If $a, b, c, d, e, f, g, h$ are positive numbers satisfying $\frac{a}{b}<\frac{c}{d}$ and $\frac{e}{f}<\frac{g}{h}$ and $b+f>d+h$, then $\frac{a+e}{b+f} < \frac{c+g}{d+h}$.

I thought it is easy to prove. But I could not. How to prove this? Thank you.

The question is a part of a bigger proof I am working on. There are two strictly concave, positive valued, strictly increasing functions $f_1$ and $f_2$ (See Figure 1). Given 4 points $x_1$, $x_2$, $x_3$ and $x_4$ such that $x_1< x_i$, $i=2, 3,4$ and $x_4> x_i$, $i=1, 2, 3$, let $d=x_2-x_1$, $b=x_4-x_3$ $c=f_1(x_2)-f_1(x_1)$, $a=f_1(x_4)-f_1(x_3)$. And given 4 points $y_1$, $y_2$, $y_3$ and $y_4$ such that $y_1< y_i$, $i=2, 3,4$ and $y_4> y_i$, $i=1, 2, 3$, let $h=y_2-y_1$, $f=y_4-y_3$ $g=f_2(y_2)-f_2(y_1)$, $e=f_2(y_4)-f_2(y_3)$.

Since the functions are concave, we have $\frac{a}{b}<\frac{c}{d}$ and $\frac{e}{f}<\frac{g}{h}$. And I thought in this setting, it is true that $\frac{a+e}{b+f} < \frac{c+g}{d+h}$ even without the restriction $b+f>d+h$.

Figure 1

  • $\begingroup$ I added a new restriction $b+f>d+h$. Refer to Keith's nice counter example for the original posting. $\endgroup$ – Sang Jul 9 '15 at 12:24
  • 1
    $\begingroup$ Your unlucky numbers are $a,b,c,d,e,f,g,h = 1, 1, 6, 5, 10, 7, 2, 1$. May I ask you what the original purpose of this question is? (I suppose it is something related to Farey series.) $\endgroup$ – darij grinberg Jul 9 '15 at 12:26
  • $\begingroup$ Hi Darij, it is part of a proof I am struggling with. I tried to describe what I am doing by adding more to the original question. Please refer to the edited question. Thank you. $\endgroup$ – Sang Jul 9 '15 at 13:32
  • $\begingroup$ This is a great example of something that seems obvious but is actually demonstrably false. In statistics this phenomenon is well-known by the name of Simpson's paradox. $\endgroup$ – Erick Wong Jul 9 '15 at 14:34

This is false.

For example, $$\frac{1}{3} < \frac{5}{12}, \quad \frac{52}{5} < \frac{11}{1}, \quad \frac{53}{8} > \frac{16}{13}.$$

  • $\begingroup$ Is there a counterexample that uses fractions in lowest terms? $\endgroup$ – G Tony Jacobs Jul 8 '15 at 16:06
  • $\begingroup$ @GTonyJacobs I've edited my answer. $\endgroup$ – Keith Jul 8 '15 at 16:09
  • $\begingroup$ Nice counterexample. $\endgroup$ – vadim123 Jul 8 '15 at 16:15
  • $\begingroup$ It is a very nice counterexample. Is there a counter example that satisfies an added restriction, $b+f>d+h$? Or the statement is true with the added restriction? $\endgroup$ – Sang Jul 9 '15 at 12:20
  • $\begingroup$ @Sang The extra restriction is irrelevant: just rescale $a,b,e,f$ by the same large integer and you'll get $b+f > d+h$ without changing any of the values being compared. You could then add $1$ to $b$ and $f$ and have a decent chance of the resulting fractions being in lowest-terms (certainly high enough that you could just pick a larger integer that works): these are strict inequalities so they are tolerant of minute adjustments. $\endgroup$ – Erick Wong Jul 9 '15 at 14:31

The updated question (with the additional constraint $b+f>d+h$) is also false. For example,

$\frac{1}{1}<\frac{3}{2}$ and $\frac{9}{4}<\frac{5}{2}$, but $\frac{1+9}{1+4} = \frac{10}{5} = \frac{8}{4} = \frac{3+5}{2+2}$.



$$\frac{a+e}{b+f} < \frac{c+g}{d+h}$$ $$(a+e)(d+h) < (c+g)(b+f)$$ $$ad+eh+ah+ed < cb+fg+cf+gb$$ here put $ah+ed = x$ and $cf+gb = y$. Now we have $$ad+eh+x < bc+fg+y$$ Now as the given $$\frac{a}{b}<\frac{c}{d} \implies { ad<bc}$$ and $$\frac{e}{f}< \frac{g}{h} \implies{eh<fg}$$ so by adding both the above eqs $$ad+eh<bc+fg$$ and by this condition we have $$ad+eh+x<bc+fg+y$$

  • 1
    $\begingroup$ How do you justify the last step? It is certainly true if $x<y$ (in particular for $\frac{a}{b}<\frac{g}{h}$ and $\frac{c}{d}<\frac{e}{f}$), but we cannot assume that. $\endgroup$ – Klaus Draeger Jul 9 '15 at 12:50

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.