1

The problem is:

In a world without gravity, a very small gun shooting point-like balls is located at the lower left end $(0, 0)$ of a $2D$ corridor. The corridor has length $L = 100\thinspace m$ and height $h = 5\thinspace m$. The gun shoots with initial velocity $v = 1\thinspace m/s$ at a random angle $\alpha$, distributed uniformly between 0 and $\pi/4$. Each time the ball bounces off the wall, its velocity component perpendicular to the wall is reduced by a fraction $\eta/(1 + \eta)$, \begin{equation} \upsilon_{vertical}\rightarrow -\left(1-\frac{\eta}{1+\eta}\right)\upsilon_{vertical}, \end{equation} where $\eta$ has an exponential distribution with mean $E(\eta) = 0.15$ and is drawn independently for every bounce


What is the expected flight time (and number of bounces). What approximation can you use to get 95% confidence interval?


I calculated it to be 250.83599 (8.32808). Using CLT, interval would be E(X)+-1.96*stddev/n where n is the number of simulations. Are those corrects? Are my derivations correct?

The paper is available at: http://www.filedropper.com/quanttestv2

----------

Below is an extract for $E[T]$.

We quickly find that: \begin{equation} T=\frac{5}{sin(\alpha)}\sum_{k=0}^{N-1}\prod_{i=0}^{k}\left( 1+\eta_i \right)+T_N \end{equation} with \begin{equation} T_N=(\frac{100}{cos(\alpha)}-(\lfloor 20tan(\alpha) \rfloor)\frac{5}{sin(\alpha)})\prod_{i=0}^{N}\left( 1+\eta_i \right) \end{equation}

\begin{align} E[T]&=E[\frac{100}{cos(\alpha)}|\alpha<=\arctan(\frac{1}{20})]+E[\frac{5}{sin(\alpha)}\sum_{k=0}^{N-1}\prod_{i=0}^{k}\left( 1+\eta_i \right)+T_N|\alpha>arctan(\frac{1}{20})] \end{align} \begin{align} E[\frac{100}{cos(\alpha)}|\alpha<=\arctan(\frac{1}{20})]&=\int_{\arctan(0)}^{\arctan(\frac{1}{20})}\frac{4}{\pi} \frac{100}{cos(x)} \thinspace dx\\ &=\frac{400}{\pi}\left[ ln|tan(\frac{x}{2}+\frac{\pi}{4})|\right]_{\arctan(0)}^{\arctan(1/20)} \end{align}

\begin{align} E[T|\alpha,\alpha>arctan(\frac{1}{20})]&=\frac{5}{sin(\alpha)}\sum_{k=0}^{N-1}E[\prod_{i=0}^{k}\left( 1+\eta_i \right)]+(\frac{100}{cos(\alpha)}-N \frac{5}{sin(\alpha)})E[\prod_{i=0}^{N}\left( 1+\eta_i \right)] \end{align}

The $\eta_i$ are independent from each other, and so are $(1+\eta_i)$.

\begin{align} E[T|\alpha,\alpha>arctan(\frac{1}{20})]&=\frac{5}{sin(\alpha)}\sum_{k=0}^{N-1}1.15^k+(\frac{100}{cos(\alpha)}-N \frac{5}{sin(\alpha)})1.15^N \end{align}

\begin{align} E[T|\alpha>arctan(\frac{1}{20})]&=\int_{arctan(\frac{1}{20})}^{\pi/4} f_\alpha(x) \left( \frac{5}{sin(x)}\sum_{k=0}^{N-1}1.15^k +(\frac{100}{cos(x)}-N \frac{5}{sin(x)})1.15^N \right) \thinspace dx \\ &=\frac{20}{\pi} \sum_{j=1}^{19} \left( \sum_{k=0}^{j-1}1.15^k \right) \left[ ln|tan(\frac{x}{2})| \right]_{\arctan(\frac{j}{20})}^{arctan(\frac{j+1}{20})}\\ &+\frac{4}{\pi}\sum_{j=1}^{19} 1.15^j \left( 100\left[ ln|tan(\frac{x}{2}+\frac{\pi}{4})| \right]_{\arctan(\frac{j}{20})}^{\arctan(\frac{j+1}{20})} - 5j\left[ ln|tan(\frac{x}{2})| \right]_{\arctan(\frac{j}{20})}^{\arctan(\frac{j+1}{20})} \right) \end{align}

\begin{align} E[T]&=6.36355+219.90387+24.56857\\ &=250.83599 \end{align}

Did
  • 279,727
phil
  • 11
  • 1
    If I understand the request, it is for Readers to go to the linked paper and verify some calculations that you've done there. Readers are more likely to respond thoughtfully to a calculation that is presented here, even if only a simplified representative calculation is done. – hardmath Oct 14 '14 at 14:58
  • I could probably skip 2 equations out of 3 and display something. Will do it in a few hours once I got access to my latex. – phil Oct 14 '14 at 15:14

1 Answers1

2

Assuming that $L=Nh$ for some integer $N\geqslant1$, the number $B$ of bounces before exiting the corridor is such that, for every $k$, $$[B\geqslant k]=[\tan\alpha\geqslant k/N],$$ hence $$E(N)=\sum_{k\geqslant1}P(B\geqslant k)=\sum_{k\geqslant1}P\left(\tan\alpha\geqslant \frac{k}N\right)=\sum_{k=1}^N\left(1-\frac4\pi\arctan\left(\frac{k}N\right)\right).$$ For $N=20$, one finds $$E(B)\approx8.328.$$ For every $N$, the sum is $N$ times a Riemann sum hence, when $N\to\infty$, $$E(B)\sim N\int_0^1\left(1-\frac4\pi\arctan x\right)\mathrm dx=\frac{\log4}\pi N.$$ For $N=20$, this approximation reads $$E(B)\approx8.825.$$ A similar approach can be applied to the time $T$ to exit the corridor since $$N\tan\alpha=B+C,$$ with $C$ in $[0,1)$ and the $k$th part of the flight, between abscissae $(k-1)h/\tan\alpha$ and $kh/\tan\alpha$ has length $h/\sin\alpha$ and is done at a speed $v_k$ such that $E(1/v_k)=(E(1+\eta))^{k-1}=\delta^{k-1}$ with $\delta=1.15$, hence $$E(T)=E\left(\sum_{k=1}^B\frac{h}{\sin\alpha}\delta^{k-1}+\frac{hC}{\sin\alpha}\delta^B\right).$$ This is equivalent to $$E(T)=L\,\tau_0+h\,\tau_{\delta,N},$$ where $$\tau_0=E\left(\frac1{\cos\alpha}\right),\qquad\tau_{\delta,N}=(\delta-1)\,\sum_{k=1}^{N}E\left(\frac{k\delta^{k-1}}{\sin\alpha}\mathbf 1_{N\tan\alpha\geqslant k}\right).$$ The term $L\tau_0$ corresponds to purely elastic bounces, that is, to $\delta=1$, described by the factor $$\tau_0=\frac4\pi \log(1+\sqrt2)\approx1.122.$$ The term $h\tau_{\delta,N}$ is the correction due to the inelasticity of the bounces, described by the factor $$\tau_{\delta,N}=\frac4\pi (\delta-1)\sum_{k=1}^Nk\delta^{k-1}\,\left(\log(1+\sqrt{1+(k/N)^2})-\log(k/N)-\log(\sqrt2+1)\right).$$ For $N=20$ and $\delta=1.15$, $\tau_{\delta,N}=55.879$ hence, for $(L,h)=(100,5)$, $$E(T)\approx L\,1.122+h\,55.879\approx391.6.$$

Did
  • 279,727
  • Hi Did. Thanks for your answer.I also found 8.328 both in my simu and my calculations. I also found 112 when no inelasticity. But I 'm not sure with your second term as I'm finding 1,210 for it (may have got it wrong). My simus and calculations give 250 – phil Oct 16 '14 at 11:09
  • Added the numerical application when L=100, h=5, and E(eta)=0.15. – Did Oct 16 '14 at 15:13
  • Thanks. But that means my derivation is wrong (no clue why, especially as my Monte carlo was giving 250). If you could spot the difference btw my explanations and yours, that would be great. I will try to understand yours more deeply tonight as well (am at work). – phil Oct 16 '14 at 15:44