Do you know Soumava's Algorithm?

Calculus Level 3

x y = y x = 43 \large x^y = y^x = 43

If x x and y y are distinct positive real numbers satisfying the equation above, find 100 ( x + y ) \lfloor 100(x+y) \rfloor .

You are allowed to use computer assistance.

You might refer to Soumava's algorithm .


The answer is 2698.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

1 solution

Harry Ray
Sep 16, 2016

This can be solved using the Newton-Raphson method but first we need to set up our function and establish the number of roots. First, note that: x y = 43 y ln ( x ) = ln ( 43 ) y = ln ( 43 ) ln ( x ) . \begin{aligned} x^y &= 43 \\ y\ln(x) &= \ln(43) \\ y &= \frac{\ln(43)}{\ln(x)}. \end{aligned} Substituting this back into the first equation gives: y x = ( ln ( 43 ) ln ( x ) ) x . y^x = \left(\frac{\ln(43)}{\ln(x)}\right)^x. The function that we want to find the root of is: f ( x ) = ( ln ( 43 ) ln ( x ) ) x 43 = e x ln ( ln ( 43 ) ln ( x ) ) 43 \begin{aligned} f(x) &= \left(\frac{\ln(43)}{\ln(x)}\right)^x - 43 \\ &= e^{x\ln\left(\frac{\ln(43)}{\ln(x)}\right)} - 43 \end{aligned}

We now need to establish the number of roots that this function has. It is actually simpler to consider x y = y x x^y = y^x . We can rearrange this as: ln ( y ) y = ln ( x ) x \frac{\ln(y)}{y} = \frac{\ln(x)}{x} Clearly x = y x = y is always a solution to this but since we need x x and y y to be distinct we are only interested in other solutions. Fortunately g ( x ) = ln ( x ) / x g(x) = \ln(x)/x is not injective. We can sketch it by noting the following facts: lim x 0 g ( x ) = g ( 1 ) = 0 lim x g ( x ) = 0 g ( x ) = 1 ln ( x ) x 2 g ( e ) = 0 \begin{aligned} \lim_{x \to 0} g(x) &= -\infty \\ g(1) &= 0 \\ \lim_{x \to \infty} g(x) &= 0 \\ g^\prime(x) &= \frac{1 - \ln(x)}{x^2} \\ g^\prime(e) &= 0 \end{aligned} Also, note that x = e x = e is the only turning point. Thus g ( x ) g(x) looks like this: So g ( x ) = k g(x) = k has precisely two solutions for any k ( 0 , 1 / e ) k \in (0, 1/e) , one solution in the range ( 1 , e ) (1, e) and the other in the range ( e , ) (e, \infty) .

We cannot directly map this back onto the original problem since we also need ln ( x ) / x = ln ( 43 ) / x y \ln(x)/x = \ln(43)/xy . What we can infer, though, is that the solution set to x y = y x x^y = y^x consists of the line y = x y = x , as well as a smooth curve from the region ( 1 , e ) × ( e , ) (1,e) \times (e, \infty) , through the point ( e , e ) (e,e) , to the region ( e , ) × ( 1 , e ) (e,\infty) \times (1,e) .

The solutions to the original problem correspond to the intersection between this curve and y = ln ( 43 ) / ln ( x ) y = \ln(43)/\ln(x) . This will have either one or three solutions, depending on whether it intersects off the line y = x y = x . Therefore, we will search for a solution to f ( x ) = 0 f(x) = 0 near x = 1.1 x = 1.1 .

Starting at x 0 = 1.1 x_0 = 1.1 , and applying the Newton-Raphson method, after 5 iterations we have x 5 = 1.156752515 x_5 = 1.156752515 . The corresponding value for y y is y 5 = ln ( 43 ) / ln ( x 5 ) = 25.82948719 y_5 = \ln(43)/\ln(x_5) = 25.82948719 . The required solution is then 100 ( x + y ) = 2698 \lfloor 100(x + y) \rfloor = \boxed{2698} .

Thanks for this neat solution with the entire concept explained with plots.

@Soumava Pal you gotta see this.

Agnishom Chattopadhyay - 4 years, 8 months ago

@Agnishom Chattopadhyay you are welcome. Of course, it's easy to just use Soumava's algorithm, but I encountered this problem in the "Curve sketching" section and wanted to give a solution that actually involved sketching curves.

Harry Ray - 4 years, 8 months ago

@Agnishom Chattopadhyay I have actually done the exact same improvement in my paper less than a month ago. The same intuition.

@Harry Ray

This is absolutely correct.

Soumava Pal - 4 years, 8 months ago

@Agnishom Chattopadhyay

In fact I think I have found the best initial values of x0 to start with to ensure fastest convergence rate.

Soumava Pal - 4 years, 8 months ago

Log in to reply

That's really cool! Analysis is not my best skill but I certainly find this interesting.

Agnishom Chattopadhyay - 4 years, 8 months ago

Log in to reply

Same here!!!

For one interval we can start with 1/x and for the other we can start with lnx.

Soumava Pal - 4 years, 7 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...