The fact that two operators that commute have a common set of
eigenfunctions can be seen as follows: assume that
is an
eigenfunction of
with eigenvalue
. Then since
and
commute,
. Comparing start and end,
must be an eigenfunction of
with eigenvalue
just
like
itself is. If there is no degeneracy of the eigenvalue,
that must mean that
equals
or is at least
proportional to it. That is the same as saying that
is an
eigenfunction of
too. (In the special case that
is
zero,
is still an eigenfunction of
, with eigenvalue
zero.)
If there is degeneracy, the eigenfunctions of
are not unique and
you can mess with them until they all do become eigenfunctions of
too. That can be shown assuming that the problem has been
approximated by a finite-dimensional one. Then
and
become
matrices and the eigenfunction become eigenvectors. Consider each
eigenvalue of
in turn. There will be more than one eigenvector
corresponding to a degenerate eigenvalue
. Now by completeness,
any eigenvector
can be written as a combination of the
eigenvectors of
, and more particularly as
where
is a combination of the eigenvectors of
with
eigenvalue
and
a combination of the eigenvectors of
with other eigenvalues.
The vectors
and
separately are still eigenvectors
of
if nonzero, since as noted above,
converts eigenvectors of
into eigenvectors with the same eigenvalue or zero. (For example,
if
was not
,
would have to make up the
difference, and
can only produce combinations of
eigenvectors of
that do not have eigenvalue
.) Now
replace the eigenvector
by either
or
,
whichever is independent of the other eigenvectors of
. Doing this
for all eigenvectors of
you achieve that the replacement
eigenvectors of
are either combinations of the eigenvectors of
with eigenvalue
or of the other eigenvectors of
. The set of
new eigenvectors of
that are combinations of the eigenvectors of
with eigenvalue
can now be taken as the replacement
eigenvectors of
with eigenvalue
. They are also eigenvectors
of
. Repeat for all eigenvalues of
.
Similar arguments can be used recursively to show that more generally, a set of operators that all commute have a common set of eigenvectors.
The operators do not really have to be Hermitian, just “diagonalizable”: they must have a complete set of eigenfunctions.
The above derivation assumed that the problem was finite dimensional, or discretized some way into a finite-dimensional one like you do in numerical solutions. The latter is open to some suspicion, because even the most accurate numerical approximation is never truly exact. Unfortunately, in the infinite dimensional case the derivation gets much trickier. However, as the hydrogen atom and harmonic oscillator eigenfunction examples indicate, typical infinite systems in nature do satisfy the relationship anyway.