3

I saw that the area where the tangent line can be drawn is divided into the corresponding divided area as shown in the picture below based on the tangent line drawn from the inflection point of the cubic function. enter image description here

In this picture,

green area: We can draw one tangent line.

pink area: We can draw three tangent lines.

And after studying more, I found that this theorem is also valid for quadratic functions and quartic functions. So I wondered if this theorem is valid for any function of any degree. However, I don't know how to prove it from the beginning. How can I prove this?

Etit: I thought of a function with degree n, which is an extreme case, and has many convex parts like this. I wondered if it was possible to generalize the number of regions where tangent lines can be drawn, even in such a complex case. If possible, I posted the question here in hopes of getting some hints. enter image description here

  • I can't understand your question- Why can you not draw a tangent line at every point of a graph of a differentiable function? – trula Aug 14 '24 at 13:30
  • @trula I edited my question just now! – user1274233 Aug 14 '24 at 13:40
  • Again: i don't understand what you mean with how many tangent lines one can drav. I can draw a tangent line at any point. so for example if I am patient at 100 points or 1000, what keeps you from drawing so many? – trula Aug 14 '24 at 13:45
  • @trula I understand what you are saying. So, in these cases, since the elementary function is differentiable multiple times, there is no obstacle to drawing the tangent line, so the only way is to draw the tangent line multiple times to fit the case? – user1274233 Aug 14 '24 at 13:52
  • 1
    I think OP is not fixing a point on the graph of the polynomial, but a generic point of the plane and he's wondering about the number of tangent lines passing through that point and tangent to the graph of the polynomial. – Kandinskij Aug 14 '24 at 15:13
  • I think you are great to see this from the original post or graphic, but probably you are right. – trula Aug 14 '24 at 15:36

1 Answers1

1

Let $r:y=mx+(-mx^\ast+y^\ast)$ be a generic line passing through the point $(x^\ast,y^\ast)\in \mathbb{R}^2$ and let $p(x)=\sum_{i=0}^n a_ix^i$ be a generic polynomial of degree $n$ (so $a_n\neq 0$). Clearly

$$r \ \ \text{is tangent to the graph of $p$}\iff \text{there is $x_0\in \mathbb{R}$ such that $\begin{cases} m(x_0-x^\ast)+y^\ast=p(x_0)\\ m=p'(x_0)\end{cases}$}$$

So $$r \ \ \text{is tangent to the graph of $p$}\iff p'(x_0)(x_0-x^\ast)+y^\ast=p(x_0) \ \ \ \text{for some $x_0\in \mathbb{R}$}$$ So we have $$r \ \ \text{is tangent to the graph of $p$}$$ $$ \iff$$ $$a_{n}(1-n)x_{0}^n+\sum_{i=1}^{n-1}(a_{i}-a_{i} i+x^\ast a_{i+1}(i+1))x_0^i+(a_0+x^\ast a_1-y^\ast)=0,\ \ \ \text{for some $x_0\in \mathbb{R}$}$$ Let $\mathcal{S}_{(x^\ast,y^\ast)}$ be the set of $x_0\in \mathbb{R}$ that solve the equation above and let $$\mathcal{T}_{(x^\ast,y^\ast)}:=\{p'(x_0):x_0\in \mathcal{S}_{(x^\ast,y^\ast)}\}.$$ I denote with $|\mathcal{S}_{(x^\ast,y^\ast)}|$ and $|\mathcal{T}_{(x^\ast,y^\ast)}|$ the cardinalities of these sets.

The number $|\mathcal{T}_{(x^\ast,y^\ast)}|$ is the number of tangent lines passing through $(x^\ast,y^\ast)$.

The number $|\mathcal{S}_{(x^\ast,y^\ast)}|$ is the number of tangent lines with their tangency multiplicities : basically each tangent line is counted once if it's tangent in just one point, twice if it's tangent in two points, thrice if it's tangent in three points and so on.

To see this subtlety consider the linear case $p(x)=a_1x+a_0$. Let $(x^\ast,y^\ast)$ be a point on the graph of $p$ (that is a line in this case). Clearly $|\mathcal{T}_{(x^\ast,y^\ast)}|=1$ because there is just one tangent line (the line $p$ itself), but $|\mathcal{S}_{(x^\ast,y^\ast)}|$ is infinite because the only tangent line is tangent in every point.

So here we have two questions:

  1. How to calculate $|\mathcal{S}_{(x^\ast,y^\ast)}|$?

  2. How to calculate $|\mathcal{T}_{(x^\ast,y^\ast)}|$?

The second one is very difficult so I'll ignore it. In this answer I'll focus on the first question (in most examples $|\mathcal{S}_{(x^\ast,y^\ast)}|=|\mathcal{T}_{(x^\ast,y^\ast)}|$).

Remember that $|\mathcal{S}_{(x^\ast,y^\ast)}|$ is the number of real solutions of the following polynomial equation:

$$a_{n}(1-n)x^n+\sum_{i=1}^{n-1}(a_{i}-a_{i} i+x^\ast a_{i+1}(i+1))x^i+(a_0+x^\ast a_1-y^\ast)=0$$

For simplicity, I'll call the polynomial above $Q(x)$. If $n=1$, the situation is trivial so I'll suppose $n\geq 2$. If $n\geq 2$ then the polynomial $Q(x)$ has degree $n$ so it has at most $n$ solutions. We have our first result.

Proposition 1: $|\mathcal{S}_{(x^\ast,y^\ast)}|\leq n$

This is compatible with your "experimental results". Moreover it's a well known fact that a polynomial of odd degree has at least a real root.

Proposition 2: $|\mathcal{S}_{(x^\ast,y^\ast)}|\geq 1$ if $n$ is odd.

This is completely compatible with your "experimental observation".


CONTINUATION OF THE ANSWER:

Let's discuss the low-degree cases i.e. $n\in \{1,2,3\}$.

n=1

As I said the case $n=1$ is trivial.

n=2

In this case:

$$Q(x)=-a_2x^2+2x^\ast a_2 x+(a_0+x^\ast a_1-y^\ast)$$

The number of zeroes of $Q$ depends on the sign of the discriminant $\Delta$. It's easy to see that $$\Delta=0\iff y^\ast=Q(x^\ast)$$ So the parabola $Q$ perfectly splits the plane in two part: one whose points have no tangent line and one whose points have two tangent lines and the points on the parabola have exactly a tangent line.

Curiosity: This result may seem an incredibly special property, but it holds not only for parabolas but for the graph of any $C^2$ function with second derivative of constant sign.

n=3

In this case

$$Q(x)=-2a_3x^3+(3x^\ast a_3-a_2)x^2+2x^\ast a_2 x+(a_0+x^\ast a_1-y^\ast)$$

Although it's not very well known, there is a discriminant also in the cubic equation

A cubic equation $$Ax^3+Bx^2+Cx+D=0$$ has three distinct real roots iff $$-27A^2D^2+18ABCD-4AC^3-4B^3D+B^2C^2>0$$ For simplicity , I'll denote the expression above $\Delta_3$. We have also that $\Delta_3=0$ iff there are $2$ distinct real roots and $\Delta_3<0$ iff there is just one real root.

For our polynomial $Q$ we have that (with an incredibly tedious computation)

$$\Delta_3=4 (p(x^\ast) - y^{\ast}) (9 a_3 a_2^2 x^\ast - 27 a_1 a_3^2 x^\ast + 27 a_3^2 y^{\ast} + a_2^3 - 27 a_0 a_3^2)$$

So basically the problem is completely controlled by the sign of the factors $(Q(x^\ast)-y^{\ast})$ an $(9 a_3 a_2^2 x^\ast - 27 a_1 a_3^2 x^\ast + 27 a_3^2 y^{\ast} + a_2^3 - 27 a_0 a_3^2)$. The sign of the first factors tells us whether $(x^\ast,y^\ast)$ is under or above the graph of our polynomial $p$. The sign of the second factors tells us if $(x^\ast,y^\ast)$ is under or above the following line $$\ell: y=\left(-\frac{a_2^2}{a_3}+a_1\right)x-\frac{a_2^3}{27a_3^2}+a_0$$ Who is this mysterious line $\ell$? Exactly the tangent line at the inflection point that you cited in your question! (Do you know how to prove this?).

n=4

TO BE CONTINUED

Kandinskij
  • 3,407
  • I sincerely appreciate your answer! – user1274233 Aug 14 '24 at 15:17
  • See this answer involving the notion of dual curve. – Jean Marie Aug 14 '24 at 16:24
  • 1
    @JeanMarie I think the setting of the question is affine and not projective. – Kandinskij Aug 14 '24 at 16:31
  • Question: Since p(x) is $\sum_{i=0}^n a_ix^i$ and p'(x) is $\sum_{i=0}^n ia_ix^{i-1}$, I thought the formula for $p'(x_0)(x_0-x^)+y^=p(x_0)$ was $(\sum_{i=1}^n ia_ix_0^{i-1})(x_0-x^)+y^=\sum_{i=0}^n a_ix_0^i$, but how can the formula $a_{n}(1-n)x^n+\sum_{i=1}^{n-1}(a_{i}-a_{i} i+x^\ast a_{i+1}(i+1))x^i+(a_0+x^\ast a_1-y^\ast)=0$ appear? – user1274233 Aug 14 '24 at 23:16
  • Hello, I tried to understand the formula by reading the answer and writing it myself, but I have a question, so I'm asking again. If possible, please answer it. I would really appreciate it. – user1274233 Aug 14 '24 at 23:17
  • In the last formula you wrote I substituted $x_0$ with $x$ to simplify the notation. Is it clear now? Nothing deep is really happening here: it's just computation (if you want I can write the computation explicitely). – Kandinskij Aug 15 '24 at 05:53
  • I'm really sorry for bothering you. Even if I consider $x_0$ and $x$ as you said, I still have some questions, such as how the formula $a_n(1-n)x^n$ was derived. If possible, could you please briefly describe the derivation process? – user1274233 Aug 15 '24 at 06:16
  • Do you agree that $\left(\sum_{i=1}^n ia_ix_0^{i-1}\right)(x_0-x^\ast)=\sum_{i=1}^n ia_ix_0^{i}-\sum_{i=1}^n x^\ast ia_ix_0^{i-1}$ ? This is just distributive property – Kandinskij Aug 15 '24 at 06:53
  • Yes, I agree it! – user1274233 Aug 15 '24 at 06:56