Home / Learn / Exam prep / GATE Physics / Lesson 1.10
All lessons

Module 1: Mathematical physics · Lesson 1.10 · about 60 min

Probability and statistics for physicists

Rules of probability, conditional probability and Bayes' theorem, counting (including identical particles), the binomial, Poisson, Gaussian, uniform and exponential distributions, moments and characteristic functions, random walks and the central limit theorem, counting statistics, propagation of errors, weighted means, least squares and chi-square.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 1 · about 60 minutes. By the end you will be able to use conditional probability and Bayes’ theorem, count configurations of distinguishable and identical particles, work with the binomial, Poisson, Gaussian, uniform and exponential distributions and know when one approximates another, handle random walks and the central limit theorem, and treat data the way an experimental physicist does: counting errors, error propagation, weighted means, least-squares lines and chi-square. This lesson closes the mathematical physics module and prepares the ground for statistical mechanics.

Physics meets probability twice: in nature (radioactive decay, thermal motion, quantum measurement) and in the laboratory (every measurement has a spread). The same few distributions cover almost everything, and the same rule, that variances of independent quantities add, explains √N.

1. Rules of probability

For events A and B: P(A∪B)=P(A)+P(B)−P(A∩B) and P(A∩B)=P(A|B)P(B). They are independent if P(A∩B)=P(A)P(B), i.e. P(A|B)=P(A). Mutually exclusive events (P(A∩B)=0) are the opposite of independent: if one happens, the other cannot.

Bayes’ theorem turns P(data|cause) into P(cause|data):

P(C|D)=P(D|C)P(C)P(D|C)P(C)+P(D|C‾)P(C‾)

Worked example. 1% of samples are contaminated; a test catches 99% of contaminated ones and falsely flags 5% of clean ones. P(C|+)=0.99×0.010.99×0.01+0.05×0.99=16. Most alarms are false, because clean samples are 99 times more common.

Counting. Arrangements of N distinguishable objects: N!; choices of k from N: (Nk). For particles in g single-particle states: distinguishable gN; identical bosons (N+g−1N) (any occupation); identical fermions (gN) (at most one per state). Three bosons in three states: 10 configurations; distinguishable: 27; fermions: 1. These counts become Bose–Einstein and Fermi–Dirac statistics. For large N, Stirling: ln⁡N!≈Nln⁡N−N (the full form N!≈2πNNNe−N is good to about 1/12N).

2. Mean, variance and moments

For a discrete variable ⟨x⟩=∑xiPi; for a continuous one ⟨x⟩=∫xp(x)dx. The variance σ2=⟨x2⟩−⟨x⟩2. Rules that hold always: ⟨aX+bY⟩=a⟨X⟩+b⟨Y⟩, and

Var(X+Y)=VarX+VarY+2Cov(X,Y)

with Cov=0 for independent variables (the converse is false: Y=X2 with symmetric X has zero covariance). So for N independent copies the variance of the sum is Nσ2 and the standard error of the mean is σ/N.

The characteristic function ϕ(k)=⟨eikx⟩ is the Fourier transform of p(x); ϕ(0)=1, moments are its derivatives at 0, and for a sum of independent variables the ϕs multiply.

σ (standard deviation)
the typical size of the spread about the mean; the square root of the variance
Standard error
the spread of a mean of N values, σ/√N
Poisson process
independent rare events at a constant average rate, such as decays or photon arrivals
Weight 1/σ²
how much a measurement counts in an average; twice as precise means four times the weight
χ²/ν
the average squared misfit per degree of freedom, in units of the error bars; about 1 for a good fit
Physical meaning
Fluctuations shrink relative to the mean as 1/√N. That is why a gas of 10²³ molecules has a sharp pressure, and why longer counting gives better rates.

3. The distributions you need

distribution formula mean variance where it appears
binomial (nk)pk(1−p)n−k np np(1−p) n yes/no trials, spins up or down
Poisson λke−λk! λ λ counts of rare independent events
Gaussian 1σ2πe−(x−μ)2/2σ2 μ σ2 sums of many small effects, measurement errors
uniform on [a,b] 1b−a a+b2 (b−a)212 rounding errors, random phases
exponential λe−λt 1/λ 1/λ2 waiting times, lifetimes

Facts examiners like:

  • Gaussian: 68.3% within 1σ, 95.4% within 2σ, 99.7% within 3σ; one tail beyond 2σ holds 2.3%. FWHM =22ln⁡2σ≈2.355σ. ⟨x4⟩=3σ4, ⟨|x−μ|⟩=σ2/π. With fixed mean and variance it has the largest entropy.
  • Poisson: P(0)=e−λ, P(k+1)/P(k)=λ/(k+1) (so the mode is ⌊λ⌋), sums of independent Poissons are Poisson, and detecting each event with probability ε gives Poisson with mean ελ.
  • Exponential: no memory; the median (half-life) is τln⁡2, below the mean τ; the probability of decaying between t1 and t2 is e−t1/τ−e−t2/τ.

Limits. Binomial with n→∞, p→0, np=λ fixed: Poisson. Binomial or Poisson with large mean: Gaussian with the same mean and variance. The figure lets you watch both limits happen.

The binomial distribution (bars) against a Poisson distribution with the same mean (dots) and a Gaussian with the same mean and variance (curve). Small p favours Poisson; large n with moderate p favours the Gaussian.

Worked example. Two independent sources give 2 and 3 counts per second on average. P(exactly 1 count in 1 s)=5e−5=0.034, because the total is Poisson with mean 5.

4. Random walks and the central limit theorem

A walker takes N independent steps of length L, right with probability p. Each step has mean L(2p−1) and variance 4p(1−p)L2, so after N steps

⟨x⟩=NL(2p−1),σx=2LNp(1−p)

For a fair walk, ⟨x⟩=0 but xrms=LN: 100 steps of 1 m spread only 10 m. With N=t/Δt this is diffusion, ⟨x2⟩=2Dt, and the distribution of x becomes the heat kernel of lesson 1.8.

The central limit theorem says this is general: the sum of many independent, identically distributed variables with finite variance is approximately Gaussian, whatever the individual distribution. (A Lorentzian has no variance and is the classic exception: averaging Lorentzian data does not narrow it.)

5. Treating data

Counting. N counts have σ=N, so 1% precision needs 104 counts. For a net rate G/tG−B/tB, variances add even though you subtract: σ=G/tG2+B/tB2.

Propagation of independent errors.

σf2=∑i(∂f∂xi)2σi2,f=xayb⇒σff=(aσxx)2+(bσyy)2

Worked example. x=2.00±0.02, y=4.0±0.08, f=x2/y=1: σff=(2×1%)2+(2%)2=2.8%. fibub’s Uncertainty propagation calculator does this for any formula.

Weighted mean. Measurements xi±σi combine with weights wi=1/σi2: x‾=∑wixi∑wi, σx‾=(∑wi)−1/2. 10.0±0.1 and 10.6±0.2 give 10.12±0.09.

Least squares and chi-square. For y=a+bx with equal errors, minimising ∑(yi−a−bxi)2 gives

b=∑(xi−x‾)(yi−y‾)∑(xi−x‾)2,a=y‾−bx‾

so the line passes through (x‾,y‾). With known errors, χ2=∑(yi−f(xi)σi)2; for ν degrees of freedom (points minus fitted parameters) a good fit has χ2/ν≈1. Much larger means a bad model or underestimated errors; much smaller means overestimated errors.

“Subtracting two measurements cancels their errors.” The values subtract; the uncertainties do not. For independent quantities, Var(X−Y)=VarX+VarY, so a small difference of two large, noisy numbers can be very uncertain. That is why weak signals on a big background need long counting times, and why 400−100 is the wrong error for a net count of 300.

  • Independence: P(A∩B)=P(A)P(B). Bayes: P(C|D)∝P(D|C)P(C); rare causes give mostly false alarms.
  • Bosons (N+g−1N), fermions (gN), distinguishable gN. ln⁡N!≈Nln⁡N−N.
  • Binomial np, np(1−p); Poisson λ, λ; uniform (b−a)2/12; exponential mean 1/λ, median ln⁡2/λ; Gaussian 68/95/99.7, FWHM 2.355σ, ⟨x4⟩=3σ4.
  • Variances of independent quantities add: σ/N, LN, G+B. Biased walk: σ=2LNpq.
  • Errors in quadrature with partial derivatives; powers multiply relative errors. Weighted mean with 1/σ2. Least-squares line through (x‾,y‾); χ2/ν≈1.

Practise: exam questions on this lesson

From warm-up to top-rank level, with full solutions. More on the probability practice page.

Try the Binomial lab, the Normal lab and the Counting statistics helper. Before this: lesson 1.9, tensors. Next: Part B begins with Lagrangian mechanics; meanwhile, quantum measurement probabilities are in lesson 4.1. Back to the GATE Physics course home.

Going on to CSIR NET? Random variables and probability distributions and measurement, errors and fitting data take these ideas into Part C.

Check yourself

1A detector records 2500 counts. What is the fractional statistical uncertainty?
2What is the variance of the number of heads in 100 tosses of a fair coin?
3You subtract a background count B from a gross count G. What is the uncertainty of G − B?

Sources

  • P. R. Bevington and D. K. Robinson, Data Reduction and Error Analysis for the Physical Sciences, 3rd ed. (2003), chapters 1 to 6
  • J. R. Taylor, An Introduction to Error Analysis, 2nd ed. (1997), chapters 3 to 8
  • F. Reif, Fundamentals of Statistical and Thermal Physics (1965), chapter 1
  • M. L. Boas, Mathematical Methods in the Physical Sciences, 3rd ed. (2006), chapter 15
  • GATE Physics (PH) syllabus, Mathematical Physics section, published by the organising institute each year

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Not affiliated with the IITs, IISc or the GATE organising institute. The general aptitude part of GATE is not covered here. Check the official GATE website for the current pattern. All questions are original.