Why the Cauchy Distribution Defies the Bell Curve

Pose the driving question with a visual hook: show two histograms side by side, one made from the average of many Gaussians (bell-shaped) and one made from the average of many Cauchys (still sharp-peaked with heavy tails).

• Repeated averaging makes Gaussians converge to a bell curve
• The same operation leaves the Cauchy distribution essentially unchanged
• This is surprising because both distributions look similar at first glance

Let the learner commit to one prediction about why the Cauchy distribution is preserved under convolution, before any mechanism is shown.

• Commit to an initial hypothesis before seeing evidence

A simulation where the learner repeatedly convolves either a standard normal distribution or a standard Cauchy distribution with itself, watching how the density evolves over iterations.

• Gaussian self-convolution visibly concentrates toward a narrower bell curve
• Cauchy self-convolution stays at the same scale and same shape
• Increase the number of convolutions to confirm the pattern persists indefinitely

Show a comparison of tail decay: Gaussian tails fall as exp(-x²) while Cauchy tails fall as 1/x², illustrating why Cauchy has no finite mean.

• Gaussian density is exp(-x²/2) — tails vanish extremely fast
• Cauchy density is 1/(π(1+x²)) — tails vanish very slowly
• Slow tail decay means extreme values appear far more often than expected

An interactive sampling tool that draws many points from either distribution; users can zoom the x-axis to see how often Cauchy draws land far from the center versus Gaussian draws.

• Gaussian samples cluster tightly near the mean
• Cauchy samples frequently include extreme outliers far from the center
• Each new Cauchy sample has a non-trivial chance of being far out in the tail

Explain the theory of stable distributions: a distribution is stable if the sum of i.i.d. copies has the same shape (up to scale). The Cauchy distribution corresponds to stability exponent α = 1, making it a fixed point of convolution.

• A stable distribution preserves its shape under summation
• The Gaussian is stable with exponent α = 2; Cauchy with α = 1
• Cauchy's α = 1 means convolving it with itself yields the same Cauchy (just rescaled)
• Because the shape is invariant, repeated convolution cannot turn it into a bell curve

Clarify the boundary: the central limit theorem requires finite mean and variance. Cauchy violates both, so the CLT does not apply, and stable distributions with α < 2 are precisely the limits of properly normalized sums.

• CLT requires finite second moment — Cauchy has none
• For α-stable distributions with α < 2, the CLT generalizes to convergence to the stable law
• Cauchy is the limiting case α = 1 of properly normalized sums of heavy-tailed variables

A transfer test where learners choose among several distributions (Gaussian, Cauchy, uniform, exponential) and predict whether repeated convolution will or will not turn it into a bell curve.

• Apply the concept of finite variance and stability to a new distribution
• Predict convergence behavior before observing the simulation result
• Distributions with heavy tails and undefined variance resist becoming bell-shaped

Resolve the opening tension by directly answering the driving question: the Cauchy distribution is preserved under convolution because it is a stable distribution of index α = 1, and its heavy tails prevent probability mass from concentrating.

• Cauchy is a fixed point of convolution, not an exception to it
• Heavy 1/x² tails ensure extreme samples keep appearing
• No finite mean means no central limit convergence to a Gaussian
• Repeated convolution cannot change the shape of a stable distribution