16
$\begingroup$

I think question title is obvious. assume we have a rectangular hyperbola chart.and we draw largest circle which fits under $y=1/x, y=0$ and $x=0$. then we continue drawing circles which are tangent to previous circle, $y= 1/x$ and $y=0$. Question is: what is radius of $n$-th circle.

enter image description here

Radius of first circle is $2 - \sqrt{2}$.

but even calculating second radius is impossible. It is over a month I am thinking on it.

PS 1: I try to find the line which connect all of circle's centers. if we call if f(x), it is clear that $$\lim_{x\to \infty}\frac{f(x)}{(1/x)}=\frac{1}{2}$$ another thing is following formula. the centers of circles that are in equal distance from y = 0 and first circle. (second circle center in on this line:) $$y=\frac{\left(r_{0}-x\right)^{2}}{4r_{0}}$$$$r_{0}=2-\sqrt{2}$$ and I get 6 formulas as follows: there are three points which are hit points of three curves. first circle and y=1/x hit point is (1,1) but two other points should be found. each of points satisfy curve formulas. also in eqach derivative of both curves are equal. and also, distance of two first circle's centers is r0+r1 which r1=y1.

$\endgroup$
4
  • 5
    $\begingroup$ Where is this question from? Knowing the context could be helpful to readers. $\endgroup$ Commented Dec 3, 2022 at 12:58
  • 2
    $\begingroup$ If you have been thinking about it for over a month, you must have tried a thing or two. What have you tried? Where do you get stuck? $\endgroup$ Commented Dec 3, 2022 at 13:23
  • 1
    $\begingroup$ I find this question in a mathematics fun forum. Even the person who ask this for first time doesn't have the answer. $\endgroup$ Commented Dec 3, 2022 at 13:43
  • 1
    $\begingroup$ If my Mathematica-assisted scratch work is correct (no guarantee there), the radius of the second circle involves the root of a degree-$7$ polynomial. $\endgroup$ Commented Dec 3, 2022 at 15:05

7 Answers 7

8
$\begingroup$

For dimensional homogeneity, let the hyperbola have equation $xy=s^2$.

At parameterized point $P=(sp,s/p)$ (for dimensionless $p$) on the hyperbola, a tangent vector is $\dfrac{p^2}{s}\dfrac{dP}{dp}=(p^2,-1)$, so that the unit downward-pointing normal is $$n := \left(-\frac{1}{\sqrt{1+p^4}}, -\frac{p^2}{\sqrt{1+p^4}}\right) \tag1$$ The center of the circle of radius $r$ tangent at $P$ will have center $K := P + r n$. For this circle to also be tangent to the $x$-axis, we require the $y$-coordinate to be $r$; solving for $r$ gives $$r = \frac{s}{p}\left(1+p^4-p^2\sqrt{1+p^4}\right)\tag2$$ so that the center is $$K = \left(\frac{s}{p}\left(2p^2-\sqrt{1+p^4}\right), \frac{s}{p}\left(1+p^4-p^2\sqrt{1+p^4}\right)\right) \tag3$$ (Sanity check: When $p=1$ and $s=1$, we have $K = (2-\sqrt{2},2-\sqrt{2})$, the correct center for the first circle.)

Now, if the circles for parameters $p$ and $q$ are tangent, then we have $$\left|K_p K_q\right|^2 = (r_p+r_q)^2 \tag4$$ This expands to a messy equation in $p$, $q$, $\sqrt{1+p^4}$, $\sqrt{1+q^4}$. A couple of rounds of squaring eliminates the roots, leaving a messier polynomial equation in $p$ and $q$: $$\begin{align} 0 = p^4 &- 12 p^3 q - 16 p^7 q + 38 p^2 q^2 + 76 p^6 q^2 + 64 p^{10} q^2 - 12 p q^3 - 176 p^5 q^3 \\ &- 112 p^9 q^3 + q^4 + 40 p^4 q^4 + 86 p^8 q^4 - 128 p^{12} q^4 - 176 p^3 q^5 - 56 p^7 q^5 \\ &+ 16 p^{11} q^5 + 76 p^2 q^6 + 516 p^6 q^6 + 172 p^{10} q^6 + 64 p^{14} q^6 - 16 p q^7 \\ &- 56 p^5 q^7 + 32 p^9 q^7 - 400 p^{13} q^7 + 86 p^4 q^8 + 392 p^8 q^8 + 625 p^{12} q^8 \\ &- 112 p^3 q^9 + 32 p^7 q^9 - 924 p^{11} q^9 + 64 p^2 q^{10} + 172 p^6 q^{10} + 1414 p^{10} q^{10}\\ &+ 64 p^{14} q^{10} + 16 p^5 q^{11} - 924 p^9 q^{11} - 400 p^{13} q^{11} - 128 p^4 q^{12} + 625 p^8 q^{12} \\ &+ 672 p^{12} q^{12} - 400 p^7 q^{13} - 400 p^{11} q^{13} + 64 p^6 q^{14} + 64 p^{10} q^{14} \end{align} \tag5$$ So, the general form is, to say the least, difficult to extract. However, for just the second circle, we can substitute $p=1$ into $(4)$ to get a degree-$7$ polynomial in $q$: $$\begin{align} 0 = 8 p^7 (2 - \sqrt2) &- p^6(73 - 48 \sqrt2 ) + 2 p^5 (23 - 16 \sqrt2) - p^4 ( 63 - 40 \sqrt2) \\ &+ 4 p^3 (5 - 4 \sqrt2)- 5p^2 (11 - 8 \sqrt2)+ 2p (7 - 4 \sqrt2) -1 \end{align} \tag6$$ Solving numerically, the only viable root is $q = 1.5831\ldots$, which corresponds to a radius of $0.3274\ldots$. (This result is consistent with a GeoGebra sketch.)

$\endgroup$
1
  • 1
    $\begingroup$ Nice. Just independently confirming your result for circle centers: I can confirm, given a tangent point $T: (x_T, 1/x_T)$ on the hyperbola $y=1/x$, the circle will have center: $C:(h,k)$ with $h = 2x_T - \sqrt{\dfrac{1}{x_T^2} + x_T^2}$ $k = \dfrac{1}{x_T} + x_T^3 - x_T^2 \sqrt{\dfrac{1}{x_T^2} + x_T^2}$ and $r = k$. One additional finding, the tangent line between two circles has the equation: $y=-\dfrac{h_n-h_{n-1}}{r_n - r_{n-1}}x+\dfrac{1}{2}\dfrac{h_n^2-h_{n-1}^2}{r_n - r_{n-1}}$ which interestingly has its x-intercept at $\dfrac{h_n + h_{n-1}}{2}$ $\endgroup$ Commented Dec 3, 2022 at 19:04
3
$\begingroup$

Consider the hyperbola equation $pq=1$. Its gradient at $\left(p,q\right)$ is $\left(q,p\right)$. If $r$ is the radius of the circle tangent at $\left(p,q\right)$ then its center is at $\left(p,q\right)-r\frac{\left(q,p\right)} {\left\Vert q,p\right\Vert}$.

The same circle is tangent at the $x$-axis when the radius $r$ is the same as the $y$-coordinate of its center. This happens when $$r=q-r\frac{p} {\left\Vert q,p\right\Vert}$$ so that $$r=\frac q {1+p/\left\Vert q,p\right\Vert}$$ and this gives the curve $$ C\left(p\right) = \left(p-r\frac q{\left\Vert q,p\right\Vert},r\right) $$ with $pq=1$, that contains all the centers of all circles tangent on the hyperbola and the $x$-axis.

For the first circle we have $p=q=1$. Then $r=\frac 1 {1+1/\sqrt2}=2-\sqrt2$.

Put $c_n=\left(p_n,q_n\right)=C\left(p_n\right)$ as the center of the $n$-th circle with radius $r_n$. Suppose $c_n$ and $r_n$ are known. The point $c_{n+1} \in C$ is the center of the $n+1$-th circle tangent to the previous one if $\left\Vert c_{n+1}-c_n\right\Vert = q_n + q_{n+1}$ since the distance between the two centers is the sum of the two radii and this is the sum of the $y$-coordinates of the centers of the circles. It is equivalent to $\left(p_n-p_{n+1}\right)^2=4 q_n q_{n+1}$ and we need $p_{n+1}>p_n$ for our problem.

This gives the conditions that let us calculate the $\left(n+1\right)$-th center and radius.

For $n=1$ we have $p_1=q_1=r_1=2-\sqrt 2 $ and the previous equation becomes $$ \left(p-\frac{p^{-2}}{\sqrt{p^{2}+p^{-2}}+p}-2+\sqrt{2}\right)^{2}=4\left(2-\sqrt{2}\right)\frac{p^{-1}\sqrt{p^{2}+p^{-2}}}{\sqrt{p^{2}+p^{-2}}+p} $$

that gives the solution $p_2\approx 1.58313$ and $r_2\approx0.327489$.

$\endgroup$
2
  • $\begingroup$ Nice! Can you please clarify why $\left(p_n-p_{n+1}\right)^2=4 q_n q_{n+1}$ is equivalent to $\left\Vert c_{n+1}-c_n\right\Vert = q_n + q_{n+1}$? $\endgroup$ Commented Sep 7, 2023 at 16:53
  • 1
    $\begingroup$ Squaring both sides of $\left\Vert c_{n+1}-c_n\right\Vert = q_n + q_{n+1}$ and simplifying gives $\left(p_n-p_{n+1}\right)^2=4 q_n q_{n+1}$. $\endgroup$ Commented Sep 8, 2023 at 20:11
1
$\begingroup$

Even though there are many answers, one can consider a trigonometric alternative;

For your question, the following geometry arises,

Trig interpretation

Where the following relation of

$$\tan{ \pi - \theta} = \frac{d}{dx}\big[\frac{1}{x}\big]$$

yields,

$$\theta = \arctan{\frac{1}{x^2}}, \theta \in [0, \pi]$$

Consider, $\phi = \pi - \theta$ and the isosceles triangle involving $\phi$, the other two angles (of the mentioned isosceles) must be $\frac{\pi - (\pi - \theta)}{2} = \frac{\theta}{2}$ each.

A bit more geometry, reveals:

More geometry

If the dashed side of the isosceles is $a$, then the following simultaneous equations arise:

$$\begin{cases}a^2 = \frac{1}{t^2\cos^2{\frac{\theta}{2}}} & ,t \text{ is the tangency point } \\ a^2 = 2r^2 - 2r^2\cos{(\pi - \theta)} \end{cases}$$

Solve for $r$, yields $$r = \frac{\sqrt{2}}{2\sqrt{\cos{(\theta)} + 1} \cdot \cos{\big( \frac{\theta}{2}\big) \cdot t}}$$

But $\theta = \arctan{\frac{1}{t^2}}$

Hence, we obtain the radius of the circle in terms of $t$, the intersection point of $y = \frac{1}{x}$ and some $n^{th}$ circle:

$$r(t) = \frac{\sqrt{2}\big(t^4 + 1\big)^\frac{1}{4}}{2t\sqrt{t^2 + \sqrt{t^4 + 1}}\cdot \sin{\big(\frac{\arctan{t^2}}{2} + \frac{\pi}{4}\big)}}$$

(To confirm this works, $r(1) = 2 - \sqrt{2}$ indeed).

From hereon, a method is to continue from @Blue 's answer $\to (4) \dots $

An alternative includes, looking at the radii between $2$ circles in terms of $r_{n}(t)$ and $\theta$

Sum of radii

If what's above if true, then (for second circle):

$$\frac{1}{t_2} = \sqrt{r(t_2)^2 - \big(t_2 - 2 + \sqrt{2} - (r(t_2) + r(1))*\cos{(\arctan{t^2})}\big)^2} + r(t_2)$$

for which my CAS is still BuFfErRiNg (self evident). Again re-confirming with @Blue 's result for $t_2 \approx 1.5831$, $r(1.5831) \approx 0.3275$. Moving on, circle three would be disastrous.

$\endgroup$
2
  • $\begingroup$ I've toyed a bit with re-parameterizing the hyperbola by $(\sqrt{\tan\theta}, \sqrt{\cot\theta})$, which incorporates some notions from your approach. If nothing else, this parameterization has the benefits that the unit normal is simply $(-\cos\theta,-\sin\theta)$ and the ugly radicals are disguised as other trig functions. So far, however, the trigonometric counterparts of my $(5)$ and $(6)$ haven't been any nicer than the polynomial forms. ... I'll note that, either way, eliminating the parameters to relate the radii directly seems to make things much, much worse in the general case. $\endgroup$ Commented Dec 4, 2022 at 15:44
  • $\begingroup$ @Blue , then this might be it. The question mostly like does not suit elementary methods, given we are trying to find a general form, which may require the help of different series (my intuition suggests the circles could be connected through a geometric series). Nonetheless, great effort in your answer. $\endgroup$ Commented Dec 4, 2022 at 15:58
0
$\begingroup$

For the $i$-th circle, let $(x_i, r_i)$ be the position of the center, and $r_i$ the radius.

For $ i = 1$, $r_1 = 2 - \sqrt{2} , x_1 = 1 - \dfrac{1}{\sqrt{2}} r_1 $

The next circle has $(x_{i+1}, r_{i+1})$ as the center. By distance from the previous circle,

$ (x_i - x_{i+1})^2 + (r_{i+1} - r_i)^2 = (r_{i+1} + r_i)^2$

so,

$ 4 r_{i+1} r_i = (x_i - x_{i+1})^2 \dots (1) $

If the tangency point of the new circle with $f(x)=\dfrac{1}{x}$, is $(t_{i+1}, \dfrac{1}{t_{i+1}})$, then, by distance,

$ (t_{i+1} - x_{i+1})^2 + (\dfrac{1}{t_{i+1}} - r_{i+1})^2 = r_{i+1}^2$

And this simplifes to

$ t_{i+1}^2 (t_{i+1} - x_{i+1})^2 + 1 - 2 r_{i+1} t_{i+1} = 0 \dots (2)$

Using the slope of the tangent,

$ \dfrac{t_{i+1} - x_{i+1}}{\dfrac{1}{t_{i+1}} - r_{i+1}} = \dfrac{1}{t_{i+1}^2} $

And this reduces to

$ t_{i+1}^3 (t_{i+1} - x_{i+1}) = 1 - r_{i+1} t_{i+1} \dots (3)$

The unknowns are: $ r_{i+1}, x_{i+1}, t_{i+1} $

Solving numerically for each $i$, generates the circles.

The figure below shows the result.

enter image description here

$\endgroup$
0
$\begingroup$

COMMENT.- The true question in this post is about the $n^{th}$ radius and this it is not hard to prove that if $(a_n,r_n)$ is the center of the $n^{th}$ circle we have the following: $$r_{2n}=r\left[\frac{(a_{2n}-a_{2n-1})(a_{2n-2}-a_{2n-3})\cdots(a_2-a_1)}{(a_{2n-1}-a_{2n-2})(a_{2n-3}-a_{2n-4})\cdots(a_3-a_2)(a_1-r)}\right]^2\\r_{2n+1}=\frac{1}{4r}\left[\frac{(a_{2n+1}-a_{2n})(a_{2n-1}-a_{2n-2})\cdots(a_3-a_2)(a_1-r)}{2r(a_{2n}-a_{2n-1})(a_{2n-2}-a_{2n-3})\cdots(a_2-a_1)}\right]^2$$ where $r=2-\sqrt2$ which is the radius $r_0$ of the first circle.

In any case, this is an alternative in search towards an explicit formulation (very complicated). Unfortunately calculation of the abscissas $a_i$ of the centers of the circles are not easy to get. However it is possible to step by step to calculated them so getting finitely many values as we could. In order to have something improving this, it is obvious that we must use tangency points of two circles and the other one of tangency with the hyperbola because if not then our circles are not determined. The task is difficult but I guess it is not impossible to solve.

$\endgroup$
2
  • $\begingroup$ $1.5831...$ is the x-coordinate of the point where the second circle touches $y=1/x$. $\endgroup$ Commented Dec 5, 2022 at 3:42
  • $\begingroup$ @Dan.- Thanks you very much dear friend. It was for that I said **If I have well understood ** because I value Blue's knowledge (My English is not good and Petronella's answer made me make the mistake) $\endgroup$ Commented Dec 5, 2022 at 10:59
0
$\begingroup$

You can use methods in the other answers to get the first few radii, say the first $k$ radii, where $r_k/r_{k-1}\approx1$, then you can easily approximate the other radii by using the following approximation formula, which I prove below:

$$r_n\approx \left({r_k}^{-2}+8(n-k)\right)^{-1/2}$$

The approximation is close even with $k=2$. In the graph below, the red circles' radii were approximated by my approximation formula with $k=2$.

enter image description here

Here is the proof of my approximation formula.

Consider two adjacent circles with radii $R$ and $r$, where $R>r$.

enter image description here

First we show that $\lim\limits_{r/R\to1}\left(\dfrac{1}{r^2}-\dfrac{1}{R^2}\right)=8$

$\lim\limits_{r/R\to1}\left(\dfrac{1}{r^2}-\dfrac{1}{R^2}\right)$
$=\lim\limits_{r/R\to1}\left(\dfrac{1}{r^2}-\dfrac{1}{R^2}\right)\left(\dfrac{CD}{AB}\right)$
$=\lim\limits_{r/R\to1}\left(\dfrac{1}{r^2}-\dfrac{1}{R^2}\right)\left(\dfrac{R+r}{(x\text{-coordinate of }B)-(x\text{-coordinate of }A)}\right)$
$=\lim\limits_{r/R\to1}\left(\dfrac{1}{r^2}-\dfrac{1}{R^2}\right)\left(\dfrac{R+r}{\frac{1}{2r}-\frac{1}{2R}}\right)$
$=\lim\limits_{r/R\to1}\left(\dfrac{R^2-r^2}{r^2R^2}\right)\left(\dfrac{R+r}{\frac{R-r}{2Rr}}\right)$
$=\lim\limits_{r/R\to1}\dfrac{2(R+r)^2}{Rr}$
$=\lim\limits_{r/R\to1}\left(\dfrac{2R}{r}+4+\dfrac{2r}{R}\right)$
$=8$

$\therefore\dfrac{1}{{(r_n)}^2}\approx \dfrac{1}{{(r_{n-1})}^2}+8 \approx \dfrac{1}{{(r_{n-2})}^2}+16 \approx \dfrac{1}{{(r_{n-3})}^2}+24 \approx... \approx \dfrac{1}{{(r_k)}^2}+8(n-k)$

$\therefore r_n\approx \left({r_k}^{-2}+8(n-k)\right)^{-1/2}$

$\endgroup$
0
$\begingroup$

The first circle $C_1$ is well-known. Now, let $C_{n}$, $n\geq 2$ be the circle with center $O_{n}(x_{n},r_{n})$ which is tangent to the circle $C_{n-1}$, $x$-axis and the hyperbola $y=\frac1x$. Let $P_{n}(a_{n},\frac{1}{a_n})$ be the point where $C_n$ is tangent to the hyperbola. From the slope condition $\large\frac{r_{n}-\frac{1}{a_{n}}}{x_{n}-a_{n}}=a_{n}^2$ and the condition $|O_{n}P_{n}^{}|^2=r_{n}^2$ which is $(\frac{1}{a_{n}}-r_{n})^2+(a_{n}-x_{n})^2=r_{n}^2$, we have $x_{n}=2a_{n}-\frac{1}{a_{n}}\sqrt{1+a_{n}^4}$ and $r_{n}=a_{n}^3+\frac{1}{a_{n}}-a_{n}\sqrt{1+a_{n}^4}.$ The condition $|O_{n}O_{n-1}|=r_{n}+r_{n-1}$ gives $(x_{n}-x_{n-1})^2=4r_{n}r_{n-1}.$ Now let $a_{n}=a, x_{n-1}=m$ and $r_{n-1}=n.$ Then we have $$8na^9-(16mn+9)a^8+(8nm^2+24m)a^7-(22m^2+16n^2)a^6+(8m^3+16n)a^5+(6-m^4-16nm)a^4+(8m^2n-8m)a^3+(2m^2-16n^2)a^2+8na-1=0$$ Examples: $x_2≈1.4617, r_2≈0.3275, a_2≈1.5831; x_3≈2.026, r_3≈0.2434, a_3≈2.081; x_4≈2.469, r_4≈0.201, a_4≈2.501$ Approximations: $a_{n}≈a_{n-1}+\frac{1}{a_{n-1}}+\frac{1}{4a_{n-1}^3}; x_{n}≈x_{n-1}++\frac{1}{x_{n-1}}+\frac{1}{4x_{n-1}^3}$ and $r_{n}≈\large\frac{r_{n-1}}{1+4r_{n-1}^2+2r_{n-1}^4}≈r_{n-1}-4r_{n-1}^3+14r_{n-1}^5.$

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.