2

Two points $P$ and $Q$ lies in the $xy$-plane. $P$ is moving from $(1,0)$ to $(0,0)$ along the $x$-axis and $Q$ is moving from $(0,0)$ to $(1,0)$ along the $y$-axis following the parametrized equations ; $$\begin{cases} P(t)=(\cos t,0)\\ Q(t)=(0,\sin t) \end{cases}$$ Let $L$ be the line segment $PQ$. What is the region that is swept by $L$ as $0\le t\le\frac\pi2$?

gif file

I think the curve in the first quadrant is $x^{\frac32}+y^{\frac32}=1$.

(1) Intuitively (using Calculus)

R(t;\Delta t) and R(t)

The lines $L(t)$ and $L(t;\Delta t)$ at time $t$ and $t+\Delta t$ can be expressed as \begin{align*} L(t)&:\frac x{\cos t}+\frac y{\sin t}=1\\ L(t+\Delta t)&:\frac x{\cos(t+\Delta t)}+\frac y{\sin(t+\Delta t)}=1\\ \end{align*} Let $R(t;\Delta t)$ be the intersection of $L(t)$ and $L(t+\Delta t)$ and let $R(t)$ be the limit of $R(t;\Delta t)$ as $\Delta t\to0$. $$ \begin{bmatrix} \frac1{\cos t} &\frac1{\sin t}\\ \frac1{\cos(t+\Delta t)} &\frac1{\sin(t+\Delta t)} \end{bmatrix} \begin{bmatrix}x\\y\end{bmatrix} = \begin{bmatrix}1\\1\end{bmatrix} $$ \begin{align*} D &=\frac1{\cos t}\cdot\frac1{\sin(t+\Delta t)}-\frac1{\sin t}\cdot\frac1{\cos(t+\Delta t)}\\ &=\frac{\sin t\cos(t+\Delta t)-\cos t\sin(t+\Delta t)}{\cos t\sin t\cos(t+\Delta t)\sin(t+\Delta t)}\\ &=\frac{\sin(t-(t+\Delta t))}{\cos t\sin t\cos(t+\Delta t)\sin(t+\Delta t)}\\ &=-\frac{\sin t}{\cos t\sin t\cos(t+\Delta t)\sin(t+\Delta t)} \end{align*} \begin{align*} R(t;\Delta t)=\begin{bmatrix}x\\y\end{bmatrix} &=\frac1D \begin{bmatrix} \frac1{\sin(t+\Delta t)} &-\frac1{\sin t}\\ -\frac1{\cos(t+\Delta t)} &\frac1{\cos t} \end{bmatrix} \begin{bmatrix}1\\1\end{bmatrix}\\ &= \begin{bmatrix} \frac{\cos t\cos(t+\Delta t)}{\sin h}\left(\sin(t+\Delta t)-\sin t\right)\\ -\frac{\sin t\sin(t+\Delta t)}{\sin h}\left(\cos(t+\Delta t)-\cos t\right) \end{bmatrix} \end{align*} $$R(t)=\lim_{\Delta t\to0}R(t;\Delta t) = (\cos^3 t,\sin^3 t)$$ We can (intuitively) think of $R(t)$ as the point at time $t$. So the curve becomes $x^{\frac32}+y^{\frac32}=1$.

(2) Rigorously? (using the Convex Optimization)

A point $R$ in the line segment $L$ at time $t$ can be expressed as a parameter $s$ ($0\le s\le 1$) \begin{align*} R(t,s) &=sP(t)+(1-s)Q(t)\\ &=(s\cos t,(1-s)\sin t)\\ &=(x,y). \end{align*}

Finding the curve in the first octant is to find the function that assign $x$ to $y$ $$f(x)=\max_{\substack{0\le t\le\frac\pi2\\0\le s\le1}}y.$$ That is, it reduces to the convex optimization problem; $$\text{Maximize} \quad\quad (1-s)\sin t$$ under the three constraints ; $$\begin{cases} s\cos t = x\\ 0\le t\le\frac\pi2\\ 0\le s\le1 \end{cases}$$

Is it possible to evaluate the exact equation $x^{\frac32}+y^{\frac32}=1$ using the convex optimization?

(3) Or does anyone can show me a rigorous way to find the curve?

shyzealot
  • 119
  • The tehnique you use is the determination of an envelope of lines. Here is an answer of mine which addresses a problem similar to your issue. See as well https://mathimages.swarthmore.edu/index.php/Envelope – Jean Marie Apr 13 '22 at 03:34
  • The problem is only similar : indeed the length of the line segment is not constant, but the essential thing in the answer I gave is the Edit explaining the method involving a "differentiation with respect to parameter", which is in fact the technique you have (re)discovered with your infinitesimal increments $\Delta t$. – Jean Marie Apr 13 '22 at 03:43
  • [+1] Everything is exact. It is the classical "sliding ladder" issue. The curve you obtain is the fourth part of an astroid. – Jean Marie Apr 13 '22 at 03:46
  • I have no idea about a possible connection with convex optimization. – Jean Marie Apr 13 '22 at 03:48
  • Another view: the astroid as an (hypo/epi)cycloid here – Jean Marie Apr 13 '22 at 03:54
  • 'envelope'! I see. Actually I've learned the notion of envelope in the differential geometry class like years ago. But, I was really bad at differential geometry, although it looks really fascinating and physics-like. Thanks for your kind response. – shyzealot Apr 13 '22 at 04:31
  • The example that you showed me looks really similar to mine. So I have to differentiate with respect to t in my case. And I understand that my approach matches the classical approach to some extent. And the name of the curve is the `astroid' I see. Thanks! – shyzealot Apr 13 '22 at 04:35
  • Thanks for your thanks. I have had (and to a certain extend still have) the same "experimental" approach, rediscovering (sometimes important) techniques/results. In some cases, it was a long time after that the light was done. Now with the Internet, and forums like this one, this time can be much shortened, fortunately. Don't hesitate to ask for other explanations... – Jean Marie Apr 13 '22 at 04:43
  • @JeanMarie Yeah. There are so many things to think about in mathematics. Thanks!! – shyzealot Apr 13 '22 at 06:20

0 Answers0