I hope it is appropriate to ask this question here:

One formulation of the abc-conjecture is

$$ c < \text{rad}(abc)^2$$

where $\gcd(a,b)=1$ and $c=a+b$. This is equivalent to ($a,b$ being arbitrary natural numbers):

$$ \frac{a+b}{\gcd(a,b)} < \text{rad}(\frac{ab(a+b)}{\gcd(a,b)^3})^2$$

Let $d_1(a,b) = 1- \frac{\gcd(a,b)^2}{ab}$ which is a proven metric on natural numbers. Let $d_2(a,b) = 1- 2 \frac{\gcd(a,b)}{a+b}$, which I suspect to be a metric on natural numbers, but I have not proved it yet. Let $$d(a,b) = d_1(a,b)+d_2(a,b)-d_1(a,b)d_2(a,b) = 1-2\frac{\gcd(a,b)^3}{ab(a+b)}$$

Then we get the equivalent formulation of the inequality above:

$$\frac{2}{1-d_2(a,b)} < \text{rad}(\frac{2}{1-d(a,b)})^2$$

which is equivalent to :

$$\frac{2}{1-d_2(a,b)} < \text{rad}(\frac{1}{1-d_1(a,b)}\cdot\frac{2}{1-d_2(a,b)} )^2$$

My question is if one can prove that $d_2$ and $d$ are distances on the natural numbers (without zero)?

Result: By the answer of @GregMartin, $d_2$ is a metric. By the other answer $d$ is also a metric.

Edit: By "symmetry" in $d_1$ and $d_2$, this interpretation also suggests that the following inequality is true , which might be trivial to prove or very difficult or might be wrong and may be of use or not in number theory:

$$\frac{1}{1-d_1(a,b)} < \text{rad}(\frac{2}{1-d(a,b)})^2$$

which is equivalent to

$$ \frac{ab}{\gcd(a,b)^2} < \text{rad}(\frac{ab(a+b)}{\gcd(a,b)^3})^2$$

(This is not easy to prove, as the $abc$ conjecture $c=a+b < ab < \text{rad}(abc)^2$ would follow for all $a,b$ such that $a+b < ab$.)

Second edit: Maybe the proof that $d_2,d$ are distances can be done with some sort of metric transformation, for example maybe with a Schoenberg transform (See 3.1, page 8 in https://arxiv.org/pdf/1004.0089.pdf) The idea, that this might be proved with a Schoenberg transform comes from the fact that:

$$d_1(a,b) = 1-\exp(-\hat{d}(a,b))$$ so $d_1$ is a Schoenberg transform of $\hat{d}(a,b) = \log( \frac{ab}{\gcd(a,b)^2}) = \log( \frac{\text{lcm}(a,b)}{\gcd(a,b)})$ which is proved to be a metric (see Encyclopedia of Distances, page 198, 10.3 )

Third edit: Here is some Sage Code to test the triangle inequality for triples (a,b,c) up to 100:

def d1(a,b):
    return 1-gcd(a,b)**2/(a*b)

def d2(a,b):
    return 1-2*gcd(a,b)/(a+b)

def d(a,b):
    return d1(a,b)+d2(a,b)-d1(a,b)*d2(a,b)

X = range(1,101)
for a in X:
    for b in X:
        for c in X:
            if d2(a,c) > d2(a,b)+d2(b,c):
                print "d2",a,b,c
            if d(a,c) > d(a,b)+d(b,c):
                print "d",a,b,c

so far with no counterexample.

Related: An inequality inspired by the abc-conjecture and two questions

  • 1
    $\begingroup$ Your speculated "by symmetry" inequality is not valid: take $a=2^j$ and $b=3^k$. $\endgroup$ – Greg Martin Sep 30 at 1:23
  • $\begingroup$ You may write something like $\ \mbox{rad}^2(x)\ $ rather than $\ \mbox{rad}(x)^2$ $\endgroup$ – Wlod AA Sep 30 at 3:53
  • 1
    $\begingroup$ @WlodAA The question clearly says one formulation . $\endgroup$ – orgesleka Sep 30 at 4:10
  • 1
    $\begingroup$ Thus, is it or is it not? Yes or No? $\endgroup$ – Wlod AA Sep 30 at 4:17
  • 1
    $\begingroup$ @GregMartin: Can there be done anything to include $a=2^j$ and $b=3^k$ in terms of the constant $K_{\epsilon}$ or $\epsilon$ in $1/(1-d_1)< K_{\epsilon} \text{rad}(2/(1-d))^{1+\epsilon}$? $\endgroup$ – orgesleka Sep 30 at 14:18

$d_2$ is indeed a metric. Abbreviating $\gcd(m,n)$ to $(m,n)$, we need to show that \begin{align*} 1-\frac{2(a,c)}{a+c} &\le 1-\frac{2(a,b)}{a+b} + 1-\frac{2(b,c)}{b+c} \end{align*} or equivalently \begin{align*} \frac{2(a,b)}{a+b} + \frac{2(b,c)}{b+c} &\le 1 + \frac{2(a,c)}{a+c}. \end{align*} Furthermore, we may assume that $\gcd(a,b,c)=1$, since we can divide everything in sight by that factor.

Note that if $a=(a,b)\alpha$ and $b=(a,b)\beta$ with $(\alpha,\beta)=1$, then $\frac{2(a,b)}{a+b} = \frac2{\alpha+\beta}$. The only unordered pairs $\{\alpha,\beta\}$ for which this is at least $\frac12$ are $\{1,1\}$, $\{1,2\}$, and $\{1,3\}$. Further, if neither $\frac{2(a,b)}{a+b}$ nor $\frac{2(b,c)}{b+c}$ is at least $\frac12$, then the inequality is automatically valid because of the $1$ on the right-hand side.

This leaves only a few cases to check. The case $\{\alpha,\beta\} = \{1,1\}$ (that is, $a=b$) is trivial. The case $\{\alpha,\beta\} = \{1,2\}$ (that is, $b=2a$) can be checked: we have $(a,c)=\gcd(a,2a,c)=1$, and so the inequality in question is \begin{align*} \frac23 + \frac{2(2,c)}{2a+c} &\le 1 + \frac2{a+c}, \end{align*} or equivalently $$ \frac{(2,c)}{2a+c} \le \frac16 + \frac1{a+c}; $$ there are only finitely many ordered pairs $(a,c)$ for which the left-hand side exceeds $\frac16$, and they can be checked by hand.

The proof for the case $\{\alpha,\beta\} = \{1,3\}$ (that is, $b=3a$) can be checked in the same way, as can the cases $a=2b$ and $a=3b$.

  • 1
    $\begingroup$ Thanks for your answer $\endgroup$ – orgesleka Sep 30 at 4:01

Not an answer but an observation.

Set $r_2(a,b,c)=d_2(a,c)/(d_2(a,b)+d_2(b,c))$ (when defined), and similarly for $r(a,b,c)$. Then Greg Martin's proof shows that the values of $r_2$ should be discrete, and indeed experimentally the values are in decreasing order


The same experiment done for $d$ gives


Thus, apart from trivial cases such as $a=b$ one should have the stronger triangle inequality $d(a,c)\le0.675(d(a,b)+d(b,c))$.

  • 1
    $\begingroup$ Thank you for this observation $\endgroup$ – orgesleka Sep 30 at 10:25

$d$ is also a metric. Proof:

First let us call a metric on natural numbers $d$ such that $d(a,b)<1$ and $d(a,b)$ is a rational number for all $a,b$ a "rational metric". Second let $d_1,d_2$ be two rational metrics such that if we set $d=d_1+d_2-d_1 d_2$ then for all $a \neq c, a \neq b$ we have $d(a,b)+d(a,c)>1$. If this is the case for $d_1,d_2$ we will call $d_1$ and $d_2$ "paired". If $d_1,d_2$ are such paired rational metrics, then $d=d_1+d_2-d_1 d_2$ is a metric. Proof:

1) $d(a,b) = 0$ iff $0 \le d_1(a,b)(1-d_2(a,b)) = -d_2(a,b) \le 0$ hence since $1-d_2(a,b)>0$ we must have $d_1(a,b) = 0$ hence $a=b$. If on the other hand $a=b$ then plugging this in $d$ and observing that $d_1(a,b)=d_2(a,b)=0$ gives us $d(a,b)=0$.

2) $d(a,b) = d(b,a)$ since $d_i(a,b) = d_i(b,a)$ for $i = 1,2$.

3) Triangle inequality: If $a=c$ or $a=b$ the triangle inequality is fullfilled and becomes an equality because of 1) : $d(b,c) \le d(a,b)+d(a,c)$ First observe that $d(x,y) < 1$ for all $x,y$. Let therefore $a\neq c, a\neq b$. Since $d_1,d_2$ are paired rational metrics we have: $d(b,c) < 1 < d(a,c)+d(a,b)$ and the triangle inequality is proved.

This proves also that $d$ is a rational metric (if $d_1,d_2$ are paired rational metrics.)

What remains to show is that $d_2(a,b) = 1-\frac{2 \gcd(a,b)}{a+b}$, $d_1(a,b) = 1-\frac{\gcd(a,b)^2}{ab}$ are paired (rational) metrics, hence it remains to show that $d(a,b) = 1- \frac{2 \gcd(a,b)^3}{ab(a+b)}$ satisfies:

$$d(a,c)+d(a,b)>1, \text{ whenever } a\neq c, a \neq b$$

The last inequality is equivalent to, after some algebra:

$$\frac{abc(a+b)(a+c)}{2} - \gcd(a,b)^3c(a+c) - \gcd(a,c)^3b(a+b)>0$$

Let $U=\gcd(a,b,c)$. Then there exist natural numbers $R,S,T,A,B,C$ such that:

$$RU = \gcd(a,b), SU = \gcd(a,c), TU = \gcd(b,c), a = RSUA, b = RTUB, c = STUC$$

Plugging this in the last inequality and after some algebra, we find:

$$1/2*(A^3*B*C*R^2*S^2*T + A^2*B^2*C*R^2*S*T^2 + A^2*B*C^2*R*S^2*T^2 + A*B^2*C^2*R*S*T^3 - 2*A*C*R^2 - 2*A*B*S^2 - 2*C^2*R*T - 2*B^2*S*T)*R^2*S^2*T*U^5 > 0 $$

We can pair each of the positive summand with a negative summand to give for example:

$$(A^3*B*C*R^2*S^2*T-2*A*C*R^2)=(A^2*B*S^2*T - 2)*A*C*R^2$$

The condition $a \neq b$ translates to $SA \neq TB$ and similarily $a \neq c$ translates to $RA \neq TC$. Suppose that $A^2*B*S^2*T - 2 \le 0$. The case $A^2*B*S^2*T=1$ contradicts $SA \neq TB$. Hence we can only have at most $A^2*B*S^2*T=2$ which leads to $A=S=1$, $BT=2$ and plugging this in the definition of $a,b$ we get $b=2a$ and $d(a,b)=\frac{2}{3}$.

Now we must show that the other pairings give the desired result:

$$( A^2*B^2*C*R^2*S*T^2-2*B**2*S*T)=(A^2*C*R^2*T - 2)*B^2*S*T$$ A similar argument to the above leads to: If $A^2*C*R^2*T = 2$, then $A=R=1$, $CT=2$ which leads to (with $S=A=1$) $a=RSUA=U,b=RTUB=2U=2a,c=STUC=2U=2a$ and it follows that $d(a,c)=\frac{2}{3}$, so $d(a,b)+d(a,c)=\frac{4}{3}>1$, and this case is done.

If $A^2*C*R^2*T > 2$ and $A^2*B*S^2*T=2$ then $$1/2*(A^3*B*C*R^2*S^2*T + A^2*B^2*C*R^2*S*T^2 + A^2*B*C^2*R*S^2*T^2 + A*B^2*C^2*R*S*T^3 - 2*A*C*R^2 - 2*A*B*S^2 - 2*C^2*R*T - 2*B^2*S*T)*R^2*S^2*T*U^5 > 0 $$ is true.

If $A^2*C*R^2*T > 2$ and $A^2*B*S^2*T>2$ then $$1/2*(A^3*B*C*R^2*S^2*T + A^2*B^2*C*R^2*S*T^2 + A^2*B*C^2*R*S^2*T^2 + A*B^2*C^2*R*S*T^3 - 2*A*C*R^2 - 2*A*B*S^2 - 2*C^2*R*T - 2*B^2*S*T)*R^2*S^2*T*U^5 > 0 $$ is true. This shows, that $d_1,d_2$ are paired metrics and completes the proof.


This question has already very good answers. I justed wanted to highlight that it is possible to shorten the proofs, using the following:

If $X_a = \{ a/k | 1 \le k \le a \}$ then $X_a \cap X_b = \gcd(a,b)$, which is straightforward to prove. Then $d_1(a,b) = 1-\gcd(a,b)^2/(ab) = 1-|X_a \cap X_b|^2 / (|X_a||X_b|)$ is the squared cosine metric (see Encyclopedia of Distances) and $d_2(a,b) = 1-2\gcd(a,b)/(a+b) = 1-2|X_a \cap X_b| / (|X_a|+|X_b|)$ is the Sorensen Metric (Encyclopedia of Distances). Hence $d_1,d_2$ are metrics of the form $d_i = 1- s_i$ where $s_i$ is a similarity. But then $s=s_1 \cdot s_2$ is also a similarity and $d=d_1 +d_2 -d_1 d_2 = 1-s=1-s_1 s_2$ is a metric.


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.