390
DOC.
393
OCTOBER
1917
a
=
ß
=
u
=
1
yields
+
8rT
1
a2
ß
-
1,
a
=
ß
=
2 //
"^4r4
1
a2
ß
-
1,
a
=
ß
-
3 //
"^4
1
a2
r4
..
a
=
1,
ß
-
ß
=
2 yields
lo2
8
r4
o
=
l,
ß
=
ß
=
3
//
_8
1
a2
r4”
ß
=
1,
OL
-
ß
-
2
//
_
8
1
a2
r4"
ß
-
1,
a
=
ß
=
3
//
“ër4
1
a2
//
lo2
=3.
II
T-1
ö
II
'S*.
II
8
r4
a
=
1,
u
=
ß
=
4
and
B
=
1,
u
=
a
=
4
yield
zero.
Kt44
is
made
up
of these
components
and
the
resulting
sum
is
zero.
Thus
I
find
exactly
the
same
thing I
had
reported
to
you
earlier,
and the
inconsistency
be-
tween
our
computations
remains.
I
have calculated
out
the
matter in
a
third
manner as well,
namely,
with formula
(52)
of
your
Annalen
paper.[5]
Owing
to
the
differentiation
on
the
left,
more
precise expressions
than
(70)
must
be used
for
the
guv's.
The calculation
is
very simple,
and
one
obtains
t11
=
t22
=
t33
=
t44
=
0.
With the best
of
intentions,
I
cannot
get anything
else,
and I do not
see
where
the
inconsistency
lies either.[6]
Because
I
have
so
much
space
left
on
the
page,
I
would also
like
to
say a
few
words
about
a
topic
I
discussed with Fokker.
But
it
is
purely
a
matter of
taste,
namely,
whether
it
is
generally
useful
to
regard
the
guv's
as
pure
numbers (V-g
established
as
not
equal
to
1).
Is it
not
more
practicable
to have different units for
distances and coordinate
lengths
in
natural
measure?
If
these
units
are
denoted
as
a
and
E,
then
guv,
would
be
of
the
dimension, a2E-2.
The centimeter
would
probably
be
the
unit of
natural
distances;
the
coordinate
lengths
could
even
be
considered
pure
numbers.