TITLE: K-theory of non-compact spaces
QUESTION [17 upvotes]: This is a question on nomenclature of $K$-theory in the topological category.
The $K$-theory of a compact space $X$ is defined as the Grothendieck group of the vectorbundles on $X$. The Atiyah-Jänich Theorem states that this is the same thing as the homotopy classes of maps $X\rightarrow \Phi(\mathbb{H})$, where $\Phi(\mathbb{H})$ is the space of Fredholm operators on some separable infinite-dimensional Hilbert space.
Now, for a non-compact space $Y$ the $K$ theory $K(Y)$ is not defined as the Grothendieck group of vector bundles on $Y$. One can do a couple of things:

Define it as the $K$-theory of its one point compactification
$K(Y):=K(Y_+)$. One needs to assume $Y$ is locally compact for this
to make sense.
Another option is to assume that $Y$ is nice, for example an
infinite $CW$ complex such as $CP^\infty$, and to define the $K$-theory of $Y$ as the limit of its finite subcomplexes. Note that $CP^\infty$ is not locally compact.
Yet another option is to define the $K$-theory of $Y$ as maps
$Y\rightarrow \Phi(\mathbb{H})$. I believe this is called representable
$K$-theory.

I have a couple of questions.

Why is 1. a good definition? I like my cohomology theories to be
      functorial under maps. Theory 1. clearly is not. I believe it is functorial
      with respect to proper maps. This reminds me of compactly supported
      cohomology. But why is this then not called compactly supported $K$-theory? 

And:

How are 2. and 3. related? 

More specifically:

What exactly does 3. describe? Are these virtual vector bundles that admit numerable trivializations?

finally:

Is there a reference where all these definitions are discussed?

REPLY [4 votes]: What exactly does 3. describe? Are these virtual vector bundles that admit numerable trivializations?

Virtual vector bundles, when defined as formal differences (i.e.,
elements in the homotopy group completion) of vector bundles,
do not form a sheaf when considered on noncompact manifolds.
The reason for this is that a compatible system of such differences
considered on some open cover might use differences
of higher and higher dimensional vector bundles as one approaches
the infinity, whereas any virtual vector bundle is a difference
of two bundles of fixed dimension.
Put differently, the homotopy group completion functor, being
a left adjoint, does not preserve the sheaf condition,
which is defined using a homotopy limit.
However, the problem can be easily rectified by replacing
the above presheaf with its sheafification.
With such an extended notion of virtual vector bundles
one then proves that their isomorphism classes are represented
by the connective K-theory spectrum.
(The above is literally true for smooth manifolds with the standard
Grothendieck topology; for topological spaces one must
take the numerable Grothendieck topology, which produces numerable bundles.)