1

I came up with a question a while ago that I think is important in conic sections, but I don't have any ideas yet on how to proceed with this type of problem (I very much wish there was a style of pure Euclidean geometry)

If we create a set of conic sections tangent to two same parallel lines in the two same points like in this image:

enter image description here

The question is, what is the family of curves, each of which is perpendicular to all the curves ?

Jean Marie
  • 88,997

1 Answers1

6

The conic curves you consider undergo four constraints : all of them pass through the 2 points $(0,-1)$ and $(0,1)$ and are tangents in those points to the 2 lines with resp. equations $y=-1$ and $y=1$.

This kind of family of conics is called a bitangent linear pencil. The curves belonging to this pencil can be given the general equation :

$$ax^2+y^2-1=0 \ \ \ \begin{cases}a>0& \ \text{ for ellipses}\\a<0& \ \text{ for hyperbolas}\\a=0& \ \ \text{for the pair of lines}\end{cases} \tag{1}$$

Remark : Equation (1) relies on a single parameter $a$, a "degree of freedom". How could it be foreseeable ? Very simply : a conic curve is determined by $5$ conditions (passing through $5$ points, or being tangent to $5$ lines or any "mixture" between those conditions), it remains $5-2-2=1$ degree of freedom.

Here are two methods, the first one (have a look at Fig. 1) provides a definitive answer. The second one is a numerical method (almost always possible for obtaining orthogonal curves) giving a satisfying graphical answer but no more.


First method : (see Fig. 1).

The set of orthogonal functions is given by equations

$$y=\sqrt{-W(n,-ke^{x^2})} \tag{2}$$

where $W$ is the Lambert function with parameters $n=0$ or $n=-1$, depending on the arbitrary positive constant $k$.

Proof of (2) in the case of the principal branch ($n=0$) :

We will abbreviate (it is classical) $W(0,x)$ into $W(x)$.

Function $W$ is known to verify the following differential equation :

$$\frac{W'(u)}{W(u)}=\frac{1}{u(1+W(u))}\tag{3}$$

Besides, as (2) can be written under the form :

$$\ln y = \frac12 \ln(-W(-ke^{x^2}))$$

Its differentiation gives :

$$\frac{y'}{y}=\frac12 \frac{-W'(-ke^{x^2})(-2kxe^{x^2})}{W(-k e^{x^2})}$$

Now use relationship (3) with $u=-k e^{x^2}$ in order to show that (5) is verified.

enter image description here

Fig. 1. Ellipses, Hyperbolas, and the two families of orthogonal curves with resp. colors red, magenta, green and blue.

Fig. 1 has been obtained with the following SAGE program implementing the $W$ function :

 var('x y')
 g=line([[0,-2.5],[0,2.5]],color='black',aspect_ratio=1,); # axes etc.
 g+=line([[-2,0],[2,0]],color='black')
 g+=plot(1,(x,-2,2),color='red')
 g+=plot(-1,(x,-2,2),color='red')
 for k in range(5):
    a=(k+1)/2;
    g+=implicit_plot(a*x^2+y^2-1,(x,-2,2),(y,-1,1),color='red'); # ellipses
    g+=implicit_plot(-a*x^2+y^2-1,(x,-2,2),(y,-2.5,2.5),color='magenta'); # hyperbolas
 v(x,p,b)=sqrt(-lambert_w(p,-b*exp(x^2)))
 r=1.8;
 for k in range(6):
    b=(k+1)/40
    g+=plot(v(x,0,b),(x,-r,r),color='green',plot_points=3000)
    g+=plot(v(x,-1,b),(x,-r,r),color='blue',plot_points=3000)   
 g

Second method (numerical solution : see Fig. 2) :

First of all, we re-write (1) under the form :

$$\frac{1-y^2}{x^2}=a\tag{4}$$

In this way, parameter $a$ disappears when (4) is differentiated with respect to $x$ :

$$\frac{-2yy'x^2-(1-y^2)2x}{x^4}=0\tag{5}$$

(5) can be considered as equivalent to :

$$y'=\frac{y^2-1}{xy}\tag{6}$$

which is the ordinary differential equation (ODE) common to all curves of the pencil of conics.

The ODE common to all orthogonal curves is obtained by replacing $y'$ in the expression above by $-\frac{1}{y'}$ (the product of slopes in any point $(x,y)$ must be $-1$) giving :

$$y'=\frac{xy}{1-y^2}\tag{7}$$

Numerical solutions of (7) can be obtained (see blue curves below) with a code in any scientific programming language having an ODE solver. I have done it with Matlab.

Some numerical inaccuracies occur around line $y=1$ ; they are due to the $1-y^2$ denominator in (7).

enter image description here

Fig. 2 : Representation of the solutions in the first quadrant, WLOG because the issue is symmetrical with respect to the $x$ and $y$ axes.

Jean Marie
  • 88,997
  • 1
    Thank you very much. I think that the scope of these tools is currently beyond my capabilities, and I expected in advance that the curves would be complex and that I would not be able to understand the ideas behind them currently. I gave you +1 for now, and later when my level improves, I will return to read the answer again, God willing. I thought that this question It is important, and that is why I presented it on the site, and perhaps later additional importance and more results for these curves will become clear – زكريا حسناوي Dec 29 '23 at 15:21
  • You mention your "capabilities" : please say what you know in "analysis". Are you self-taught ? Do you know derivation for example ? It is important to go beyond conic curves... – Jean Marie Dec 29 '23 at 18:58
  • 2
    Yes, I am a self-taught person. My geometric skills are good, but I am a beginner in calculus. I have some geometry knowledge that goes beyond conic sections, but it is difficult to explain what I know and what I do not know. – زكريا حسناوي Dec 29 '23 at 19:12
  • 1
    I have made an important "edit" to my previous answer with a nicer figure. I admire your geometric skills, coming from a self-taught person, and the way you discover hidden properties. Surely, you will progress by learning new algebraic techniques. – Jean Marie Dec 31 '23 at 05:59
  • Thank you very much. Yes, the new format is similar to what I imagined. Thank you very much for the advice. I will try to improve my algebraic skills in the coming periods, and of course I will continue from time to time to submit more of my work. I am happy with your answers to my questions. You are helping me develop my content and level. – زكريا حسناوي Dec 31 '23 at 09:08
  • About a different question about "bitangent pencil of conics" see my answer here. This question is asked in the context of projective geometry (where figures can be deformed by the "large" set of projective transformations but where angles are - in principle - prohibited) whereas your question is in the context of euclidean geometry (where in particular right angles make sense) with a poorer set of transformations (only rotations, translations, symmetries) but we did not have to bother about these transformations in the question here. – Jean Marie Feb 11 '24 at 08:54
  • 2
    This is really interesting – Claude Leibovici May 04 '24 at 15:59
  • @زكريا حسناوي As this question has been asked 5 months ago and you have received only this answer with 6 upvotes, you should validate it ; it is important for the "good health" of this site that convenient answers to "old" questions are validated, a way to "close" a question. – Jean Marie Jun 04 '24 at 07:05