Lab Methods Domain Guide

Uncertainties add in quadrature, not in a straight line, because independent errors are as likely to cancel as to reinforce. That is why a small uncertainty combined with a large one barely matters: $\sqrt{10^{2}+3^{2}} = 10.4$, so the $3$ contributes almost nothing. The practical lesson is to improve the dominant term and ignore the rest, which is also how experimentalists decide where to spend their effort.

Counting experiments carry their own error bar for free. If a detector records $N$ counts, the uncertainty is $\sqrt{N}$ with no repeat runs needed, so $100$ counts is a 10% measurement and $10{,}000$ counts is a 1% measurement. Because the relative error falls as $1/\sqrt{N}$, halving it costs four times the counting time — a diminishing return that the exam likes to test as a ratio question.

Averaging is a treatment for one disease only. It shrinks random scatter as $1/\sqrt{N}$ and leaves systematic offsets untouched, so a beautifully precise result can still be badly wrong. Whenever a question offers 'take more measurements' as a remedy, first ask whether the error described is random or systematic — that single distinction answers a surprising number of items outright.

ETS lists laboratory methods among the content areas of the GRE Physics Test, and our learning model groups error analysis and instrumentation beneath it. Treat this block as guaranteed marks: unlike quantum mechanics or electromagnetism, it asks you to apply a handful of rules rather than to recall a body of theory, and the rules fit on one page.

Start with the propagation rules, because most items reduce to them. For a sum or difference $q = a \pm b$, absolute uncertainties combine in quadrature: $\delta q = \sqrt{(\delta a)^{2}+(\delta b)^{2}}$. For a product or quotient $q = ab$ or $a/b$, relative uncertainties combine the same way: $\dfrac{\delta q}{q} = \sqrt{\left(\dfrac{\delta a}{a}\right)^{2}+\left(\dfrac{\delta b}{b}\right)^{2}}$. For a power $q = a^{n}$, the relative uncertainty multiplies: $\dfrac{\delta q}{q} = |n|\dfrac{\delta a}{a}$, so a squared quantity doubles its relative uncertainty and a square root halves it. Note the asymmetry that trips people up: a difference of two nearly equal numbers keeps their absolute uncertainty while shrinking the value, so its relative uncertainty can be enormous — which is why subtracting a large background from a large signal is a poor experimental design.

Counting statistics is the second block. Radioactive decays, photon arrivals, and cosmic-ray events are independent random events, so their counts follow a Poisson distribution whose standard deviation is the square root of the mean. That gives $\sigma_N = \sqrt{N}$ for $N$ recorded counts and a relative uncertainty of $1/\sqrt{N}$. Two consequences are tested repeatedly: the uncertainty on a rate $R = N/t$ is $\sqrt{N}/t$, and improving the relative precision by a factor $k$ requires $k^{2}$ times the counting time. Background must be measured separately and subtracted, and its own uncertainty propagates into the net result by the sum-and-difference rule.

Averaging and fitting turn many measurements into one number. The mean of $N$ readings has a standard error $\sigma/\sqrt{N}$, where $\sigma$ is the spread of the individual readings — so the scatter of the data does not shrink, only the uncertainty in their centre does. When two quantities are related non-linearly, linearise before fitting: plotting $T^{2}$ against $L$ for a pendulum gives a straight line of slope $4\pi^{2}/g$, and plotting $\ln N$ against $t$ for a decay gives a straight line of slope $-\lambda$. The exam favours this because it converts a physics question into a question about a slope or an intercept, and because a non-zero intercept where theory predicts zero is the classic fingerprint of a systematic error.

Instrumentation is mostly about not disturbing what you measure. An ammeter goes in series and must have very low resistance so it does not throttle the current it reads; a voltmeter goes in parallel and must have very high resistance so it does not divert current from the element it reads across. Getting either backwards is not a small error: an ammeter placed in parallel across a source is a short circuit. An oscilloscope has a high input impedance and displays voltage against time, so a period is read from the horizontal scale and an amplitude from the vertical one. A Wheatstone bridge measures resistance by nulling — adjusting a known resistance until no current flows through the detector — which is intrinsically more precise than a deflection measurement because at balance the detector's own properties do not matter.

Detectors and dimensional reasoning finish the domain. Geiger counters, scintillators with photomultipliers, and semiconductor detectors all convert radiation into countable pulses and all suffer dead time, during which a detector cannot register a second event — so the observed rate underestimates the true rate at high intensities. Energy resolution describes how well a detector separates two nearby energies. Finally, dimensional analysis is the cheapest error check available anywhere on the paper: an expression whose units do not reduce correctly is wrong regardless of how plausible it looks, and on a multiple-choice exam a units check alone can often eliminate three of five options without doing any physics at all.