From: mueckenh on

Tim Peters schrieb:

> [mueckenh(a)rz.fh-augsburg.de]
> >>> An uncountable countable set
> >>>
> >>> There is no bijective mapping f : |N --> M,
> >>> where M contains the set of all finite subsets of |N
> >>> and, in addition, the set K = {k e |N : k /e f(k)} of all natural
> >>> numbers k which are mapped on subsets not containing k.
>
> [Dik T. Winter]
> >> Ah, I see. The set M is defined depending on f. Well in that case f
> >> is clearly not a bijection between N and M.
>
> I'm not sure I'd be so generous here. K isn't a definite set before f is
> specified (meaning that for K to provably exist by the axiom of separation,
> f has to provably exist first), but to specify f K has to be a definite set
> first (since K must be the image under f of some natural k, f can't be
> proven to exist unless K can be proven to exist first).

Of course f does not exist. But the reason has nothing at all to do
with cardinalities. This error, however, is the foundation of
Hessenberg's proof of Cantor's theorem.

A set is required which is the image of k if it is not the image of k.
A barber is required who shaves himself if he does not.
>
> IOW, I doubt this argument can be made in ZFC, just as it's not possible
> under ZFC to prove that
>
> F = {k e |N : k e F}
> or
> G = {k e |N : k /e G}
>
> exist.

On the other hand, why should any mapping not exist? If you claim that
a surjective mapping |N --> P(|N) must contain the set K in any case,
then there is no arguing why my mapping should not exist.

Regards, WM

From: mueckenh on

Virgil schrieb:

> In article <1150782120.872330.199120(a)c74g2000cwc.googlegroups.com>,
> mueckenh(a)rz.fh-augsburg.de wrote:
>
> > Daryl McCullough schrieb:
> >
> > > mueckenh(a)rz.fh-augsburg.de says...
> > >
> > > >> >> > There is no bijective mapping f : |N --> M,
> > > >> >> > where M contains the set of all finite subsets of |N
> > > >> >> > and, in addition, the set K = {k e |N : k /e f(k)} of all natural
> > > >> >> > numbers k which are mapped on subsets not containing k.
> > >
> > > >Above I spelled out clearly that M contains the set of all finite
> > > >subsets of |N and one more set K. What is unclear?
> > >
> > > Okay, so M is not a fixed set, but is a function of f. So to make
> > > this precise, let's put in the functional dependence:
> > >
> > > K(f) = { k in N | k is not an element of f(k) }
> > > M(f) = { A in P(N) | A is finite, or A = K(f) }
> > >
> > > where P(N) means the set of all subsets of N.
> > >
> > > Then what you are saying is that
> > >
> > > forall f: N -> P(N),
> > > f is not a bijection between N and M(f)
> > >
> > > That's true. But that doesn't mean that M(f) is uncountable.
> >
> > If it is proven impossible to set up a mapping between M and |N, then M
> > is uncountable.
>
> But you have not proven it impossible, you merely argue that ONE
> function between N and M(f) is not a bijection.
>
>
>
> WE have some f: N --> P(N), which we know is not a surjection since
> there is no n in N for which one can have f(n) = K(f).
>
> But that does no mean that there is no surjection between M(f) and N.
>
> In fact, given a function f for which K(f) exists, it is not difficult
> to build a bijection between M(f) and N as follows:
>
> For each S in M(F) define g: M(f) --> N by:
> If K is finite then
> if 0 e N
> then g(S) = sum_{s e S} 2^s
> else g(S) = sum_{s e S} 2^(s-1)
> endif
> else (K being infinite)
> g(K) = the first natural
> and
> if 0 e N
> then g(S) = sum_{s e S} 2^s+1
> else g(S) = sum_{s e S} 2^(s-1)+1
> endif
> endif
>
> In all these cases, g will be a bijection from M(f) to N.
>
>
>
> >
> > > We can prove
> > >
> > > forall f: N -> P(N), exists g: N -> P(N)
> > > g is a bijection between N and M(f)
> >
> > No, it is not, because M(f) is incomplete without K. But K does not
> > exist.
>
> But you have not proven it never exists, and for some functions, f, it
> Does exist.
>
> For example, let f(n) = {} for all n, then
> K = {k in N : not k e f(k) } = N.
>
> So that for SOME functions f, f, K(f) exists.
>
> The reason that Meucken's argument fails here is that there is no
> necessity for the function f to be a surjection onto M(f).
>
> The desired bijection can be established by an entirely different
> function between M(f) and N, as shown above.

Only if K(f) would exist.
>
>
>
>
>
>
>
> > The reason for its non-existence is not lacking number of
> > elements of |N but an impredicable definition.
>
> Wrong, as shown by a specific example above.

Your example ist wrong because K(f) does not exist.

Regards WM

From: Dik T. Winter on
In article <1150728187.749465.36710(a)g10g2000cwb.googlegroups.com> mueckenh(a)rz.fh-augsburg.de writes:
> Dik T. Winter schrieb:
> > In article <1150718841.726873.223390(a)g10g2000cwb.googlegroups.com> mueckenh(a)rz.fh-augsburg.de writes:
....
> > > > > An uncountable countable set
> > > > >
> > > > > There is no bijective mapping f : |N --> M,
> > > > > where M contains the set of all finite subsets of |N
> > > > > and, in addition, the set K = {k e |N : k /e f(k)} of all natural
> > > > > numbers k which are mapped on subsets not containing k.
> > > >
> > > > But is the set K in M? Pray give a proof.
> > >
> > > Of course it is, by definition, for without K M would not be M.
> >
> > Ah, I see. The set M is defined depending on f. Well in that case f
> > is clearly not a bijection between N and M.
>
> That is correct.
>
> > This does not tell us that
> > there is *no* bijection (say g) between N and M.
>
> M depends on f. And when you take g, then M depends on g. That is the
> essence of M.

In that case M is not a set. And in order to tell whether that is
uncountable or not you have to first define cardinality on such
non-sets.

> > So also M(f) is countable.
>
> That is not at all the way a bijection has to be constructed between |N
> and M. That is simply impossible!

A bijection can be constructed, you only can not name if f.

> But if you like tricks of this kind, then you can biject |N and its
> power set P(|N) \ K. Subsequently take
> 1. g(0) = K(f)
> 2. g(n) = f(n - 1) when n > 0
> and we see that Cantor's theorem is not proven by Hessenberg's proof,
> because this proof makes use of the impredicable definition of a set
> which does not and cannot exist, as well as the barber who shaves all
> men of his village.

Sorry, this is plain nonsense. You can not construct a bijection between
P(N) \ K. The difference between this and your construction with M(f)
is that the target M(f) depends on the bijection you are constructing.
P(N) is a properly defined set that does not depend on anything but N.
But you are wrong. For every mapping f, the set K does exist (and is
an element of P(N), which does not depend on f), but it is not in the
image of f, which it should be if f is a bijection.

> Of course, one may define a mapping from |N on P(|N), for instance n
> --> {n}, where K is well defined. But the set {f, k, K}, essential for
> Hessenberg's proof, does not exist. This shows that the customary
> proof of Cantor's theorem makes use of a set which does not exist,
> namely {f, k, K}. Therefore, its non-existence does not prove anything
> about surjectivity of mappings.

The triple {f, k, K} does indeed not exist, but if f is a bijection it
should exist. It is just this contradiction that shows that f is not
a bijection.
--
dik t. winter, cwi, kruislaan 413, 1098 sj amsterdam, nederland, +31205924131
home: bovenover 215, 1025 jn amsterdam, nederland; http://www.cwi.nl/~dik/
From: Dik T. Winter on
In article <1150796061.943937.27660(a)p79g2000cwp.googlegroups.com> mueckenh(a)rz.fh-augsburg.de writes:
....
> > [Dik T. Winter]
> > >> Ah, I see. The set M is defined depending on f. Well in that case f
> > >> is clearly not a bijection between N and M.
> >
> > I'm not sure I'd be so generous here. K isn't a definite set before f is
> > specified (meaning that for K to provably exist by the axiom of separation,
> > f has to provably exist first), but to specify f K has to be a definite set
> > first (since K must be the image under f of some natural k, f can't be
> > proven to exist unless K can be proven to exist first).
>
> Of course f does not exist. But the reason has nothing at all to do
> with cardinalities. This error, however, is the foundation of
> Hessenberg's proof of Cantor's theorem.

Oh, but I think I disagree with Tim. Given any f: N -> S where S is the
set of finite subsets of N, K(f) and M(f) do exist.

> A set is required which is the image of k if it is not the image of k.
> A barber is required who shaves himself if he does not.

But here is your confusion.
1. Given a mapping f: N -> P(N), the set K(f) constructed according to
Hessenberg *does* exist. If f were a bijection it is required (by
the *definition* of bijection) that K(f) (because it is an element
of P(N)) is the image of some element of N, but it is not, showing
that f is not a bijection. P(N) does not depend on f, so there is
no mapping between N and P(N) that is a bijection.
2. Given a mapping f: N -> S, the sets K(f) and M(f) do exist. If f
were a bijection between N and M(f), K(f) should be in the image,
it is not, so f is not a bijection. From this you can *not*
conclude that M(f) is uncountable, because there can be a bijection
g: N -> M(f). You can at most conclude that the union of *all* M(f)'s
is uncountable. And that is easy, that union is just P(N).

> On the other hand, why should any mapping not exist? If you claim that
> a surjective mapping |N --> P(|N) must contain the set K in any case,
> then there is no arguing why my mapping should not exist.

You mapping does exist. See (2) above.
--
dik t. winter, cwi, kruislaan 413, 1098 sj amsterdam, nederland, +31205924131
home: bovenover 215, 1025 jn amsterdam, nederland; http://www.cwi.nl/~dik/
From: Daryl McCullough on
mueckenh(a)rz.fh-augsburg.de says...

>Daryl McCullough schrieb:

>> Okay, so M is not a fixed set, but is a function of f. So to make
>> this precise, let's put in the functional dependence:
>>
>> K(f) = { k in N | k is not an element of f(k) }
>> M(f) = { A in P(N) | A is finite, or A = K(f) }
>>
>> where P(N) means the set of all subsets of N.
>>
>> Then what you are saying is that
>>
>> forall f: N -> P(N),
>> f is not a bijection between N and M(f)
>>
>> That's true. But that doesn't mean that M(f) is uncountable.
>
>If it is proven impossible to set up a mapping between M and |N, then M
>is uncountable.

Once again, M is not a fixed set. It is a *function* of f.
For each function f, there is a corresponding K_f and a
corresponding M_f.

Let's look at an example: Let f(x) be a mapping from N to P(N)
as follows:

f(0) = {}
f(1) = { 0 }
f(2^a_1 3^a_2 5^a_3 ... p_n^a_n)
= { x_1, x_2, ..., x_n }
where x_1 = a_1,
x_{n+1} = x_n + a_{n+1}

As I already pointed out, K(f) = { 0, 1, 2, ... } = N.
M(f) = { A | A is a finite subset of N, or A = K(f) }

M(f) is countable, since there is a bijection g:

g(0) = N
g(x+1) = f(x)

>That is the definition of uncountability.

By the definition of "uncountable" M(f) is countable.

>Cp. Cantor's diagonal proof:
>There it is even possible to set up a bijektion between {numbers of the
>list & diagonal number}, (after the diagonal number has been
>constructed) and |N. Nevertheless uncountablilty is claimed.

The difference is that there is a proof that P(N) is uncountable,
but there is a proof that M(f) *is* countable.

What we can show is that for every f,

f is not a surjection from N to M(f).

But since M(f) is a subset of P(N), we have

f is not a surjection from N to M(f)
implies
f is not a surjection from N to P(N)

So we conclude (Cantor's theorem)

forall f, f is not a surjection from N to P(N).

>> We can prove
>>
>> forall f: N -> P(N), exists g: N -> P(N)
>> g is a bijection between N and M(f)
>
>No, it is not, because M(f) is incomplete without K. But K does not
>exist.

On the contrary, for every f, there is a corresponding K(f), and
there is a corresponding M(f).

You give me an f, and I will show what the corresponding K(f) is.
We already saw one example: for the f defined explicitly above,
K(f) turns out to be all of N.

>The reason for its non-existence is not lacking number of
>elements of |N but an impredicable definition.

Since K(f) *does* exist, it's a little silly to talk about the
reasons for its nonexistence.

Perhaps this would be easier for you if you start out
considering finite sets, instead of N.

Let N_1 = { 0 }. Then P(N_1) = { {}, {0} }.
Let f be any function from N_1 to P(N_1). There
are two such functions: f_1 and f_2 where

f_1(0) = {}
f_2(0) = {0}

Now consider K(f_1).

K(f_1) = { x in N_1 | x is not an element of f_1(x) }
= { 0 }

Clearly, K(f_1) is not in the image of f_1. So f_1 is not
a bijection between N_1 and P(N_1).

Now consider K(f_2).

K(f_2) = { x in N_1 | x is not an element of f_2(x) }
= {}

Clearly, K(f_2) is not in the image of f_2. So f_2 is not
a bijection between N_1 and P(N_1).

Note, K(f_1) is *not* equal to K(f_2).

Since f_1 and f_2 are all the functions from N_1 to P(N_1),
it is clear that there are no bijections from N_1 to P(N_1).

We can try the same thing using N_2 = { 0, 1}. We will find
that there is no bijection from N_2 to P(N_2). The same is
true for N_3 = { 0, 1, 2 } and N_4, and N_5, etc.

We can generalize: for *any* set A (finite or not), and for
any function f from A to P(A), we can define

K(f) = { x in A | x is not an element of f(x) }

It is clearly the case that K(f) is not in the image of f.
It is also clearly the case that K(f) is an element of P(A).
So we conclude:

f is not a bijection between A and P(A).

--
Daryl McCullough
Ithaca, NY

First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Prev: integral problem
Next: Prime numbers