Basic Probability Distributions

Fundamental probability distributions in physics: Gaussian normal, Poisson, binomial, and exponential models.

What it is

A probability distribution is a rigorous mathematical model that maps probabilities to the possible outcomes of a physical measurement or the microstates of a system. Discrete variables (such as particle counts in a radiation detector) are modeled via a probability mass function $P(X = k)$; continuous variables (such as molecular velocities in a gas or photon energies) are formulated via a probability density function (PDF) $f(x)$, where the probability of observing a value in $[x, x + dx]$ is $dP = f(x) dx$.

In experimental and theoretical physics, four canonical distributions dominate:

1. Binomial Distribution: Models the number of successes $k$ in $N$ independent trials with fixed success probability $p$: $P(k; N, p) = \binom{N}{k} p^k (1 - p)^{N-k}$
2. Poisson Distribution: The limiting form of the binomial distribution as $N \to \infty$ and $p \to 0$ with constant mean rate $\lambda = N p$. It governs rare, independent discrete events in space or time (radioactive decay, photon arrivals on CCD pixels): $P(k; \lambda) = \frac{\lambda^k e^{-\lambda}}{k!}, \quad \mu = \lambda, \quad \sigma^2 = \lambda$
3. Normal (Gaussian) Distribution: The universal consequence of the Central Limit Theorem; arises whenever a physical observable represents the cumulative sum of numerous small, independent microscopic perturbations: $f(x; \mu, \sigma) = \frac{1}{\sigma \sqrt{2\pi}} \exp\left( -\frac{(x - \mu)^2}{2\sigma^2} \right)$
4. Exponential Distribution: Models waiting times between successive Poisson events (particle lifetimes, radioactive decay intervals): $f(t; \lambda) = \lambda e^{-\lambda t} \quad (t \geq 0), \quad \mu = \frac{1}{\lambda}, \quad \sigma = \frac{1}{\lambda}$

Why it appears in physics

Statistical mechanics, thermodynamics, quantum mechanics, and instrumentation are governed by probabilistic laws.

• In statistical physics, the Maxwell-Boltzmann molecular velocity distribution arises by maximizing entropy under particle and energy constraints, manifesting as a Gaussian for each Cartesian velocity component $v_x, v_y, v_z$.
• In nuclear and particle physics, radioactive emission from an unstable isotope is a memoryless stochastic process: counts recorded per second on a Geiger-Müller detector follow Poisson statistics with mathematical precision. Its defining property $\sigma = \sqrt{\mu}$ dictates that relative counting uncertainty scales as $\sigma_N / N = 1/\sqrt{N}$, dictating integration times required for target signal-to-noise ratios.
• In experimental metrology, the Central Limit Theorem proves that random noise from multiple uncorrelated sources converges asymptotically to a Gaussian profile, justifying standard deviation as the standard uncertainty metric.

Visual or geometric idea

• Binomial and Poisson: Discrete histograms plotted over non-negative integers $k = 0, 1, 2, \dots$. At low rates $\lambda < 5$, Poisson distributions exhibit pronounced positive skewness with extended right-hand tails. As $\lambda$ increases ($\lambda > 20$), skewness vanishes and the discrete bars align closely with the continuous envelope of a Gaussian curve with $\mu = \lambda$ and $\sigma = \sqrt{\lambda}$.
• Gaussian: A symmetric bell-shaped curve centered at mean $\mu$, possessing inflection points precisely at $\mu \pm \sigma$. The total integral equals 1; the cumulative areas within $[\mu - \sigma, \mu + \sigma]$, $[\mu - 2\sigma, \mu + 2\sigma]$, and $[\mu - 3\sigma, \mu + 3\sigma]$ cover $68.27\%$, $95.45\%$, and $99.73\%$ of the total probability, respectively.
• Exponential: A monotonically declining curve starting at $\lambda$ at $t=0$ and decaying asymptotically to zero. Its geometric memoryless property means that conditional tails $f(t + t_0 | t > t_0)$ duplicate the original shape under rescaling.

Minimal symbolic language

1. Normalization condition: $\sum_{k} P(k) = 1 \quad (\text{discrete}), \qquad \int_{-\infty}^\infty f(x) dx = 1 \quad (\text{continuous})$
2. Statistical moments (expectation and variance): $\mu = E[X] = \int_{-\infty}^\infty x f(x) dx$ $\sigma^2 = \operatorname{Var}(X) = E[(X - \mu)^2] = \int_{-\infty}^\infty (x - \mu)^2 f(x) dx = E[X^2] - \mu^2$
3. Cumulative distribution function (CDF): $F(x) = P(X \leq x) = \int_{-\infty}^x f(u) du$
4. Standard normal score ($z$-score): $z = \frac{x - \mu}{\sigma}, \qquad f(z) = \frac{1}{\sqrt{2\pi}} e^{-z^2/2}$

What the result means

Fitting empirical data to theoretical distributions enables rigorous interpretation:

• If radioactive decay counts exhibit sample variance $s^2 \gg \bar{k}$, counting statistics are not purely Poisson; an external noise source or instrument drift induces overdispersion.
• Appropriate distribution choices yield valid confidence intervals, rigorous hypothesis testing (such as $5\sigma$ statistical significance required in particle discovery at CERN, representing random false-alarm probability $p < 3 \times 10^{-7}$), and accurate sensor saturation estimates.

When not to use it

1. Never use Gaussian distributions for small discrete counts: If mean detection rate is $\lambda = 2$, a Gaussian PDF assigns finite probability to unphysical negative counts ($k < 0$). Poisson statistics must be preserved.
2. Never assume Poisson when events interact: If particle detection blinds the detector during dead-time periods, or triggers avalanche Townsend discharges, independence fails and Poisson models underestimate true variance.
3. Never apply Gaussian models to heavy-tailed or scale-free phenomena: Fully developed turbulence, cosmic-ray energy cascades, or self-organized criticality exhibit power-law distributions (Cauchy, Lévy, Pareto) where variances can diverge.

Operational summary

• Check variable domain: discrete counts vs continuous physical dimensions.
• For independent counting processes with small individual success rates over time/space, assign a Poisson distribution with $\sigma = \sqrt{\lambda}$.
• For continuous observables perturbed by numerous additive independent fluctuations, adopt the Gaussian model $\mathcal{N}(\mu, \sigma^2)$.
• For decay times or waiting times between Poisson events, employ the exponential density $f(t) = \lambda e^{-\lambda t}$.
• Apply normalization and moment calculations to determine probabilities, error bounds, and detection thresholds.

Procedure for Selecting and Applying Probability Distributions

Before calculating

Before starting probability calculations, characterize the physical process:

• Is the observable continuous (length, voltage, mass, temperature)?
• Is it a discrete counting sequence of non-negative integer events ($k = 0, 1, 2, \dots$)?
• Is it an elapsed continuous waiting time until an event occurs? Verify that the system is stationary and that successive observations are statistically independent.

Step 1: identify the mathematical object

Select the appropriate canonical model based on physical constraints:

• Rare independent counts within a temporal or spatial window: Poisson distribution with rate parameter $\lambda$.
• Continuous physical quantity perturbed by multiple random noises: Normal distribution $\mathcal{N}(\mu, \sigma^2)$.
• Repeated Bernoulli trials with binary outcomes (success/failure): Binomial distribution $\mathcal{B}(N, p)$.
• Decay time or lifespan of metastable states: Exponential distribution $\text{Exp}(\lambda)$.

Step 2: choose the operation

Clarify the required mathematical calculation:

• Discrete point probability: Evaluate probability mass function $P(X = k)$.
• Interval probability:
  • For discrete variables: finite sum $P(a \leq X \leq b) = \sum_{k=a}^b P(X = k)$.
  • For continuous densities: definite integral $P(a \leq X \leq b) = \int_a^b f(x) dx = F(b) - F(a)$.
• Counting uncertainty: For Poisson counts, the standard deviation is directly $\sigma = \sqrt{\lambda} \approx \sqrt{N_{\text{counts}}}$.

Step 3: preserve units and signs

• For discrete distributions, $P(X = k)$ is dimensionless and strictly bounded in $[0, 1]$.
• For continuous PDFs $f(x)$, physical units are the reciprocal of the random variable: $[f(x)] = [x]^{-1}$. Consequently, $f(x)$ can exceed 1 without violating probability conservation, provided the integral $\int f(x) dx = 1$ is preserved.
• The standard $z$-score $z = (x - \mu)/\sigma$ is dimensionless.

Step 4: calculate

• For Gaussian distributions, compute $z$-scores for boundaries and evaluate the standard normal CDF $\Phi(z)$ or error function $\operatorname{erf}$: $P(X \leq x) = \Phi\left( \frac{x - \mu}{\sigma} \right) = \frac{1}{2} \left[ 1 + \operatorname{erf}\left( \frac{x - \mu}{\sigma \sqrt{2}} \right) \right]$
• For Poisson processes with $\lambda > 20$, approximate via $\mathcal{N}(\lambda, \lambda)$ incorporating continuity correction: $P(X = k) \approx \int_{k - 0.5}^{k + 0.5} \frac{1}{\sqrt{2\pi \lambda}} \exp\left( -\frac{(u - \lambda)^2}{2\lambda} \right) du$

Step 5: interpret the result

• Report probabilities as percentages or scientific fractions.
• Interpret findings within the physical context: does this represent false-alarm probability exceeding detection thresholds? Does it define the $95\%$ calibration envelope?
• When measuring radioactive activity, report $N_{\text{counts}} \pm \sqrt{N_{\text{counts}}}$, demonstrating that relative uncertainty decreases as $1/\sqrt{N}$.

Self-check checklist

• Does total integrated probability equal 1?
• Are all computed probabilities strictly bounded in $[0, 1]$?
• Does the continuous probability density $f(x)$ carry dimensions $[x]^{-1}$?
• Did I restrict Poisson evaluations strictly to non-negative integers $k \in \mathbb{N}_0$?
• If approximating Poisson by Gaussian, is $\lambda \geq 20$ with half-step ($0.5$) continuity correction applied?

Guided example: Radioactive counting with a Geiger-Müller detector

Physical situation

A nuclear physics laboratory measures environmental background radiation using a Geiger-Müller tube. Over a 10-minute calibration run, exactly 120 ionization pulses are detected. A fast 1-minute test is subsequently carried out to evaluate whether an ore sample exhibits radioactive activity above background. We seek to:

1. Model the expected background rate in 1 minute and its statistical uncertainty.
2. Determine the probability of registering exactly 15 counts in 1 minute solely due to random background fluctuations.
3. Compute the count threshold per minute needed to declare positive radioactive detection at a $3\sigma$ significance level ($p < 0.0013$).

Mathematical translation

Nuclear decay and background ionization are rare, independent, discrete events occurring in continuous time, obeying Poisson statistics $P(k; \lambda) = \frac{\lambda^k e^{-\lambda}}{k!}$.

1. Mean rate per minute: $\lambda = \frac{120 \text{ counts}}{10 \text{ min}} = 12.0 \text{ counts/min}$ The Poisson variance is $\sigma^2 = \lambda$, yielding standard deviation: $\sigma = \sqrt{\lambda} = \sqrt{12.0} \approx 3.464 \text{ counts/min}$
2. Point probability for $k = 15$ with $\lambda = 12$: $P(15; 12) = \frac{12^{15} e^{-12}}{15!}$

Step-by-step calculation

1. Point probability calculation: $12^{15} \approx 1.540702 \times 10^{16}$ $e^{-12} \approx 6.144212 \times 10^{-6}$ $15! = 1\,307\,674\,368\,000 \approx 1.307674 \times 10^{12}$ $P(15; 12) = \frac{1.540702 \times 10^{16} \cdot 6.144212 \times 10^{-6}}{1.307674 \times 10^{12}} = \frac{9.4664 \times 10^{10}}{1.307674 \times 10^{12}} \approx 0.07239 \quad (7.24\%)$
2. $3\sigma$ detection threshold calculation: With $\lambda = 12$ high enough for quasi-Gaussian behavior, the one-tailed $3\sigma$ threshold is: $k_{\text{threshold}} = \lambda + 3 \sigma = 12.0 + 3(3.464) = 12.0 + 10.39 = 22.39$ Rounding up: observing $k \geq 23$ counts in 1 minute indicates that random background fluctuation probability is less than $0.13\%$.

Reading the result

Observing 15 counts in 1 minute represents an ordinary background fluctuation ($7.24\%$ point probability). Conversely, registering 23 or more counts in 1 minute provides $99.87\%$ statistical confidence that the sample emits radiation above background.

Common mistake avoided

Assigning an arbitrary instrument tolerance to counting rates. In Poisson counting, the fundamental floor of uncertainty is dictated by nature and equals exactly $\sqrt{N}$.

Inverse example: Particle mean lifetime from exponential decay

Physical situation

In a tracking detector, decay times of stopped cosmic-ray muons are measured: $\mu^- \to e^- + \bar{\nu}_e + \nu_\mu$. It is observed that $50\%$ of all muons decay within $t_{1/2} = 1.523 \,\mu\text{s}$ (experimental half-life).

Given result

Empirical median of decay times: $t_{\text{median}} = t_{1/2} = 1.523 \,\mu\text{s}$. The cumulative probability at this time is $P(T \leq t_{1/2}) = 0.50$.

Analytical reconstruction

Decay times follow an exponential probability density: $f(t) = \lambda e^{-\lambda t} \quad (t \geq 0)$ The cumulative distribution function is: $F(t) = \int_0^t \lambda e^{-\lambda u} du = 1 - e^{-\lambda t}$ Setting to $0.50$: $1 - e^{-\lambda t_{1/2}} = 0.50 \implies e^{-\lambda t_{1/2}} = 0.50$ Taking natural logarithms: $-\lambda t_{1/2} = \ln(0.50) = -\ln(2) \implies \lambda = \frac{\ln(2)}{t_{1/2}}$ The mean lifetime is the expectation value: $\tau = \mu_t = E[T] = \frac{1}{\lambda} = \frac{t_{1/2}}{\ln(2)}$ Substituting empirical data: $\tau = \frac{1.523 \,\mu\text{s}}{0.693147} \approx 2.197 \,\mu\text{s}$

What physical question it answers

What fraction of muons survives past $3\tau \approx 6.59 \,\mu\text{s}$? $P(T > 3\tau) = 1 - F(3\tau) = e^{-\lambda (3/\lambda)} = e^{-3} \approx 0.0498 \quad (4.98\%)$ Only roughly $5\%$ of muons remain undecayed after 3 mean lifetimes.

Limit verification

As $t \to 0$, $F(0) = 0$ (no particles decayed initially). As $t \to \infty$, $F(t) \to 1$ (all unstable states eventually decay). The standard deviation of the exponential distribution matches the mean: $\sigma_t = \tau = 2.197 \,\mu\text{s}$.

Common Mistakes in Fundamental Probability Distributions

1. Using Gaussian approximations for small counting rates

It seems reasonable because the Gaussian bell curve is familiar and readily integrated via built-in software functions. It fails because for $\lambda < 10$, Poisson distributions are noticeably skewed and physically bounded at zero ($k \geq 0$). A Gaussian model with $\mu = 2$ and $\sigma = \sqrt{2} \approx 1.414$ allocates roughly $8\%$ probability to impossible negative counts ($k < 0$). It is detected when reporting detection bounds based on Gaussian error functions when $\lambda < 15$. It is corrected by strictly using the exact Poisson mass function $P(k) = \lambda^k e^{-\lambda} / k!$ for small sample rates. Mini-example: for $\lambda = 1$, the exact Poisson probability of zero counts is $e^{-1} \approx 0.368$ ($36.8\%$), whereas a naive Gaussian model yields a distorted estimate near $24\%$.

2. Neglecting continuity correction when approximating discrete sums by continuous integrals

It seems reasonable because at $\lambda = 50$, the discrete envelope visually matches a continuous Gaussian. It fails because a discrete random variable concentrates finite mass at single integers, while integrating a continuous PDF at a single point yields identically zero. It is detected when someone computes $P(X = k)$ by evaluating $f(k)$ directly instead of integrating from $k - 0.5$ to $k + 0.5$. It is corrected by expanding discrete interval boundaries by $\pm 0.5$ when transitioning to continuous integration. Mini-example: for $P(X = 25)$ with $\lambda = 25$, the PDF height is $\frac{1}{\sqrt{50\pi}} \approx 0.0798$, which accurately approximates the integral over $[24.5, 25.5]$, but tail conditions like $P(X \geq 30)$ must integrate from $29.5$, not $30.0$.

3. Confusing probability density with probability

It seems reasonable because on a density curve, the vertical height $f(x)$ appears to indicate "how probable" an outcome is. It fails because $f(x)$ is a density carrying dimensional units $[x]^{-1}$. If $x$ is measured in millimeters instead of meters, $f(x)$ scales up by a factor of 1000 and can easily exceed unity. It is detected whenever someone claims a probability is $1.5$ or $2.8$ from peak heights of narrow distributions. It is corrected by remembering that dimensionless probability requires integration over an interval: $P = \int_a^b f(x) dx$. Mini-example: for a normal distribution with $\sigma = 0.01 \text{ s}$, the peak density is $f(0) \approx 40 \text{ s}^{-1}$; the probability over a $0.001 \text{ s}$ interval is $40 \times 0.001 = 0.04$, which is strictly valid.

4. Assuming Poisson variance is independent of the mean

It seems reasonable because in Gaussian distributions, mean $\mu$ and variance $\sigma^2$ are uncoupled free parameters. It fails because in Poisson statistics, mean and variance are tied by physical identity: $\sigma^2 = \lambda = \mu$. A Poisson process with mean 100 and standard deviation 2 cannot exist. It is detected when experimental counting uncertainties are reported without linking them to $\sqrt{N_{\text{counts}}}$. It is corrected by assigning $\sigma = \sqrt{N}$ to raw counting observables. Mini-example: recording 10,000 photon counts establishes an irreducible statistical uncertainty of $\sqrt{10000} = 100$ counts ($1\%$ relative noise).

5. Applying exponential decay to systems with wear or fatigue

It seems reasonable because exponential curves decay smoothly, resembling how everyday equipment deteriorates. It fails because the exponential distribution is uniquely memoryless: $P(T > t + s | T > s) = P(T > t)$. This implies an old component has the exact same failure probability in the next minute as a brand-new one. It is detected when material fatigue or mechanical friction is modeled via simple exponential distributions. It is corrected by adopting Weibull or log-normal distributions for aging mechanical components, reserving exponential models strictly for spontaneous nuclear decay and quantum transitions. Mini-example: Carbon-14 nuclei do not age (a 10,000-year-old nucleus has the identical decay probability today as a newly formed one); a turbine blade subjected to cyclic stress accumulates microcracks, violating memorylessness.

Quick detection rule

• If counting discrete events: verify if $\lambda < 20$. If so, Gaussian models are forbidden.
• If calculating continuous probabilities: check dimensions. If the value is not a dimensionless quantity between 0 and 1, a dimensional mistake occurred.

Physical Applications of Fundamental Probability Distributions

1. Astrophysics: photon shot noise in astronomical CCD sensors

Appears when imaging distant galaxies or faint exoplanets using CCD cameras on space telescopes (Hubble, JWST). It is calculated by modeling the photon flux incident on each detector pixel as a Poisson process with rate $\lambda = \Phi_{\text{photons}} \cdot A \cdot \Delta t \cdot \eta_{\text{quantum}}$. The result defines the irreducible shot noise $\sigma = \sqrt{\lambda}$, establishing the fundamental photometric detection limit that no hardware engineering can surpass. The critical danger is conflating Poisson photon shot noise with Gaussian electronic readout noise; readout noise is fixed per frame, whereas shot noise scales with the square root of source brightness. As a final sanity check, the signal-to-noise ratio $\text{SNR} = \lambda / \sqrt{\lambda + \sigma_{\text{read}}^2}$ is evaluated to determine optimal exposure duration before diffuse sky background saturates the detector dynamic range.

2. Nuclear physics: radiocarbon dating (Carbon-14)

Appears in archaeometry and nuclear geophysics to date organic samples by measuring the residual beta decay rate $^{14}\text{C} \to {}^{14}\text{N} + e^- + \bar{\nu}_e$. It is calculated using the exponential decay law $N(t) = N_0 e^{-\lambda t}$, where individual nuclear decay lifetimes follow an exponential distribution $f(t) = \lambda e^{-\lambda t}$ with mean lifetime $\tau = 1/\lambda = T_{1/2} / \ln(2) \approx 8267 \text{ yr}$. The result determines specimen age from isotopic ratio measurements in accelerator mass spectrometry (AMS). The critical danger is neglecting that for specimens older than 40,000 years, decay counts drop to tiny numbers where asymmetric Poisson statistics dominate; treating them with symmetric Gaussian error bars introduces severe chronological bias. As a final sanity check, derived ages are cross-calibrated against dendrochronological tree-ring records that account for historical atmospheric radiocarbon fluctuations.

3. Solid-state physics: Maxwell-Boltzmann thermal velocity distributions

Appears in kinetic theory and semiconductor physics to model thermal velocities of conduction electrons and gas molecules. It is calculated by recognizing that each Cartesian velocity component ($v_x, v_y, v_z$) independently follows a normal distribution $\mathcal{N}(0, k_B T / m)$. Speed $v = \sqrt{v_x^2 + v_y^2 + v_z^2}$ then follows the Maxwell speed distribution: $f(v) = \left( \frac{m}{2\pi k_B T} \right)^{3/2} 4\pi v^2 \exp\left( -\frac{m v^2}{2 k_B T} \right)$ The result provides the most probable speed $v_p = \sqrt{2 k_B T / m}$, mean speed $\langle v \rangle = \sqrt{8 k_B T / \pi m}$, and root-mean-square speed $v_{\text{rms}} = \sqrt{3 k_B T / m}$. The critical danger is assuming speed $v$ is normally distributed; speed is strictly non-negative, and its density vanishes quadratically at $v=0$ due to the spherical momentum volume element $4\pi v^2$. As a final sanity check, the equipartition theorem is verified by calculating the mean kinetic energy $\langle \frac{1}{2} m v^2 \rangle = \frac{3}{2} k_B T$.

4. Medical physics: cell survival modeling in radiation therapy

Appears in radiation oncology to predict cellular survival fraction $S(D)$ following ionizing radiation dose $D$ via the linear-quadratic model. It is calculated by modeling double-strand DNA lethal lesions as a spatial Poisson process across the cell nucleus: survival requires zero lethal hits: $S = P(k=0) = e^{-\langle k \rangle} = \exp(-\alpha D - \beta D^2)$. The result gives the survival curve used to optimize clinical dose fractionation, maximizing tumor eradication while sparing normal tissue. The critical danger is assuming a deterministic linear response while overlooking stochastic microscopic energy deposition (microdosimetry), which underestimates biological damage in hypoxic tumor regions. As a final sanity check, at low doses ($D \to 0$) survival responds linearly with slope $-\alpha$, while at high doses the quadratic dual-track mechanism $-\beta D^2$ takes over.

5. High-energy physics: particle discovery criteria ($5\sigma$ standard)

Appears in collider experiments (such as the Higgs boson discovery at CERN LHC) when searching for resonances above continuous standard-model backgrounds. It is calculated by integrating the standard normal tail: local statistical significance $Z$ is defined as $Z = \frac{N_{\text{obs}} - N_{\text{bkg}}}{\sigma_{\text{bkg}}}$. The result provides formal statistical confidence; particle physics demands $Z \geq 5\sigma$, corresponding to a one-sided $p$-value $p = 1 - \Phi(5) \approx 2.87 \times 10^{-7}$ (less than a 1-in-3.5-million chance of random background fluctuation). The critical danger is the "Look-Elsewhere Effect" (LEE): scanning a wide invariant mass window exponentially increases the global probability of observing a local $3\sigma$ or $4\sigma$ anomaly, prompting spurious discovery claims without trial-factor adjustments. As a final sanity check, global significance is quantified through complete Monte Carlo background simulations involving millions of synthetic pseudo-experiments.

Guided Practice: Fundamental Probability Distributions

Level 1: recognition

Select the most appropriate probability model for each physical scenario:

1. The number of thermal photons striking an astronomical CCD pixel in a 50 ms exposure with an average rate of 4 photons.
2. The Cartesian velocity component $v_x$ of gas molecules in thermal equilibrium at temperature $T$.
3. The elapsed waiting time between consecutive cosmic-ray detections on a plastic scintillator.
4. The count of spins oriented "up" in a paramagnetic system of $N=10$ independent atoms under an external magnetic field.

Solution:

1. Poisson distribution (discrete count of rare independent events in continuous exposure time, small mean $\lambda = 4$).
2. Normal (Gaussian) distribution (Maxwell-Boltzmann velocity distribution, $\mu = 0$, $\sigma^2 = k_B T / m$).
3. Exponential distribution (continuous waiting time between memoryless Poisson events).
4. Binomial distribution (number of successes in $N$ independent Bernoulli trials with probability $p$).

Level 2: direct application

The dielectric breakdown voltage of ceramic insulating disks follows a normal distribution with mean $\mu = 25.0 \text{ kV}$ and standard deviation $\sigma = 2.0 \text{ kV}$. Calculate the probability that a randomly chosen disk survives a proof test of $28.0 \text{ kV}$ without breaking down (i.e., breakdown voltage $X \geq 28.0 \text{ kV}$).

Solution:

1. Compute the standard $z$-score for $x = 28.0 \text{ kV}$: $z = \frac{x - \mu}{\sigma} = \frac{28.0 - 25.0}{2.0} = +1.5$
2. The standard normal cumulative CDF value at $z = 1.5$ is $\Phi(1.5) \approx 0.9332$.
3. Survival probability corresponds to the upper tail: $P(X \geq 28.0) = 1 - \Phi(1.5) = 1 - 0.9332 = 0.0668 \quad (6.68\%)$ Only $6.68\%$ of ceramic disks withstand $28.0 \text{ kV}$.

Level 3: tool selection

A neutron monitor records on average $\lambda = 400$ neutrons per minute inside a research reactor core. We wish to estimate the probability of detecting more than 430 neutrons in a given minute. Should we evaluate the exact Poisson sum or use a Gaussian approximation?

Solution: Because $\lambda = 400 \gg 20$, the Gaussian approximation is extremely accurate and avoids computing astronomically large factorials like $400!$.

1. Equivalent Gaussian parameters: $\mu = \lambda = 400$, $\sigma = \sqrt{\lambda} = \sqrt{400} = 20$.
2. Apply continuity correction for $k > 430$ (equivalent to $k \geq 431$): select threshold $x = 430.5$.
3. Compute $z$-score: $z = \frac{430.5 - 400}{20} = \frac{30.5}{20} = 1.525$
4. Probability: $\Phi(1.525) \approx 0.9364 \implies P(X > 430) = 1 - 0.9364 = 0.0636 \quad (6.36\%)$

Level 4: physical interpretation

Show that for a Poisson distribution, the ratio of standard deviation to mean (relative noise or coefficient of variation) is $\frac{\sigma}{\mu} = \frac{1}{\sqrt{\lambda}}$. What does this imply for the signal-to-noise ratio (SNR) in photon-counting spectroscopy and X-ray imaging?

Solution: For Poisson processes, mean is $\mu = \lambda$ and variance is $\sigma^2 = \lambda$, so $\sigma = \sqrt{\lambda}$. Relative noise is: $\frac{\sigma}{\mu} = \frac{\sqrt{\lambda}}{\lambda} = \frac{1}{\sqrt{\lambda}}$ Signal-to-noise ratio is $\text{SNR} = \frac{\text{Signal}}{\text{Noise}} = \frac{\mu}{\sigma} = \sqrt{\lambda}$. Physical implication: Quadrupling integration time or exposure dose multiplies count $\lambda$ by 4, but shot noise increases by only $\sqrt{4} = 2$. Consequently, SNR exactly doubles (scaling as $\sqrt{N}$). To gain a $10\times$ improvement in image SNR, one must collect $100\times$ more counts.

Level 5: intentional mistake

A student asserts: "In a normal distribution $\mathcal{N}(0, \sigma^2)$, if the standard deviation is small, say $\sigma = 0.1$, the peak density is $f(0) = \frac{1}{0.1 \sqrt{2\pi}} \approx 3.989$. Since probability cannot exceed 1, the formula must be unnormalized or mathematically flawed". Identify the conceptual blunder.

Conceptual error: The student confuses a discrete probability mass function $P(X = k)$ with a continuous probability density function $f(x)$.

1. For continuous random variables, exact point probability is zero: $P(X = 0) = 0$.
2. Normalization requires the integrated total area to equal 1: $\int_{-\infty}^\infty f(x) dx = 1$.
3. The density $f(x)$ carries physical units $[x]^{-1}$. If $x$ is measured in meters with $\sigma = 0.1 \text{ m}$, then $f(0) = 3.989 \text{ m}^{-1}$. Integrating over a narrow interval $\Delta x = 0.01 \text{ m}$ gives probability $P \approx f(0) \Delta x = 3.989 \times 0.01 = 0.03989$ ($3.99\% < 1$), fully obeying probability axioms.