to the measurement result. Each row should specify source, estimate, distribution, standard uncertainty, sensitivity coefficient, contribution to combined uncertainty and degrees of freedom where appropriate. The budget should include significant influences, not just convenient instrument specifications. Sensitivity analysis helps identify which contributions dominate and where improvements are worthwhile.
Expanded uncertainty U=k u_c is reported with a coverage factor and stated coverage interpretation. A factor near 2 often corresponds to approximately 95% coverage for near-normal distributions with sufficient degrees of freedom, but this is not universal. Small samples may require an effective degrees-of-freedom calculation and Student's t coverage factor. Avoid claiming a confidence level without a justified distributional model.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
20. Scientific reporting, ethics and data integrity
A scientific report should explain the question, theory, apparatus, method, results, uncertainty, interpretation and limitations. Tables require headings and units; graphs require labelled axes, scales, legends and uncertainty information where meaningful. Equations should define symbols and assumptions. Raw observations should be preserved, while corrected and derived values should be clearly distinguished.
Data integrity requires accurate recording, secure storage, traceable changes and honest reporting. Do not fabricate, selectively suppress or manipulate measurements to match an expected result. Outliers should be investigated using pre-established criteria and documented reasoning. Image processing, smoothing, baseline subtraction and filtering must be disclosed when they could influence interpretation.
Ethical practice includes safe work, responsible use of resources, proper authorship, acknowledgement of contributions and transparent conflict-of-interest statements. Negative results can be scientifically valuable. A report should distinguish what was observed from what is inferred, and should not claim causal conclusions from a design that only establishes correlation.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
21. Advanced experimental methods and model validation
Advanced experiments often combine multiple sensors and physical models. A coupled measurement model may include thermal expansion, electrical resistance, optical path length and environmental corrections. The full model should be derived before data collection, and the sensitivity of the final result to each input should be evaluated. Independent measurements of key correction terms can reduce reliance on assumed constants.
Model validation compares predictions with observations across conditions not used to tune the model. Residuals should be examined for structure and scale. Parameter identifiability asks whether the available data can uniquely constrain model parameters; strongly correlated parameters may produce apparently good fits but large uncertainties. Design experiments to maximise information, for example by extending the range or selecting conditions where candidate models differ.
When model discrepancy is important, include it in the uncertainty assessment or explicitly report that the uncertainty only represents measurement inputs conditional on the model. A precise numerical fit cannot compensate for omitted physics. Sensitivity studies, reference experiments and independent techniques provide stronger validation than repeatedly fitting the same dataset.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
22. Planning an advanced research experiment
A research plan begins with a focused question and a review of relevant physical theory and prior measurement methods. Translate the question into a measurable quantity, define an operational measurand and identify the expected signal, noise and timescale. Select a measurement principle based on sensitivity, dynamic range, selectivity, invasiveness, cost and safety. Consider whether an independent method can validate the result.
Before construction, prepare a risk assessment, equipment list, calibration plan, uncertainty budget, data-management plan and analysis strategy. Use simulations or analytical estimates to identify required tolerances. Design fixtures to control alignment, temperature, vibration and electromagnetic interference. Pilot the method and update the plan based on observed limitations while recording changes.
The final research record should enable another competent investigator to reconstruct the setup and analysis. Include instrument settings, calibration records, raw files, scripts, software versions, sample identifiers and environmental logs. State what remains uncertain and what additional experiment would most effectively reduce that uncertainty. A good experiment is not defined by complexity but by the quality of the evidence it produces.
Experimental workflow. Define the measurand and measurement equation before collecting data. List every input quantity and identify which are measured, corrected, controlled or assumed. Select instruments with adequate range, resolution, bandwidth and uncertainty. Check calibration status and establish a traceability route. Plan repeats, reference checks and control measurements. Record raw observations with units and timestamps, then apply corrections transparently. Evaluate uncertainty and inspect residuals before drawing a physical conclusion.
Critical evaluation. Ask whether the instrument perturbs the system, whether the assumed physical model is valid over the measured range, whether the sampling interval resolves the relevant dynamics and whether environmental conditions remain stable. Consider the difference between instrument specification and actual performance in the experiment. Manufacturer accuracy limits are not automatically standard uncertainties; their meaning and distribution must be interpreted. A result should be no more precise than the evidence supports.
Worked numerical examples
Calculations use standard uncertainty unless stated otherwise. Uncertainty contributions are combined only under the stated independence and model assumptions.
Example 1. Mean and uncertainty of repeated readings
Readings 10.2, 10.4, 10.3, 10.1 and 10.5 mm give mean 10.30 mm. The sample standard deviation is 0.158 mm, so standard uncertainty of the mean is 0.158/√5=0.071 mm. Report 10.30 mm with u_A≈0.07 mm, while considering calibration and resolution contributions separately.
Example 2. Resolution contribution
A digital instrument displays in 0.01 V increments. Rounding lies within ±0.005 V. For a rectangular distribution, u=0.005/√3=0.00289 V, equivalently 0.01/√12.
Example 3. Expanded uncertainty
A calibration certificate states U=0.12 °C with k=2. The corresponding standard uncertainty is u=0.06 °C, assuming the stated coverage factor and model are applicable.
Example 4. Uncertainty in a sum
For L=a+b with independent u(a)=0.2 cm and u(b)=0.3 cm, u(L)=√(0.2²+0.3²)=0.36 cm.
Example 5. Uncertainty in a product
For A=xy with independent relative uncertainties 1.0% and 2.0%, relative standard uncertainty is √(0.01²+0.02²)=2.24%.
Example 6. Four-wire resistance
A sample has 0.250 V across its sensed terminals when 50.0 mA flows through current leads. R=V/I=5.00 Ω. Kelvin sensing excludes most lead/contact voltage drops from the measured voltage.
Example 7. Interferometer displacement
A Michelson interferometer at λ=632.8 nm records 120 fringes during mirror motion. Δx=Nλ/2=120×632.8 nm/2=37.97 μm.
Example 8. Poisson counting uncertainty
A detector records 400 counts in 20 s. Rate is 20 s⁻¹ and standard uncertainty is √400/20=1 s⁻¹, assuming negligible background and dead-time effects.
Example 9. RC rise-time correction
Measured rise time is 12 ns and instrument rise time 8 ns. Estimated signal rise time is √(12²−8²)=8.94 ns under the Gaussian-response approximation.
Example 10. Thermal calibration fit
A linear calibration gives T=(a+bV). If b=25.0 °C V⁻¹ and voltage standard uncertainty is 2.0 mV, the voltage contribution is 25×0.002=0.050 °C, before coefficient and reference uncertainties.
Example 11. Weighted mean
Two independent estimates are 5.0±0.2 and 5.6±0.4. Inverse-variance weights are 25 and 6.25. Weighted mean=(25×5+6.25×5.6)/31.25=5.12; standard uncertainty=1/√31.25=0.179.
Example 12. ADC quantisation
A 16-bit converter spans 0–5.000 V. Δ=5/65536=76.3 μV; rectangular quantisation standard uncertainty is 22.0 μV, excluding other ADC errors.
Example 13. Uncertainty of a period-derived frequency
For f=1/T and small uncertainty, u(f)/f≈u(T)/T. If T=2.000±0.004 s, f=0.5000 Hz and u(f)=0.0010 Hz.
Example 14. Thermal expansion correction
A 1.000 m steel scale with α=11.5×10⁻⁶ K⁻¹ changes length by ΔL=αLΔT=0.230 mm for a 20 K temperature rise. The correction is significant for sub-millimetre work.
Example 15. Background-subtracted counts
Sample and background measurements yield 900 and 100 counts over equal times. Net count is 800; independent Poisson standard uncertainty is √(900+100)=31.6 counts.
Example 16. Power uncertainty
For P=VI, V=12.0±0.1 V and I=2.00±0.02 A independent, P=24.0 W and relative uncertainty √[(0.1/12)²+(0.02/2)²]=1.18%, giving u(P)≈0.28 W.
Advanced laboratory investigations
1. Calibration of a digital voltmeter
Use a calibrated voltage source or reference calibrator with uncertainty suitably below the required instrument tolerance. Select points across each intended range, including zero, intermediate values and near full scale. Approach points from increasing and decreasing directions to test hysteresis. Record reference value, indication, correction, repeatability and environmental conditions. Fit a correction function only when justified by residuals. Verify selected points independently and report as-found and as-left performance.
2. Four-terminal resistance and bridge comparison
Measure a low-resistance specimen using separate current and potential leads. Reverse current to estimate and cancel thermoelectric EMFs where appropriate. Compare the result with a bridge or calibrated reference resistor. Control specimen temperature and current to avoid self-heating. Repeat after remounting to assess contact and positioning effects. Include reference resistor uncertainty, current stability, voltage resolution, temperature coefficient and repeatability in the uncertainty budget.
3. Oscilloscope timebase and rise-time evaluation
Apply a stable reference waveform with known frequency and rise time. Measure period over many cycles to reduce cursor quantisation, compare the timebase with a traceable frequency reference and evaluate trigger jitter. Measure a fast edge using different probe configurations and bandwidth limits. Estimate instrument rise-time contribution using the appropriate response model. Record sample rate, bandwidth, interpolation mode, acquisition length and probe compensation.
4. Michelson interferometer displacement measurement
Align the two arms to obtain high-contrast fringes. Use a stabilised laser with known wavelength and a calibrated translation stage. Move the mirror through a measured displacement while counting fringes or recording phase. Repeat in both directions to test backlash and hysteresis. Correct for alignment cosine error and, where required, refractive index of air. Include wavelength, fringe count, phase estimation, stage calibration, environmental conditions and vibration in the uncertainty analysis.
5. Spectrometer wavelength calibration
Record several known spectral lines spanning the intended wavelength interval. Fit a wavelength-versus-pixel or wavelength-versus-angle calibration relation and inspect residuals for curvature. Determine repeatability by repeated scans and evaluate slit-width effects on line position. Measure a reference line not used in the fit to validate the calibration. Report spectral resolution separately from wavelength accuracy and include instrument response when estimating line widths.
6. Calorimeter heat-capacity experiment
Calibrate the calorimeter using a known electrical energy input or a well-characterised mixing procedure. Measure sample mass, initial and final temperatures, and container response. Estimate heat exchange with surroundings using a cooling curve or an appropriate correction model. Repeat at different energy inputs to test linearity. Include thermometer calibration, mass, timing, calorimeter constant, heat loss and thermal equilibration in the uncertainty budget.
7. Photon-counting and background subtraction
Measure background counts with the same detector settings and counting interval as the sample. Repeat the sample and background runs to test stability. Check detector dead time and pulse threshold, and verify that the count rate is within the detector's linear range. Apply background subtraction and combine independent counting uncertainties. For energy spectra, calibrate channels with reference peaks and evaluate resolution. Follow institutional radiation authorisation, shielding, monitoring and handling rules at all times.
8. Data acquisition validation
Feed known DC levels, sine waves and pulses into the acquisition chain. Verify gain, offset, linearity, sample timing and channel synchronisation. Test alias rejection by applying controlled out-of-band signals. Compare digitised data with a calibrated reference instrument and inspect raw codes for clipping or missing values. Validate software using synthetic datasets with known answers and retain configuration files, firmware versions and analysis scripts.
Reporting and safety
Prepare a complete apparatus diagram, calibration record, raw-data table, analysis, residual plots, uncertainty budget and discussion. Identify all corrections and state whether uncertainties are standard or expanded. Use isolated low-voltage supplies where possible, observe laser eye-safety controls, secure optical components, manage cryogenic and vacuum hazards, and follow approved radiation procedures. Safety controls are part of experimental design, not an appendix added after the experiment.
Essential equations and reporting checklist
Mean: x̄=Σxᵢ/n; sample standard deviation: s=[Σ(xᵢ−x̄)²/(n−1)]¹ᐟ²
Type A standard uncertainty of mean: u_A=s/√n
Rectangular resolution contribution: u=d/√12
Expanded uncertainty: U=k u_c
Uncorrelated propagation: u_c²(y)=Σ(∂f/∂xᵢ)²u²(xᵢ)
Correlated propagation: u_c²(y)=ΣᵢΣⱼ cᵢcⱼu(xᵢ,xⱼ)
Weighted mean: x̄_w=Σwᵢxᵢ/Σwᵢ, with wᵢ=1/uᵢ²
Michelson displacement: Δx=Nλ/2
Poisson count standard deviation: σ_N≈√N
Rise-time correction: 1/t_r,sys²≈1/t_r,meas²−1/t_r,inst²
Resistance: R=V/I; AC impedance: Z(ω)=V(ω)/I(ω)
Frequency from period: f=1/T; u(f)/f≈u(T)/T for small uncertainty
- Measurand and measurement model clearly defined.
- Instrument identity, range, resolution and calibration recorded.
- Reference traceability and environmental conditions documented.
- Raw data preserved separately from corrected and derived values.
- Random and systematic influences considered.
- Uncertainty budget includes sensitivity coefficients and correlations.
- Fit residuals and model limitations examined.
- Results reported with units, uncertainty and coverage convention.
Common experimental mistakes
- Calling a calibration sticker proof of traceability without checking the certificate and scope.
- Confusing Type A/Type B methods with random/systematic error categories.
- Dividing instrument accuracy limits by √3 without establishing a rectangular distribution.
- Combining correlated inputs as if they were independent.
- Reporting expanded uncertainty without a coverage factor or interpretation.
- Using a high R² value as the only test of a physical model.
- Ignoring instrument loading, bandwidth, timing jitter and environmental effects.
- Subtracting background counts without combining their uncertainties.
- Failing to preserve raw data and analysis provenance.
- Presenting a fitted line without residuals, parameter units or uncertainty.
University examination questions with model answers
1. Distinguish calibration, adjustment and verification.
Model answer: Calibration establishes indication-to-reference relationships; adjustment changes the instrument; verification confirms specified requirements are met.
2. Explain metrological traceability.
Model answer: It is a documented unbroken chain of calibrations to a stated reference, with each link contributing uncertainty.
3. Compare Type A and Type B evaluation.
Model answer: Type A uses statistical analysis of observations; Type B uses other information such as certificates, resolution and specifications. Both yield standard uncertainties.
4. Why does repeating a measurement not remove systematic error?
Model answer: A stable bias shifts repeated observations in the same way. Repetition estimates scatter but does not identify or correct the bias.
5. Derive standard uncertainty of the mean.
Model answer: For n independent observations with sample standard deviation s, standard uncertainty of the mean is s/√n.
6. Explain uncertainty propagation for a product.
Model answer: For independent x and y, relative variance of xy is approximately the sum of squared relative standard uncertainties.
7. When are covariance terms required?
Model answer: When input quantities are correlated, for example through shared standards or common environmental corrections.
8. What do residual plots reveal?
Model answer: Patterns reveal model inadequacy, drift, heteroscedasticity, outliers or unmodelled physical effects.
9. Explain four-terminal resistance measurement.
Model answer: Separate current leads establish current while high-impedance voltage leads sense potential directly at the specimen, reducing lead resistance influence.
10. Why must AC meter waveform capability be stated?
Model answer: Average-responding sine-calibrated meters may misread distorted waveforms; true-RMS performance is limited by bandwidth and crest factor.
11. Describe oscilloscope probe loading.
Model answer: Probe resistance and capacitance load the circuit, potentially changing amplitude, bandwidth and rise time.
12. Derive displacement per fringe for Michelson interferometry.
Model answer: Mirror motion Δx changes round-trip path by 2Δx; one fringe corresponds to path change λ, so Δx=λ/2.
13. List key optical spectrometer calibration issues.
Model answer: Wavelength scale, line-spread function, stray light, detector response, slit width and saturation.
14. What is calorimeter constant?
Model answer: The heat capacity of the calorimeter assembly, determined by calibration and included in the energy balance.
15. Explain dead time in radiation counting.
Model answer: A detector is temporarily unable to register a subsequent event after a pulse; observed rate may be lower than true rate.
16. Why is anti-alias filtering required?
Model answer: Frequency components above Nyquist can fold into the sampled band and become indistinguishable from lower frequencies.
17. What is repeatability versus reproducibility?
Model answer: Repeatability holds conditions largely fixed; reproducibility changes specified conditions such as lab, operator or equipment.
18. Why is uncertainty budget useful?
Model answer: It identifies contributions and dominant sources, guides improvement and makes the reported result transparent.
19. What does expanded uncertainty mean?
Model answer: It is combined standard uncertainty multiplied by a stated coverage factor, with coverage interpretation dependent on distribution and degrees of freedom.
20. How should outliers be treated?
Model answer: Investigate physical and procedural causes, apply predefined justified criteria, document decisions and never remove data solely because it disagrees.
21. Explain model validation.
Model answer: Compare model predictions with independent data and inspect residuals, parameter plausibility and behaviour across conditions.
22. What information belongs in raw-data metadata?
Model answer: Units, timestamps, instrument identity and settings, calibration, sample ID, environmental conditions, operator and processing provenance.
23. Why does temperature matter in precision experiments?
Model answer: It changes dimensions, resistance, refractive index, sensor response, electronic offsets and material properties.
24. State the purpose of a control measurement.
Model answer: It monitors background, drift, contamination or system response independently of the primary sample measurement.
25. What is the role of ISO/IEC 17025?
Model answer: It specifies competence requirements for testing and calibration laboratories, including technical and management requirements.
Advanced uncertainty analysis: correlated measurements, Monte Carlo methods and decision rules
1. Covariance in a differential measurement
Suppose a small voltage difference is determined from two readings made with the same voltmeter: V=V₂−V₁. Its variance is u²(V)=u²(V₂)+u²(V₁)−2u(V₁,V₂). If the readings share a common zero offset, that offset is correlated and may cancel in the difference. Treating the two readings as independent would overestimate or underestimate the uncertainty depending on the actual covariance. This is why the measurement sequence and instrument model matter. A reversal method can deliberately expose offset terms: measure the specimen in both polarities, combine the readings to isolate the odd-in-current signal and use the even component to estimate thermal EMF or offset. The method only works when the offset remains sufficiently stable during the reversal cycle.
2. Sensitivity coefficients and dominant contributors
The sensitivity coefficient cᵢ=∂f/∂xᵢ quantifies how the reported result changes when an input changes. The contribution to combined variance is cᵢ²u²(xᵢ) for an independent input. A large input uncertainty does not necessarily dominate: its sensitivity may be small. Conversely, a small uncertainty in a highly influential input may dominate. Rank the variance contributions to identify whether improved calibration, better temperature control, more repetitions or a different measurement principle would most effectively improve the result. Avoid spending effort reducing a minor contribution while leaving the dominant source unchanged.
3. Effective degrees of freedom and coverage
When uncertainty components have finite degrees of freedom, the effective degrees of freedom of the combined standard uncertainty can be estimated with the Welch–Satterthwaite relation: ν_eff=u_c⁴/Σ(u_i⁴/ν_i), for independent components expressed as standard uncertainty contributions. A Student's t coverage factor can then be selected for the desired coverage probability and effective degrees of freedom. Components evaluated from certificates or assumed distributions may have effectively large degrees of freedom only when that assumption is justified. A coverage factor of two is a useful approximation in many circumstances, but not a universal guarantee of 95% coverage.
4. Monte Carlo propagation procedure
For a nonlinear model y=f(x₁,…,x_n), assign a probability distribution to each input using the available evidence. Generate a large set of input vectors, preserving correlations where known, evaluate the model for every vector and form the output distribution. Estimate the central value, standard uncertainty and coverage interval from the resulting samples. Check convergence by increasing the number of trials and comparing quantiles. Monte Carlo propagation is especially useful when the output distribution is asymmetric, bounded or multimodal. It does not remove the need to justify the input distributions, correlations or physical model.
5. Guarding against drift and time correlation
Repeated observations collected close together may be correlated because of thermal drift, slow environmental variation, instrument filtering or a slowly changing sample. In that case s/√n can be too small because the observations do not provide n independent pieces of information. Plot readings against time, inspect autocorrelation and compare blocks of measurements. Randomise the order of conditions, interleave reference readings and allow adequate settling time. If drift is unavoidable, include a time-dependent term in the model or use a measurement sequence that estimates drift independently.
6. Decision rules and conformity assessment
A measurement is often used to decide whether a device, material or process meets a tolerance. A decision rule defines how uncertainty is considered when making that statement. A simple acceptance rule compares the measured value directly with the tolerance limit, but this can increase the risk of accepting a nonconforming item near the boundary. Guard bands move the acceptance limit inward to account for measurement uncertainty and reduce false acceptance risk. The appropriate rule depends on the consequences of false acceptance and false rejection, the uncertainty, applicable standards and agreement between customer and laboratory. State the decision rule with the result rather than implying that uncertainty automatically determines pass or fail.
7. Validation of a complete measurement procedure
Validation demonstrates that a procedure is fit for its intended use. Relevant characteristics may include range, repeatability, intermediate precision, reproducibility, bias, linearity, detection limit, robustness and uncertainty. The required studies depend on the physical quantity and application. A method intended to resolve a tiny change needs a different validation emphasis from one intended to classify a sample across a broad range. Use reference materials, independent methods and deliberately varied conditions to test the assumptions that matter most. Record acceptance criteria before reviewing final performance where practical.
8. Research data stewardship and reproducible computation
Store raw files in read-only or versioned form and create derived datasets through documented scripts. Each file should carry a unique experiment identifier, timestamp, sample identity, units, instrument settings, calibration version and operator or automated system identity. Analysis scripts should state software and library versions, input filenames, parameter choices and processing steps. Preserve plots and tables generated from the same data pipeline. Backups should be tested by restoration, not merely assumed to exist. A reproducible analysis allows another researcher to regenerate reported values from the original observations without hidden manual edits.
Further study
Unit 17: Mathematical Physics develops the mathematical tools used to formulate, solve and interpret advanced physical models.
Continue to Unit 17: Mathematical Physics →