From 755f048b2cfcb949701ff0844c4c30f76fc90d00 Mon Sep 17 00:00:00 2001 From: Erik Strand <erik.strand@cba.mit.edu> Date: Wed, 27 Feb 2019 00:47:32 -0500 Subject: [PATCH] Add a proof --- _psets/3.md | 25 +++++++++++++++++++++---- 1 file changed, 21 insertions(+), 4 deletions(-) diff --git a/_psets/3.md b/_psets/3.md index 644b13a..ea146db 100644 --- a/_psets/3.md +++ b/_psets/3.md @@ -217,13 +217,13 @@ So as long as the error rate isn't $$1/2$$ (i.e. zero information gets through), of majority voting the error rate can be made arbitrarily small. Let's prove this. Since $$f_0$$ is a polynomial, it's continuous. By inspection there are three -solutions to $$x = 3 x^2 - 2 x^3$$: zero, one half, and one. Thus $$f_0$$ has three -[fixed points](https://en.wikipedia.org/wiki/Fixed_point_(mathematics)). This suffices to show that -for any $$x$$, if $$f_n(x)$$ converges it must converge to zero, one half, or one. +[fixed points](https://en.wikipedia.org/wiki/Fixed_point_(mathematics)): zero, one half, and one. +This suffices to show that for any $$x$$, if $$f_n(x)$$ converges it must converge to zero, one +half, or one (I'll prove this below). Now fix any $$x$$ such that $$0 < x < 1/2$$. Because $$f_0$$ is a cubic polynomial, it can't cross the line $$y = x$$ more than three times. We've already noted that it does cross this line exactly -three times (at zero, one half, and one). So the fact that +three times, namely at its fixed points. So the fact that $$ 3 \cdot \left( \frac{1}{4} \right)^2 - 2 \cdot \left( \frac{1}{4} \right)^3 @@ -257,6 +257,23 @@ $$ This symmetry establishes the claim. +For completeness let's prove the claim that a sequence generated by recursive application of a +continuous function $$f$$ can only converge to a fixed point of $$f$$. I'll do this in two steps. + +First, we must establish that if a sequence $$x_0, x_1, \ldots$$ in the domain of $$f$$ converges to +a point $$x_\infty$$ (also in the domain of $$f$$), then the sequence $$f(x_0), f(x_1), \ldots$$ +converges to $$f(x_\infty)$$. Fix any $$\epsilon > 0$$. Since $$f$$ is continuous, there is some +$$\delta$$ such that $$\lvert x - x_\infty \rvert < \delta$$ implies $$\lvert f(x) - f(x_\infty) +\rvert < \epsilon$$. Since $$x_0, x_1, \ldots$$ converges to $$x_\infty$$, there is some $$N$$ such +that $$n > N$$ implies $$\lvert x_n - x_\infty \rvert < \delta$$. Thus $$n > N$$ implies $$\lvert +f(x_n) - f(x_\infty) \rvert < \epsilon$$ which establishes the claim. + +Second, we can show that any sequence $$x_0, x_1, \ldots$$ generated by successive application of +$$f$$ that converges must converge to a fixed point of $$f$$. Say $$x_0, x_1, \ldots$$ converges to +$$x_\infty$$. Since $$x_{n + 1} = f(x_n)$$ for all $$n$$, this means that $$f(x_0), f(x_1), \ldots$$ +converges to $$x_\infty$$. But by the result just proven, $$f(x_0), f(x_1), \ldots$$ converges to +$$f(x_\infty)$$. Since limits are unique this means $$x_\infty = f(x_\infty)$$. + ### Behavior of leading term For small $$\epsilon_0$$, the cubic term in $$f_0(\epsilon_0)$$ is approximately zero. So let's -- GitLab