5

Find all real solutions of the logarithmic equation $$\log_2 x + \log_x 9 = \log_2 (x+9)$$ without using derivatives.

My teacher said that this problem can be solved using only basic facts on logarithmic functions and increasing/decreasing functions, so there is no need to use derivatives to verify all solutions.


For someone who is interested in a solution using derivatives, we note that $x > 0$ and $x \ne 1$. By direct manipulations and substituting $x = 3^t$, we get $$ \begin{align} \log_2(3^t) + \log_{3^t}(9) &= \log_2(3^t+9) \\ \frac{2}{t} &= \;\log_2 \left(\frac{3^t+9}{3^t} \right) \\ t &= \frac{2\ln2}{\ln(1+3^{2-t})}. \end{align} $$ Letting $f(t) = \frac{2\ln 2}{\ln(1+3^{2-t})}$, we find that $$ f''(t)=\frac{2(\ln 2)(\ln 3)^2(18-3^t\ln(1+3^{2-t}))}{(3^t+9)^2(\ln(1+3^{2-t}))^3}. $$ We use the fact that $\ln(1+x) \le x$ for all $x \in \mathbb{R} \implies 18-3^t\ln(1+3^{2-t}) \ge 18-3^t(3^{2-t}) = 9>0$. Therefore, $f''(t) > 0$ for all $t \in \mathbb{R}$, and we can conclude that $f$ is strictly convex on $\mathbb{R}$.

Since $f$ is strictly convex, the equation $f(t)=t$ must have at most $2$ solutions. By inspection, $t \in \{1,2\}$ satisfies this equation. Hence, the equation $\log_2 x + \log_x 9 = \log_2 (x+9)$ has $2$ solutions, which are $x \in \{3^1,3^2\} = \{3,9\}$.

Integreek
  • 8,530
Nuddun
  • 321
  • 1
    "....without derivatives" seems unlikely. It's really hard to grab a hold of the fact that both $3,9$ are solutions but that in all other cases the functions maintain some distance between one another without appealing to repeated rate of changes like a second derivative. It seems that it is precisely concavity itself that allows one to be able to show that these distances minimize (to equality) again even after $x=3.$ – Derek Luna Jun 21 '25 at 02:59
  • 4
    Change of base to base $e$, and algebra, assuming $x \neq 1$, yields $$ \ln(9) \ln(2) = \ln(x) \ln\left(\frac{x+9}{x}\right) $$

    First $x = 9$ is a natural solution. Note $9$ is a perfect square so this is also equal to $\ln(3) \ln(4)$. This yields the other natural solution $x = 3$. Not sure about the increasing and decreasing though.

    – Pavan C. Jun 21 '25 at 03:18
  • @PavanC. I would argue that these solutions are natural even if only using the original expression provided. – Derek Luna Jun 21 '25 at 15:20
  • 1
    @DerekLuna I agree, if you have a good feel for $\log_x$ and $\log_2$ and how they may play together, you would see those natural solutions immediately. But if I'm in a precalculus class (maybe just learned logs or how to use them properly), I would do some algebraic manipulation as described. – Pavan C. Jun 21 '25 at 17:18

3 Answers3

0

The following is simply a speculation that this is not possible

It appears that any such proof may simply be a reformulation of various ideas related to derivatives (and $n$th derivatives in general).

Consider that we must, at some stage, have the following cases:

$1 \leq x < 3, 3 <x< 9,$ and $x >9$.

These intervals are required to analyze the behavior of the functions near the exact solutions.

$\log_2 x + \log_x 9 = \log_2 (x+9) \implies \log_2\frac{x}{x+9} +\log_x9 = 0.$

Heuristically, one may propose that the cases we considered imply the following, respectively: $\log_2\frac{x}{x+9} +\log_x9 > 0, \log_2\frac{x}{x+9} +\log_x9 < 0,\log_2\frac{x}{x+9} +\log_x9 > 0$.

Thus, as an example, we may consider $x > 9$. We want to somehow show that $x>9 \implies \log_2\frac{x}{x+9} +\log_x9 > 0$.

Notice that $\log_2\frac{x}{x+9}< 0, \log_x9 >0$, and that both functions approach $0$.

In essence, the only way to show the sum is nonzero on this interval, and on any entire interval, is to show that one of them approaches $0$ faster than the other, which is equivalent to using derivatives.

Derek Luna
  • 2,869
  • Your last paragraph is pessimistic. You want to show that the equation$$\ln(x)(\ln(x+9)-\ln(x))=\ln(2)\ln(9),$$has no solutions with $x>9$. It is straightforward to show that$$\ln(x+9)-\ln(x)<\sum_{k=0}^8\frac{1}{x+k}<\frac{9}{x},$$so then it suffices to show that$$9\frac{\ln(x)}{x}<\ln(2)\ln(9).$$Of course $\frac{\ln(x)}{x}$ is decreasing for $x>e$, and for $x=18$ you already have$$9\frac{\ln(x)}{x}=9\frac{\ln(18)}{18}=\frac12\ln(18)<\ln(2)\ln(9).$$This is based on a the very crude upper bound $\frac{9}{x}$. With some care the upper bound $\sum_{k=0}^8\frac{1}{x+k}$ may get you down to $x=9$. – Servaes Jun 21 '25 at 20:05
  • 1
    Is using the decreasing behavior of a known ratio of functions not just masking a derivative? How would this be described, proven, and/or understood without appealing to derivatives? It’s decreasing precisely because its denominator (a function) increases faster than its numerator (a function). – Derek Luna Jun 21 '25 at 20:49
  • Certainly not. Taking derivatives involves taking a limit of a ratio. I do not need to define a derivative to know that a function like $\frac{x^2+x+1}{3x+1}$ is increasing for $x>1$. Of course one may wonder how $\ln(x)$ is defined, without using derivatives (or integrals?). – Servaes Jun 22 '25 at 06:15
  • In particular, to show that $\frac{\ln(x)}{x}$ is decreasing, it suffice to show that $x<y$ implies $x^y>y^x$. Setting $y=(1+c)x$ with $c>0$ this is equivalent to $x>(1+c)^{\tfrac{1}{c}}$ for all $c>1$. – Servaes Jun 22 '25 at 06:25
  • @Servaes Of course, if you disagree, you are welcome to provide a full answer which includes all details... no one seems to be stopping you from doing that. – Derek Luna Jun 22 '25 at 15:00
  • However, I do agree with your point as it pertains to rational functions. – Derek Luna Jun 22 '25 at 17:18
  • I don't agree with your assessment that it is likely impossible. That is not an answer to the original question. Which is why I posted a comment here. I do not have an answer to the original question. – Servaes Jun 22 '25 at 17:20
  • That's perfectly fine, that's why it's titled as a speculation instead of assessment or proof. – Derek Luna Jun 22 '25 at 17:27
  • @Servaes I believe you can define $e$ (and hence $\ln(x)$) with limits without getting into derivatives/integrals – Debalanced Jun 22 '25 at 21:52
0

This is an unfinished draft. Perhaps it contains some ideas that someone else can use towards a complete proof. So far this draft shows that any solution must satisfy $e<x\leq9$.

Let $x$ be a positive real number such that $$\log_2(x)+\log_x(9)=\log_2(x+9).\tag{1}$$ For $\log_x(9)$ to make sense we must have $x\neq1$. We can convert to natural logarithms by noting that $$\frac{\ln(9)}{\ln(x)}=\log_x(9)=\log_2(x+9)-\log_2(x)=\frac{\ln(x+9)-\ln(x)}{\ln(2)},$$ and hence we have $$\ln(x)\Big(\ln(x+9)-\ln(x)\Big)=\ln(2)\ln(9).\tag{2}$$ To find all solution without using derivatives, it matters to know your definition of the natural logarithm for positive reals. I'll take it to be the unique inverse of the exponential function $x\ \mapsto\ e^x$. I'll use the following two facts about the exponential function, proofs can be found elsewhere on this forum:

Observation 1: The function $f(t):=\left(1+\frac1t\right)^t$ is strictly increasing with $\lim_{t\to\infty}f(t)=e$.

Proof. See for example here.$\qquad\square$

Observation 2: The function $g(t):=\left(1+\frac1t\right)^{t+1}$ is strictly decreasing with $\lim_{g\to\infty}f(t)=e$.

Proof. See for example here.$\qquad\square$

Lemma 3: For all $t>0$ we have $$\ln(t+1)-\ln(t)<\frac1t.$$

Proof. By elementary algebraic manipulation this is equivalent to $$\ln\left(1+\frac1t\right)<\frac1t \qquad\text{ and hence to }\qquad e>\left(1+\frac1t\right)^t, $$ which holds for all $t>0$ by Observation 1. $\qquad\square$

Substituting $t=\tfrac x9$ into equation $(2)$ we get $$\ln(2)\ln(9)=\ln(9t)\Big(\ln(t+1)-\ln(t)\Big),$$ and hence applying Lemma 3 yields $$\ln(2)\ln(9)<\frac{\ln(9t)}{t}, \qquad\text{ or equivalently}\qquad \frac{\ln(x)}{x}>\frac{\ln(2)\ln(9)}{9}.$$

Lemma 4: The function $\frac{\ln(x)}{x}$ is strictly increasing for $x<e$, and strictly decreasing for $x>e$.

Proof. Let $x>y>e$. Then we want to show that $\frac{\ln(x)}{x}<\frac{\ln(y)}{y}$, or equivalently $x^y<y^x$. Writing $x:=(1+c)y$ with $c>0$, it suffices to shows that $$\left((1+c)y\right)^y<y^{(1+c)y},\qquad\text{ or equivalently }\qquad y>(1+c)^{\tfrac1c}.$$ With Observation 1 this follows immediately from the fact that $y>e$. The proof for $y<x<e$ is entirely analogous.$\qquad\square$

This shows that the left hand side of equation $(2)$ is monotonically increasing for $x<e$, because $$2^{\ln(9)}=\left(1+\frac9x\right)^{\ln(x)}=\left(\left(1+\frac9x\right)^{\frac{x}{9}}\right)^{9\frac{\ln(x)}{x}}.$$ The base $\left(1+\frac9x\right)^{\frac{x}{9}}$ is strictly increasing by Observation 1, and the exponent is strictly increasing for $x<e$ by Lemma 4. So for $x\leq e$ we have $$\left(1+\frac9x\right)^{\ln(x)}\leq \left(1+\frac9e\right)^{\ln(e)}=1+\frac9e.$$ It is a matter of elementary algebra to show that $2^{\ln(9)}>1+\tfrac9e$, and so we conclude that any solution to equation $(1)$ must have $x>e$.

Similarly we can obtain a better upper bound by using Observation 2 and a slight variation on Lemma 4:

Lemma 5: The function $\frac{\ln(x)}{x+9}$ is strictly decreasing for $x>9$.

Proof. Let $x>y>9$. Then we want to show that $\frac{\ln(x)}{x+9}<\frac{\ln(y)}{y+9}$, or equivalently $x^{y+9}<y^{x+9}$. Writing $x:=(1+c)y$ with $c>0$, it suffices to shows that $$\left((1+c)y\right)^{y+9}<y^{(1+c)y+9},\qquad\text{ or equivalently }\qquad y^{\tfrac{y}{y+9}}>(1+c)^{\tfrac1c}.$$ By Observation 1 we have $(1+c)^{\tfrac1c}<e$ and it is clear that $y^{\tfrac{y}{y+9}}$ is strictly increasing for $y>1$. For $y=9$ we have $y^{\tfrac{y}{y+9}}=3>e$, which completes the proof.$\qquad\square$

Now as before we can decompose the left hand side of $(2)$ as follows: $$2^{\ln(9)}=\left(1+\frac9x\right)^{\ln(x)}=\left(\left(1+\frac9x\right)^{\frac{x+9}{9}}\right)^{9\frac{\ln(x)}{x+9}}.$$ As before, the base $\left(1+\frac9x\right)^{\frac{x+9}{9}}$ is strictly decreasing by Observation 2, and the exponent is strictly decreasing for $x>9$ by Lemma 5. For $x=9$ we of course find that $\left(1+\frac9x\right)^{\ln(x)}=2^{\ln(9)}$, so this shows that there are no solutions with $x>9$.

Some remarks

  1. The proof of Lemma 5 shows that we can improve this upper bound for $x$ a bit; we readily see that $y^{\tfrac{y}{y+9}}>e$ for $y=9$, but it is also clear that this is true for some smaller values of $y$.
    With the help of a computer this goes down to $y\approx8.2$ and so
    the only solution with $x>8.2$ is $x=9$.
  2. The proof of Observation 2 can be sharpened to show that $h(t)=(1+\tfrac1t)^{t+\tfrac12}$ is strictly decreasing, again with limit $e$ of course. Then we can use the same argument as in the proofs of Lemma's 4 and 5 for the function $\frac{\ln(x)}{x+9/2}$, to show that it is strictly decreasing for $x>6$ (or even $x>5.86$ with the help of a computer), and hence conclude that $x=9$ is the unique solution with $x>6$.
Servaes
  • 67,306
  • 8
  • 82
  • 171
0

$\newcommand{\deq}{\stackrel{\triangle}{=}}$ $\DeclareMathOperator{\sgn}{sgn}$ Write $$g(t)\deq(1+3^{2-t})^t\text{.}$$ Evidently, $g(t)<1$ if $t<0$ and $g(t)=(1+3^{2-t})^t <1 + \frac{t3^{2-t}}{1-(t-1)3^{2-t}}=1 + \frac{t}{3^{t-2}-(t-1)} <1 + \frac{t}{1+2t-4-t+1}= 1 +\frac{t}{t-2}<4$ if $t>3$ by Bernoulli's inequality, so that the goal is to show that $$\begin{align} g(t)&>4 & t&\in(1,2)\text{,} \\ g(t)&<4 & t&\in(0,1)\cup(2,3)\text{.} \end{align}$$ Our strategy is to produce a rational approximation $h(t)$ to $g(t)$ such that

$$\begin{align} g(t)&>h(t) & t&\in(1,2)\text{,} \\ g(t)&<h(t) & t&\in(0,1)\cup(2,3)\text{,} \end{align}$$ then to use a computer algebra implementation of Vincent's theorem to decide whether $$\begin{align} h(t)&>4 & t&\in(1,2)\text{,} \\ h(t)&<4 & t&\in(0,1)\cup(2,3)\text{.} \end{align}$$

To produce such an approximation without differential calculus, we invoke the following fact:

The function $\digamma^x(s)\deq (1+x)^s$ is strictly absolutely monotone if $x>0$ and strictly completely monotone if $x<0$.

(A function $s\mapsto \phi(s)$ is said to be strictly absolutely monotone if all of its divided differences $\Delta^n\phi[s_0,s_1,\ldots,s_n]$ at pairwise-distinct inputs $\{s_0,s_1,\ldots,s_n\}$ are strictly positive. It is strictly completely monotone if $s\mapsto \phi(-s)$ is strictly absolutely monotone. Strict absolute monotonicity implies being positive, strictly increasing, and strictly convex.)

The consequence of this fact that we use is that the convergent $p^n_x(s)$ of degree $n-1$ of the Newton series of $\digamma^x(s)$ $$p^x_n(s)\deq \sum_{k=0}^{n-1}x^k\binom{s}{k} $$ has a remainder of predictable sign: $$\begin{align}\frac{(\sgn x)^n}{\binom{s}{n}}\left(\digamma^x(s)-p^x_n(s)\right)\propto (\sgn x)^n\Delta^n\digamma^x[s,0,\ldots,n-1] > 0\text{.} \end{align}$$

So define $$h(t)\deq \left. \frac{p_6^z(t+2)}{(1+z)^2}\right\rvert_{z=\tfrac{1}{3}p_4^2(3-t)}\text{.}$$

It follows, then, that $h$ bounds $g$ in the "right way" as described above, as this plot illustrates:

Plot of <span class=$g(t)$ and $h(t)$" />

Once the CAS has checked that $h(t)>4$ on $t\in(1,2)$ and $h(t)<4$ on $t\in(0,1)\cup(2,3)$ the demonstration is finished.

K B Dave
  • 9,458