From: quasi on
On Tue, 03 Aug 2010 00:27:59 -0500, quasi <quasi(a)null.set> wrote:

>On Sat, 31 Jul 2010 05:26:14 -0700 (PDT), Gerry <gerry(a)math.mq.edu.au>
>wrote:
>
>>On Jul 31, 10:10 pm, Kent Holing <K...(a)statoil.com> wrote:
>>> Can anybody prove or disprove that the cubic
>>> r x^3 - (r^2 + r + 1)x^2 - (r^3 - 1)x + r^2(r^2 + r + 1) = 0 is irreducible for all integers r /= 0?
>>> Kent Holing,
>>> Norway
>>
>>No doubt you have observed that unless r^2 + r + 1 is squareful
>>this polynomial is irreducible by Eisenstein.
>
>Based on divisibility considerations, I think I've shown more --
>namely that if a nonzero integer r is such that the polynomial in
>question is reducible, then r^2 + r + 1 must be a perfect cube.
>
>The argument goes as follows ...
>
>Assume r is a nonzero integer such that the polynomial is reducible.
>
>Let f(x) be the univariate polynomial in x for the given r. Since f is
>cubic, the reducibility of f implies that f has a rational root. Since
>the leading coefficient is r, the polynomial g(y) defined by
>
> g(y) = r^2 f(y/r)
>
>must have an integral root. Expanding g(y) yields
>
> g(y) = y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4)
>
>By slight abuse of notation, let y be the integral root. Thus
>
> y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4) = 0.
>
>Let p be any prime factor of r^2 + r + 1.
>
>It's clear that r^2 + r + 1 divides y^3, so p|^3, hence p|y.
>
>Suppose p^k || r^2 + r + 1 (which means k is the greatest positive
>integer such that p^k | (r^2 + r + 1).
>
>Consider the powers of p which divide each of the 4 terms of g(y).
>
> p^a || y^3
> p^b || (r^2 + r + 1) y^2
> p^c || -(r^4 - r) y
> p^k || (r^6 + r^5 + r^4)
>
>Then k < c and k < b, hence we must have k = a.
>
>But a is a multiple of 3, hence so is k.
>
>Therefore r^2 + r + 1 is a perfect cube, as claimed.

Ok, I see how to prove it now.

The polynomial f(x) with r integral and nonzero can't be reducible.

Assume instead there is a nonzero integer r for which f(x) is
reducible.

Then as I argued in my prior message (quoted above), the polynomial

g(y) = y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4)

must have an integer root (where now y is an indeterminate).

But by direct substitution,

g(r^2) = r^3
g(r^2 + 1) = -r^3

So g has a real but non-integral root strictly between

r^2 and r^2+1.

Also

g(-r^2-1) = -4r^4 - 3r^3 - 6r^2 - 2r - 2
g(-r^2+1) = 4r^4 + r^3 - 2r^2

But g(-r^2-1) is negative (even for all real r).

and g(-r^2+1) is positive (since |r| >= 1).

Also, g(-r^2) = -r^3 which is nonzero.

It follows that g has a real but non-integral root strictly between

-r^2 -1 and -r^2+1.

Hence the 3rd root must be the integral root.

Let the 3 roots be y1, y2, y3, where

r^2 < y1 < r^2 + 1
-r^2-1 < y2 < -r^2+1

and y3 is the integral root.

Adding the above inequalities yields

-1 < y1 + y2 < 2

But the sum of the roots is r^2 + r + 1. It follows that

r^2 + r - 1 < y3 < r^2 + r + 2

Since y3 is an integer, we must have

y3 = r^2 + r

or

y3 = r^2 + r + 1

By divisibility considerations, it would have to be the latter one,
but forget divisibility -- we don't need it. Instead just substitute
each of the above for y3 and solve for r. As expected, the resulting
equation for r has no integral roots.

Therefore f(x) is always irreducible.

Note that no use was made of the fact, proved in prior message, that
r^2 + r + 1 would have to be a perfect cube.

For this problem, inequality considerations trumped divisibility.

quasi
From: quasi on
On Tue, 03 Aug 2010 02:16:14 -0500, quasi <quasi(a)null.set> wrote:

>On Tue, 03 Aug 2010 00:27:59 -0500, quasi <quasi(a)null.set> wrote:
>
>>On Sat, 31 Jul 2010 05:26:14 -0700 (PDT), Gerry <gerry(a)math.mq.edu.au>
>>wrote:
>>
>>>On Jul 31, 10:10 pm, Kent Holing <K...(a)statoil.com> wrote:
>>>> Can anybody prove or disprove that the cubic
>>>> r x^3 - (r^2 + r + 1)x^2 - (r^3 - 1)x + r^2(r^2 + r + 1) = 0 is irreducible for all integers r /= 0?
>>>> Kent Holing,
>>>> Norway
>>>
>>>No doubt you have observed that unless r^2 + r + 1 is squareful
>>>this polynomial is irreducible by Eisenstein.
>>
>>Based on divisibility considerations, I think I've shown more --
>>namely that if a nonzero integer r is such that the polynomial in
>>question is reducible, then r^2 + r + 1 must be a perfect cube.
>>
>>The argument goes as follows ...
>>
>>Assume r is a nonzero integer such that the polynomial is reducible.
>>
>>Let f(x) be the univariate polynomial in x for the given r. Since f is
>>cubic, the reducibility of f implies that f has a rational root. Since
>>the leading coefficient is r, the polynomial g(y) defined by
>>
>> g(y) = r^2 f(y/r)
>>
>>must have an integral root. Expanding g(y) yields
>>
>> g(y) = y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4)
>>
>>By slight abuse of notation, let y be the integral root. Thus
>>
>> y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4) = 0.
>>
>>Let p be any prime factor of r^2 + r + 1.
>>
>>It's clear that r^2 + r + 1 divides y^3, so p|^3, hence p|y.
>>
>>Suppose p^k || r^2 + r + 1 (which means k is the greatest positive
>>integer such that p^k | (r^2 + r + 1).
>>
>>Consider the powers of p which divide each of the 4 terms of g(y).
>>
>> p^a || y^3
>> p^b || (r^2 + r + 1) y^2
>> p^c || -(r^4 - r) y
>> p^k || (r^6 + r^5 + r^4)
>>
>>Then k < c and k < b, hence we must have k = a.
>>
>>But a is a multiple of 3, hence so is k.
>>
>>Therefore r^2 + r + 1 is a perfect cube, as claimed.
>
>Ok, I see how to prove it now.
>
>The polynomial f(x) with r integral and nonzero can't be reducible.
>
>Assume instead there is a nonzero integer r for which f(x) is
>reducible.
>
>Then as I argued in my prior message (quoted above), the polynomial
>
> g(y) = y^3 - (r^2 + r + 1) y^2 - (r^4 - r) y + (r^6 + r^5 + r^4)
>
>must have an integer root (where now y is an indeterminate).
>
>But by direct substitution,
>
> g(r^2) = r^3
> g(r^2 + 1) = -r^3
>
>So g has a real but non-integral root strictly between
>
> r^2 and r^2+1.
>
>Also
>
> g(-r^2-1) = -4r^4 - 3r^3 - 6r^2 - 2r - 2
> g(-r^2+1) = 4r^4 + r^3 - 2r^2
>
>But g(-r^2-1) is negative (even for all real r).
>
>and g(-r^2+1) is positive (since |r| >= 1).
>
>Also, g(-r^2) = -r^3 which is nonzero.
>
>It follows that g has a real but non-integral root strictly between
>
> -r^2 -1 and -r^2+1.
>
>Hence the 3rd root must be the integral root.
>
>Let the 3 roots be y1, y2, y3, where
>
> r^2 < y1 < r^2 + 1
> -r^2-1 < y2 < -r^2+1
>
>and y3 is the integral root.
>
>Adding the above inequalities yields
>
> -1 < y1 + y2 < 2
>
>But the sum of the roots is r^2 + r + 1. It follows that
>
> r^2 + r - 1 < y3 < r^2 + r + 2
>
> Since y3 is an integer, we must have
>
> y3 = r^2 + r
>
>or
>
> y3 = r^2 + r + 1
>
>By divisibility considerations, it would have to be the latter one,
>but forget divisibility -- we don't need it. Instead just substitute
>each of the above for y3 and solve for r. As expected, the resulting
>equation for r has no integral roots.

More precisely, no nonzero integral roots.

In fact,

g(r^2 + r) = -r^3

and

g(r^2 + r + 1) = r^3 + r^2 + r

>Therefore f(x) is always irreducible.
>
>Note that no use was made of the fact, proved in prior message, that
>r^2 + r + 1 would have to be a perfect cube.
>
>For this problem, inequality considerations trumped divisibility.

quasi
From: Rob Johnson on
In article <rbisrael.20100802214432$3c1d(a)news.acm.uiuc.edu>,
Robert Israel <israel(a)math.MyUniversitysInitials.ca> wrote:
>Gerry Myerson <gerry(a)maths.mq.edi.ai.i2u4email> writes:
>
>> In article <rbisrael.20100801195326$60e4(a)news.acm.uiuc.edu>,
>> Robert Israel <israel(a)math.MyUniversitysInitials.ca> wrote:
>>
>> > Gerry <gerry(a)math.mq.edu.au> writes:
>> >
>> > > On Jul 31, 10:10 pm, Kent Holing <K...(a)statoil.com> wrote:
>> > > > Can anybody prove or disprove that the cubic
>> > > > r x^3 - (r^2 + r + 1)x^2 - (r^3 - 1)x + r^2(r^2 + r + 1) = 0 is
>> > > > irreduc> > > ible for all integers r /= 0?
>> > > > Kent Holing,
>> > > > Norway
>> > >
>> > > No doubt you have observed that unless r^2 + r + 1 is squareful
>> > > this polynomial is irreducible by Eisenstein.
>> > > --
>> > > GM
>> >
>> > And it happens to be irreducible for r=18 and r=88916, which are the
>> > first two
>> > positive integers r for which r^2 + r + 1 is squareful. What's the next
>> > r?
>> > This sequence doesn't seem to be in the OEIS.
>>
>> Squareful numbers are of the form a^2 b^3.
>>
>> r^2 + r + 1 = a^2 b^3 is
>>
>> (2 r + 1)^2 - b^3 (2 a)^2 = -3,
>>
>> so it's Pellish - if it has one solution, it has infinitely many.
>> r = 18 leads to b = 7, so we're interested in solutions to
>> x^2 - 343 y^2 = -3
>> with x odd and y even. The next one (after x = 37, y = 2,
>> which corresponds to r = 17) has about 20 digits and can
>> be found using http://www.alpertron.com.ar/QUAD.HTM
>>
>> r = 88916 comes from b = 13 and gives another infinite family.
>>
>> None of this answers Robert's question, of course, as to what
>> the next integer is, although it does give a bound.
>>
>
>Trying all b <= 500, I get 228594845490 for b = 199, then 9494342272670 for
>b=19, and 690845140450082 and 1262403975253755261 for b=7. Kent's polynomial
>is irreducible for each of these r values.

I haven't been able to make this work any better than the previous
attempt, but if

P(x) = r x^3 - (r^2 + r + 1)x^2 - (r^3 - 1)x + r^2(r^2 + r + 1)

then

P(x-r+1) = r x^3 - (4r^2 - 2r + 1) x^2 + (r - 1)(4r^2 - 2r + 1) x

+ r(4r^2 - 2r + 1)

Thus, P(x-r+1) does not satisfy Eisenstein if 4r^2 - 2r + 1 is
squareful. We get the same type of Pell equation

(2c)^2 d^3 - (4r-1)^2 = 3 [1]

Perhaps it might be possible to show that the same r cannot work for
[1] and [2]

(2a)^2 b^3 - (2r+1)^2 = 3 [2]

at the same time.

Rob Johnson <rob(a)trash.whim.org>
take out the trash before replying
to view any ASCII art, display article in a monospaced font