Why Repeated Convolution Becomes a Bell Curve
Repeated convolution of any probability distribution with itself converges to a Gaussian, and this happens because convolution is the operation of adding independent random variables, whose sums the Central Limit Theorem forces toward the normal distribution.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 8 scenes and explore, respond, and learn as you go.
Why does convolving a distribution with itself over and over always converge to a Gaussian shape?
You blur an image once and it softens. Blur it a hundred times and something remarkable happens: the result looks like the same bell-shaped curve no matter what you started with.
Convolving a distribution with itself spreads it out, so it should keep getting flatter and wider — but it does not disappear. Something specific and stable is being approached.
An interactive simulation showing distributions evolving step by step under repeated self-convolution, with side-by-side comparison to a Gaussian curve.
Repeated convolution converges to a Gaussian because of a deeper fact: convolution corresponds to adding independent random variables, and the Central Limit Theorem guarantees that sums of independent variables approach a normal distribution.
Convolving flattens and widens a distribution, so after enough repetitions it might just spread out into a uniform flat line. The Gaussian shape feels too specific to be inevitable.
- Fourier-domain proofs of the Central Limit Theorem
- Edge cases like the Cauchy distribution (which is a special boundary case)
- Discrete versus continuous convolution technicalities
- Image-processing blur kernels beyond illustration
- 01The Mystery of Repeated BlurringslideQuestion
Introduce the driving question with a concrete image-blurring analogy: every blur is a convolution, and after many blurs the result looks like a bell curve no matter what the picture was.
- Convolving a distribution with itself spreads it out
- Repeating this many times should keep spreading it
- Yet something stable and specific appears: the Gaussian
- 02Your First GuessquizPrediction
Ask the learner to commit to an answer before any explanation: what shape does repeated self-convergence approach?
- Make one independent prediction
- Compare intuition against the eventual answer
- 03Watch Any Shape Become a Bell CurveinteractiveEvidence
Let the learner pick a starting distribution (uniform, triangular, exponential, two-peaked) and repeatedly convolve it with itself, watching each step overlay a growing Gaussian for comparison.
- Choose any starting shape
- Press convolve to add one more independent copy
- Overlay the Gaussian to see the match grow
- 04Convolution as Adding Random VariablesinteractiveEvidence
Show that convolving two distributions is exactly the same as summing two independent random variables drawn from those distributions, with a draggable histogram that updates as draws accumulate.
- Drag to draw from two distributions
- See the sum distribution emerge as a convolution
- Notice repeated convolution equals summing many copies
- 05The Central Limit Theorem Steps InslideExplanation
Explain why the sum of independent random variables — regardless of their individual shapes — converges to a Gaussian: characteristic functions multiply, and any well-behaved function raised to a high power becomes a quadratic in its logarithm.
- Repeated convolution is summing many independent copies
- Characteristic functions multiply under convolution
- A high power of a smooth function is dominated by its quadratic Taylor term
- A quadratic in log-space is a Gaussian in original space
- 06When It Fails: The Cauchy CounterexampleslideBoundary
Show the boundary case where repeated convolution does NOT converge to a Gaussian: the Cauchy distribution, which is its own stable distribution and reproduces itself under convolution rather than narrowing toward a bell.
- Heavy-tailed distributions can be stable under convolution
- Cauchy plus Cauchy is still Cauchy
- The result depends on the tails being light enough for averaging to work
- 07Apply It to a New SituationinteractiveTransfer
Test whether the learner can transfer the idea: given a new context (sum of many small independent errors in measurement), predict the shape of the total error distribution and verify by simulation.
- Identify the independent contributions being summed
- Predict the resulting shape
- Verify with the convolution simulator
- 08The Answer, DirectlyslideResolution
Restate the answer clearly: repeated self-convolution converges to a Gaussian because convolution is addition of independent random variables, and the Central Limit Theorem forces such sums toward the normal distribution for any starting distribution with finite variance.
- Convolution equals summing independent copies
- Many independent sums approach a Gaussian by CLT
- Cauchy is the boundary case where the theorem's assumptions break
Discussion threads for a Stage aren't available yet.
