Why Self-Convolution Becomes a Bell Curve

Pose the driving question: start with a sharp square pulse, convolve it with itself once, then again, then again. Each step looks like simple smoothing, yet the result converges to a precise bell-shaped curve. Frame this as a puzzle rather than a lesson.

• Define self-convolution of a shape with itself
• Show the surprising endpoint: a Gaussian, not a vague blob
• State the driving question explicitly

Let the learner manipulate a slider that controls how many times a starting shape (square, triangle, or skewed bump) is convolved with itself. Before revealing the full sequence, ask what curve they expect after many iterations and invite them to choose between 'a smooth blob with no special form', 'the original shape, just wider', 'a Gaussian bell curve', or 'depends on the starting shape'.

• Pick a starting shape
• Slide the iteration count
• Commit to a prediction before the evidence unfolds

Present a static montage of the square pulse after 1, 2, 3, 5, and 10 self-convolutions. Show how corners soften, then triangular segments round into smooth arcs, then approach a symmetric bell. Overlay the true Gaussian with matching variance on the final frame so the match is unmistakable.

• Corners disappear first, then segments round out
• Shape becomes symmetric very quickly
• Final curve overlays a perfect Gaussian

Let the learner switch between three starting shapes (square, triangle, sharply skewed double bump) and watch each one iterated to n=10. All three converge to visually the same bell curve, scaled and shifted. This shows the endpoint does NOT depend on the starting shape — evidence that something universal is at work.

• Switch among different starting shapes
• Run each to the same iteration count
• Observe all curves overlap near a single Gaussian shape

Reinterpret convolution of a distribution with itself: picking a value from f and adding it to an independent pick from f gives the distribution f*f. Iterate n times and you get the distribution of a sum of n independent copies. By the Central Limit Theorem, that sum converges to a Gaussian as n grows, regardless of the original distribution's shape.

• f*f equals the distribution of X1 + X2 with X1, X2 independent draws from f
• n iterations = sum of n independent draws
• Central Limit Theorem forces convergence to a Gaussian
• The Gaussian is the unique fixed point: G*G = G when variances add correctly

Note the limits of the result. Heavy-tailed distributions (like Cauchy) have infinite variance, so the Central Limit Theorem does not apply and iterated self-convolution does not collapse to a Gaussian — the curve stays in the same family but does not narrow. Point out that the Gaussian is special precisely because it is its own fixed point under convolution.

• Infinite-variance distributions break the convergence
• Finite variance is the hidden assumption
• Gaussian is the unique self-reproducing fixed point

One focused transfer question: present a new scenario (a skewed triangular distribution iterated 20 times) and ask whether it will converge to a Gaussian. The correct answer is yes, because finite variance lets the Central Limit Theorem apply; the distractor options test common misconceptions (no, because the shape is asymmetric; no, only symmetric shapes converge; yes, but to the original triangle scaled up).

• Recognize that asymmetry is not an obstacle
• Identify finite variance as the real requirement
• Distinguish the limit shape from the starting shape

Directly answer the driving question. Repeated self-convergence builds a sum of independent contributions, so the Central Limit Theorem drives any finite-variance starting shape toward a Gaussian. The Gaussian is special because it is the unique distribution that reproduces itself under convolution — making it the inevitable destination of this process.

• Restate the answer clearly
• Name the Central Limit Theorem as the mechanism
• Highlight the Gaussian as the unique fixed point