Sequence satisfying an |
n
By Yatir Halevi on Monday, September 30,
2002 - 09:39 am:
Prove that for all
,
.
Thanks,
Yatir
By Michael Doré on Monday, September 30, 2002 - 06:50 pm:
The sum means
where the sum on the right
hand side is over all divisors of
(apart from
itself). So for instance:
Anyway, I think the following works. We have:
where the sum is over all positive
which divide into
. The
on the
right hand side is the number of subsets of
and so this
suggests that we're counting subsets of
in some way.
But how do we incorporate the
condition? Well we can write the numbers
1, 2, ...,
on a circle, and we can represent a subset by colouring red
all the integers in the subset (and black otherwise). Now it may be that if you
rotate the circle a certain amount then the colouring still looks the same - i.e.
the sequence of reds and blacks is symmetric under a rotation of angle
units.
Note that we would have to have
- so this gives us a way of incorporating
this condition.
So let's claim that
is equal to
defined as the number of subsets of
such that if you represent the subset by colouring red numbers
on a circle then the sequence of reds and blacks isn't symmetric under any
non-trivial rotation.
To do this it suffices to show that:
where the sum is over all natural
because then
follows by
induction. I'll leave this to you - write back if you need help.
So anyway we've shown that
is the number of subsets on
which aren't symmetric under any rotation when we write them on a circle. Let
be the set of these subsets. We have
so it sufficies to show that
divides into
. Well this is quite easy. Define two members of
to
be equivalent if you can switch between their representations on the circle by
rotation. It is clear that each equivalence class has exactly
elements, and
is a disjoint union of all the equivalence classes, so
is a multiple
of
, so we're done.
By Yatir Halevi on Thursday, October 03,
2002 - 01:14 pm:
(by definition)
par
Though it is hardly a combinatorical proof...
Yatir
By Michael Doré on Thursday, October 03, 2002 - 02:04 pm:
Don't understand.
is defined as the sequence
satisfying your recurrence relationship. Why is
by definition?
is defined as the number of subsets of
such that if
you represent this subset by colouring in numbers on a circle, the sequence of
colours changes under any non-trivial rotation.
All of this rotation stuff is hard to explain. So let's forget about rotations
and circles and stuff, and I'll try and explain more clearly.
can be
defined as the number of subsets
of
with the property
that for any
in
there exists
,
in
with
(mod
) and such that exactly one of
,
is in
.
So we must show that
because then we're done as clearly
. The
way to show that
is to show that
satisfies exactly the same
recurrence relationship as
. (Because it is obvious that the recurrence
relationship admits a unique solution, so it follows that
.) So you
need to show:
where
is as defined in my second paragraph. This is, of course, a
combinatorial task.
A hint: if
is a subset of
let
be the minimal
integer
in
such that for all
,
in
with
(mod
) we have
in
in
. Note that
always exists because the property is satisfied with
(although this
will usually not be the minimal allowed integer with this property.
If
is in
define
as the set of subsets of
of
such that
.
Clearly the set of subsets of
is the disjoint union of
,
...,
. So:
Now prove that
if
doesn't divide into
. If
show that
. Substitute this in, and you have the required recurrence,
which completes the proof.