10

We have $y = x^2 + 1$ and are trying to find the point closest to $(4, 0)$. For context, this is for a high school calculus class.

Clearly, $l = \sqrt{(x - 4)^2 + ((x^2 + 1) - 0)^2}$, from which we get $l^2 = (x - 4)^2 + ((x^2 + 1) - 0)^2$ and so $2 l \frac{\text{d}l}{\text{d}x} = 4x^3 + 6x - 8$. Setting this equal to zero to find maxima and minima, we get $4x^3 + 6x - 8 = 0$. Dividing everything by $2$, we finally have $2x^3 + 3x - 4 = 0$. How do I solve this cubic? It seems that the root is irrational, so the rational root theorem is failing me, and I don't know what else to try. Am I missing something obvious? Did my teacher make a typo?

3 Answers3

8

I don't think there's a high school way to solve this. I am using the standard hyperbolic substitution for depressed cubic with one real root.

$$2x^3 + 3x - 4 = 0 \tag{1}$$

Since one root is real and coefficient of $x>0$, we use: $$\sinh(3\theta) = 4\sinh^3 \theta + 3\sinh \theta$$

Multiplying $(1)$ by $2$ gives $4x^3 + 6x = 8$. Let $x = \sqrt{2}\sinh \theta$. $$ 4(\sqrt{2}\sinh \theta)^3 + 6(\sqrt{2}\sinh \theta) = 8 $$ $$ 8\sqrt{2}\sinh^3 \theta + 6\sqrt{2}\sinh \theta = 8 $$ Dividing by $2\sqrt{2}$: $$ 4\sinh^3 \theta + 3\sinh \theta = \frac{4}{\sqrt{2}} = 2\sqrt{2} $$ $$ \sinh(3\theta) = 2\sqrt{2} \implies \theta = \frac{1}{3}\text{arcsinh}(2\sqrt{2}) $$

Using $\text{arcsinh}(z) = \ln(z+\sqrt{z^2+1})$: $$ \theta = \frac{1}{3}\ln(2\sqrt{2}+3) = \frac{1}{3}\ln((\sqrt{2}+1)^2) = \frac{2}{3}\ln(\sqrt{2}+1) $$ Substituting into $$x = \sqrt{2}\sinh \theta = \frac{\sqrt{2}}{2}(e^\theta - e^{-\theta})$$ $$ x = \frac{1}{\sqrt{2}}\left( (\sqrt{2}+1)^{2/3} - (\sqrt{2}-1)^{2/3} \right) $$

  • 3
    That's a pretty neat answer! Never thought you could use a hyperbolic sub to solve this equation...but it works! – bjcolby15 Dec 12 '25 at 17:29
  • 2
    @bjcolby15 most people only see the cubic equation as Cardano's formula, but in fact the general theory of the cubic equation should show that the case where there is only one single solution is a hyperbolic substitution, and the case where there are three solutions is a trigonometric substitution, where in the trigonometric case you can choose either of cosine or sine and still get answers regardless of the choice (the hyperbolic case you have to use the one that is suitable to the case at hand). As such, this is a very known thing. – naturallyInconsistent Dec 12 '25 at 20:36
  • @naturallyInconsistent True, that...I'm so used to seeing trig substitutions for root finding that discovering this is amazing. Still a neat answer, though :-) – bjcolby15 Dec 12 '25 at 22:13
7

You have been given the exact solution by @匚ㄖㄥᗪ乇ᗪ.

You can have a good approximation of the real root of $$2x^3 + 3x - 4 = 0 $$ knowing (by inspection or plotting) that the solution is "close" to $1$.

Let $x=1-\epsilon$ and expand. Consider now that you look for the zero of $$g(\epsilon)=1-9 \epsilon +6 \epsilon ^2-2 \epsilon^3$$ Neglecting the highest power, solving the quadratic gives $$\epsilon \sim \frac{9-\sqrt{57}}{12}=0.120847 \quad \implies \quad x \sim 0.879153$$ while the exact solution is $0.879615$.

If you want to improve, knowing this result, repeat the same process with $x=1-\frac {12}{100}-\epsilon=\frac {22}{25}-\epsilon$

$$g(\epsilon)=\frac{46}{15625}-\frac{4779 \epsilon}{625}+\frac{132 \epsilon ^2}{25}-2\epsilon ^3$$ and neglect the second and third power. This gives $$\epsilon=\frac{46}{119475}\quad \implies \quad x \sim \frac{105092}{119475}=0.879615$$ which correspond to an absolute error of $10^{-7}$.

Sooner or later, you will also learn that we can approximate functions by rational functions. Around $x=1$ $$2x^3 + 3x - 4 \sim \frac{2 x^2+71 x-64}{2 x^2-10 x+17}$$ To see how decent it is, plot the two functions between $x=0$ and $x=2$.

Solving for zero the numerator gives $$x \sim \frac{3\sqrt{617}-71}{4}=0.879614$$

Edit (for the fun)

Amazing could be to continue with rational approximations, writing $$2x^3 + 3x - 4 \sim \frac {a_nx^2+b_nx+c_n}{1+\sum_{k=1}^n \alpha_k\,x^k }$$ As a function of $n$,below are the quadratic equations to solve and, converted to decimals, the approximate root

$$\left( \begin{array}{ccc} n & \text{quadratic equation} & \text{solution} \\ 0 & 6 x^2-3 x-2 & 0.879152869605896 \\ 1 & 9 x^2+8 x-14 & 0.879597254201444 \\ 2 & 2 x^2+71 x-64 & 0.879613522561331 \\ 3 & 69 x^2-138 x+68 & 0.879614146914231 \\ 4 & 414 x^2-205 x-140 & 0.879614885414259 \\ 5 & 619 x^2+556 x-968 & 0.879614880026944 \\ 6 & 126 x^2+4905 x-4412 & 0.879614879828974 \\ 7 & 4779 x^2-9506 x+4664 & 0.879614879821710 \\ 8 & 28570 x^2-14003 x-9788 & 0.879614879812342 \\ 9 & 42573 x^2+38640 x-66928 & 0.879614879812398 \\ 10 & 7866 x^2+338855 x-304148 & 0.879614879812399 \\ \end{array} \right)$$

How to build these approximations

To approximate a function $f(x)$ around a given point $x_0$, the most typical is a Taylor series expansion. More efficient are the $[m,n]$ Padé approximants which write $$f(x) =\frac {\sum_{k=0}^m \alpha_k\,(x-x_0)^m}{1+\sum_{k=1}^n \beta_k\,(x-x_0)^k }$$ For brevity, I shall use $t=x-x_0$.

In the present case, with $x_0=1$ $$f(x)=2x^3+3x-4 \quad \implies \quad f(t)= 1+9 t+6 t^2+2 t^3$$ and the $[2,3]$ Padé approximants write $$\frac {\sum_{k=0}^2 \alpha_k\,t^m}{1+\sum_{k=1}^3 \beta_k\,t^k }$$ Now, cross multiplying, consider $$\Big(1+\beta_1\,t+\beta_2\,t^2+\beta_3\,t^3\Big)\Big( 1+9 t+6 t^2+2 t^3\Big)=\alpha_0+\alpha_1\,t+\alpha_2\,t^2$$ Put every on the same side, expand and group by powers; this will give a fifth degree polynomial $$\sum_{n=0}^5 \gamma_n\,t^n=0$$ where $$ \gamma_0=1-\alpha_0 \qquad \gamma_1=9-\alpha_1+\beta_1 \qquad \gamma_2= 6-\alpha_2+9\beta_1+\beta_2$$ $$ \gamma_3=2+6\beta_1+9\beta_2+\beta_3\qquad \gamma_4=2\beta_1+6\beta_2+9\beta_3\qquad \gamma_5=2\beta_2+6\beta_3$$

Cancelling all of them is simple (six linear equations for six variables); it gives $$\alpha_0=1 \quad \alpha_1=0 \quad \alpha_2=-69\quad \beta_1=-9 \quad \beta_2=6 \quad \beta_3=-2$$

So, the numerator we need to cancel is $$1-69t^2=1-69(x-1)^2 \qquad \implies \qquad x=1-\frac{1}{\sqrt{69}}=0.879614$$

We could have been more lazy, considering instead the $[1,n]$ Padé approximant; it would generate the sequence of rational numbers $$\left\{\frac{8}{9},\frac{22}{25},\frac{548}{623},\frac{4552}{5175},\frac{548}{623},\frac{314092}{357079}, \frac{2609060}{2966139},\frac{2408068}{2737639},\frac{180027332}{204666083},\cdots\right\}$$

7

Here is the middle school way for the cubic equation : $$2x^3+3x-4=0 \iff 8x^3-16 + 12x =0\tag{1}$$

We intend to use the known formula $$a^3 +b^3 +c^3 -3abc = (a+b+c)(a^2+b^2+c^2-ab-bc-ca) \tag{2}$$ with $c = 2x$.

We will find $(a,b)$ such that $a^3+b^3 =-16$ and $ab =-2$. It suffices to solve the quadratic equation $$X^2+16X-8=0 \iff(a,b)=\left(\sqrt[3]{-8-6\sqrt{2}}, \sqrt[3]{-8+6\sqrt{2}},\right)$$

As the equation $(2)$ has 1 unique solution $a+b+c=0\iff x=-\frac{a+b}{2}$, therefore, the equation $(1)$ has 1 unique solution $$\color{red}{x = -\frac{a+b}{2}= \frac{1}{2} \left(\sqrt[3]{8+6\sqrt{2}}+ \sqrt[3]{8-6\sqrt{2}} \right)} $$

NN2
  • 20,543