5
$\begingroup$

We are given $n\in \mathbb N^+$ and $p\in[\frac{1}{2},\frac{n+1}{n+2}]$.

Our goal is to find $t\in[0,1]$ such that $$(1-p)\sum_{i=0}^{n}t^i = p\sum_{i=0}^{n}(1-t)^i$$

Is there a closed-form solution $t(n,p)$?

How about a close formula for some non-trivial $p$, e.g. for $g(n)\triangleq t(n,0.6)$?


A few observations:

  • This is equivalent, for $t\neq0,1$, to: $$(1-p)t(1-t^{n+1})=p(1-t)(1-(1-t)^{n+1})$$

  • $\forall n:t(n,\frac{1}{2})=\frac{1}{2}$

  • $\forall p:t(1,p)=3p-1$

  • $\forall n:t(n,\frac{n+1}{n+2})=1$

  • $\forall n:t(n,p)$ is monotonically increasing in $p$.

  • $\forall p:t(2,p)=\frac{2 p + 1-\sqrt{-3+28 p-28 p^2}}{4 p - 2}$


If this is not possible, is it possible to bound it with simple function? e.g. I think I showed $$p\leq t(n,p)\leq \frac{(n+2)p - 1}{n}$$

Which works great for large $n$, but not so much for small values.

Can we give tighter bound for $t$?

$\endgroup$
4
  • 1
    $\begingroup$ Probably not. Most irreducible polynomials of degree $n>4$ have a non-solvable Galois group. Of course it would help if you defined more precisely what you mean by "closed form". Solvability in radicals is equivalent to a solvable Galois group for the polynomial. $\endgroup$
    – GH from MO
    Nov 17, 2014 at 14:50
  • 1
    $\begingroup$ I think I found closed form for n in {3,5}. $\endgroup$
    – joro
    Nov 17, 2014 at 15:35
  • $\begingroup$ @joro - I have a closed form for $t(3,p)$, would love to hear about $t(5,p)$ ! $\endgroup$
    – R B
    Nov 17, 2014 at 15:36
  • 3
    $\begingroup$ The Galois group of the equation for $p=3/5$ and $n=7$ is $S_7$, which is not soluble. So don't expect a solution by radicals for $n=7$. $\endgroup$ Nov 17, 2014 at 17:26

2 Answers 2

2
$\begingroup$

Too long for comment, here is partial answer for $n=5$ per R B's request.

According to Maple:

n:=5:q:=(1-p)*sum(t^i,i=0..n)-p*sum( (1-t)^i,i=0 .. n):so:=solve(q,t):

so[1] and so[2] are complex and this is messy for latex, so:

convert(so[3],string);convert(so[4],string);convert(so[5],string);

"1/6*(-4116*p^2-28+1428*p+2744*p^3+12*(-43218*p^3+21903*p^2+21609*p^4-294*p\
    +9)^(1/2))^(1/3)-6*(49/9*p+2/9-49/9*p^2)/(-4116*p^2-28+1428*p+2744*p^3+\
    12*(-43218*p^3+21903*p^2+21609*p^4-294*p+9)^(1/2))^(1/3)-2/3+7/3*p"

"-1/12*(-4116*p^2-28+1428*p+2744*p^3+12*(-43218*p^3+21903*p^2+21609*p^4-294\
    *p+9)^(1/2))^(1/3)+3*(49/9*p+2/9-49/9*p^2)/(-4116*p^2-28+1428*p+2744*p^\
    3+12*(-43218*p^3+21903*p^2+21609*p^4-294*p+9)^(1/2))^(1/3)-2/3+7/3*p+1/\
    2*I*3^(1/2)*(1/6*(-4116*p^2-28+1428*p+2744*p^3+12*(-43218*p^3+21903*p^2\
    +21609*p^4-294*p+9)^(1/2))^(1/3)+6*(49/9*p+2/9-49/9*p^2)/(-4116*p^2-28+\
    1428*p+2744*p^3+12*(-43218*p^3+21903*p^2+21609*p^4-294*p+9)^(1/2))^(1/3\
    ))"

"-1/12*(-4116*p^2-28+1428*p+2744*p^3+12*(-43218*p^3+21903*p^2+21609*p^4-294\
    *p+9)^(1/2))^(1/3)+3*(49/9*p+2/9-49/9*p^2)/(-4116*p^2-28+1428*p+2744*p^\
    3+12*(-43218*p^3+21903*p^2+21609*p^4-294*p+9)^(1/2))^(1/3)-2/3+7/3*p-1/\
    2*I*3^(1/2)*(1/6*(-4116*p^2-28+1428*p+2744*p^3+12*(-43218*p^3+21903*p^2\
    +21609*p^4-294*p+9)^(1/2))^(1/3)+6*(49/9*p+2/9-49/9*p^2)/(-4116*p^2-28+\
    1428*p+2744*p^3+12*(-43218*p^3+21903*p^2+21609*p^4-294*p+9)^(1/2))^(1/3\
    ))"

Experimentally so[3] is real and in the requested range.

$\endgroup$
3
  • 3
    $\begingroup$ For $n=5$ you're solving the cubic $${t}^{3}-7\,p{t}^{2}+2\,{t}^{2}+7\,pt+2\,t-7\,p+1$$ $\endgroup$ Nov 17, 2014 at 18:16
  • $\begingroup$ @RobertIsrael for n=5 I think I am solving higher degree than cubic, which might be reducible. $\endgroup$
    – joro
    Nov 17, 2014 at 18:26
  • 2
    $\begingroup$ Yes, the point is that it factors, and the cubic factor is the one that contains $p$, so that's what you're actually solving. $\endgroup$ Nov 17, 2014 at 18:35
2
$\begingroup$

You might find this useful.
Write the equation as $$ p = \dfrac{t (t^{n+1}-1)}{(1-t)^{n+2} + t^{n+2}-1} = \dfrac{(1-s)((1-s)^{n+1}-1)}{s^{n+2}+(1-s)^{n+2}-1}$$ where $s = 1-t$. If $n \ge 3$, I get $$ p = \dfrac{n+1}{n+2} - \dfrac{(n+1)}{2(n+2)} s - \dfrac{(n^2+4n+3)}{12(n+2)} s^2 + \ldots $$ so we should have $$ t > \dfrac{2(n+2)p}{n+1} - 1 $$ for $p$ near $(n+1)/(n+2)$. Indeed this appears to hold for $1/2 \le p \le 1$ and all $n \ge 1$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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