Just wondering is there any way to solve this equation using some algebraic or calculus tricks but without graphs.

Solve for $n$ $$2^{2n-3}=32(n-1)$$

I solved this intuitively or to be more specific we can solving sigh equations by substituting $n=1,2,3,...$ but what if the solution of $n$ not belongs to integer set.

Any ideas ? I just want a general algebraic approach.

  • 2
    $\begingroup$ Look up the lambert W function $\endgroup$ – Shogun May 6 at 4:44
  • 1
    $\begingroup$ If there's a solution that isn't an integer, then there's no algebraic trick to find it. The best you can do is numerical methods, such as Newton's Method, q.v. $\endgroup$ – Gerry Myerson May 6 at 4:44
  • 1
    $\begingroup$ @GerryMyerson Small correction: the similar equation $4^{2n-1}=32(n-1)$ has a non integer root ($n=\frac 32$) that's easy to find by inspection. I think you meant irrational rather than integral. $\endgroup$ – Deepak May 6 at 6:25
  • $\begingroup$ @Deepak, yes, careless of me. Though I was referring to the particular equation OP is asking about. $\endgroup$ – Gerry Myerson May 6 at 6:53

The solutions are $$1 - \frac{W(-\ln(2)/32)}{2 \ln(2)} $$ where $W$ is a branch of the Lambert W function. The real solutions are obtained with the principal branch (approximately $1.015974889$) and the "$-1$" branch ($5$).

EDIT: In general, if you suspect an equation involving exponentials may have a solution of the form $x = a + b W(c)$, the approach is to try to put it into the form $z e^z = c$ with a substitution $x = a + b z$. In your case, $2^{2x-3} = 32(x-1)$ can be written as $$ (x-1) \exp((3-2x)\ln(2)) = 1/32$$ If $x = 1 - y$, that's $$ - 64 y \exp(2 \ln(2) y) = 1 $$ so with $z = 2 \ln(2) y$ it's $$ z e^z = - \frac{\ln(2)}{32} $$ And then $z = W(-\ln(2)/32)$. Since $-e^{-1} < -\ln(2)/32 < 0$, you get real solutions with either the principal or the $-1$ branch.

  • $\begingroup$ I don't want the answer... As I said I know the answer...I want the approach. $\endgroup$ – Aman Rajput May 6 at 5:24
  • 1
    $\begingroup$ You didn‘t sound like you knew the general solution in your opening question? Did you even look up the Lambert W function? The approach is quite straightforward once you know what it is. $\endgroup$ – Kezer May 6 at 5:29
  • $\begingroup$ How do you know the answer, Aman, if you don't know an approach? $\endgroup$ – Gerry Myerson May 6 at 6:55
  • $\begingroup$ Why doesn't your answer shows that n=5 is also a solution ? $\endgroup$ – Aman Rajput May 6 at 15:23
  • $\begingroup$ @AmanRajput: It does: look at the line that says ... and the "-1" branch (5). $\endgroup$ – NickD May 6 at 17:13

Consider that you look for the zero's of function $$f(x)=2^{2x-3}-32(x-1)$$ By inspection, $f(0)=\frac{257}{8}$, $f(1)=\frac 12$, $f(2)=-30$; so, a solution close to $x=1$.

Perform a Taylor expansion around $x=1$ (this is equivalent to the first interation of Newton method); it will give $$f(x)=\frac{1}{2}+(x-1) (\log (2)-32)+O\left((x-1)^2\right)$$ Ignoring the higher order terms, this would give, as an estimate, $$x=\frac{2 \log (2)-65}{2 (\log (2)-32)}\approx 1.015970944$$

We could do better building a $[1,n]$ Padé approximant; in such a way, the estimate is obtained solving a linear equation in $(x-1)$. For example, the simplest (corresponding to the first iteration of Halley method) would be $$f(x)=\frac{\frac{1}{2}+\frac{ \left(2048+\log ^2(2)-128 \log (2)\right)}{2 (\log (2)-32)}(x-1)}{1-\frac{ \log ^2(2)}{\log (2)-32}(x-1)}$$ giving $$x=\frac{2080+\log ^2(2)-129 \log (2)}{2048+\log ^2(2)-128 \log (2)}\approx 1.015974860$$ We could continue the process and get the following results

$$\left( \begin{array}{cc} n & x_{(n)} \\ 0 & 1.0159709442173479658 \\ 1 & 1.0159748596445568773 \\ 2 & 1.0159748895153359604 \\ 3 & 1.0159748896898660072 \\ 4 & 1.0159748896907116852 \\ 5 & 1.0159748896907153030 \\ 6 & 1.0159748896907153175 \end{array} \right)$$


If we are concerned by the roots, consider that $$f'(x)=2^{2 x-2} \log (2)-32 \qquad \text{and} \qquad f''(x)=2^{2 x-1} \log ^2(2) \,\, > 0 \,\,\forall x$$ The first derivative cancels at $$x_*=1+\frac{\log \left(\frac{32}{\log (2)}\right)}{2 \log (2)}\implies f(x_*)=\frac{16 \left(1+\log \left(\frac{\log (2)}{32}\right)\right)}{\log (2)} <0$$ So, two roots. To approximate them, build a Taylor series around $x_*$ to get $$f(x)=f(x_*)+\frac 12 f''(x_*) (x-x_*)^2+O((x-x_*)^2)$$ and solve the quadratic; this will give $x_1=2.05$ and $x_2=5.48$. Using thse guesses, Newton method should converge quite fast.

  • $\begingroup$ Why doesn't your solution shows that n=5 is also a valid solution ? $\endgroup$ – Aman Rajput May 6 at 15:23
  • $\begingroup$ I guess that's because you said that you found the integer solution and you wanted to find the non-integral one. $\endgroup$ – NickD May 6 at 18:01
  • $\begingroup$ @AmanRajput. Have a look at my edit. $\endgroup$ – Claude Leibovici May 7 at 3:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.