I'll give you my two cents worth on the HDevice--a long two cents worth.  I
like the HDevice--it gives both very precise and accurate values *if you
are careful*.  In fact, I think it can give results that are as good as can
be obtained by any other method.  In a recent project, I analyzed a lab
standard [quality control], similar in composition to the samples we were
running, and the precision was  0.5%o [n=135] and the mean was within
0.1%o of the value we obtained for the standard when we calibrated it
directly against VSMOW and SLAP.  I also have a duplicate data set on 40-50
unknowns run by the HDevice and by offline preparation with Uranium and
analyzed in another lab.  The agreement was very good--for only one sample
was there a difference in reported values that was greater than 1%o.
However, the device has some quirks to be aware of.

To fill with Cr:  Invert the qtz reactor so the small aperture neck is
down.  Push some quartz wool through, into the bulb of the reactor, and
pack it down into the narrow end of the small aperture neck.  Add Cr
powder, constantly tapping the reactor to settle the Cr.  Get as much
powder into the reactor blub as you can.  The reactors are expensive and
you want to analyze as many samples as possible--the Cr is cheap relative
to the reactor.  Then pack 2-3 wads of quartz wool down into the wide
aperture neck to hold the Cr in place and to clean out loose powder.

Be sure to use two septa on the top and change them each time you change
reactors.  Depending on the grain size of the Cr, I get about 150
injections [1 microliter per injection].  Make sure the reaction time is
long enough [2 minutes?] and set the temperature from 800-900C.  Don't try
to push the number of samples you can run because if you exhaust the Cr you
will get water into the inlet.  If you take a partially exhausted reactor
out, you will see a distinct boundary between reacted and unreacted Cr
powder.  Apparently H2 is reduced just as soon as the expanded water vapor
reaches hot Cr.  When installing a new reactor, bring the temperature up
slowly in 100C increments and pump it thoroughly before use.

Some mention has been made of memory in the syringe.  If you have an
autosampler, program it to purge the needle [precontaminate it] with the
subsequent water several times before making the next injection.  This
should eliminate the memory in the needle.  However, this is not the only
memory effect.  I find there is about 1% memory in the Cr itself--no matter
how many times I purge the needle first.  For example, if I analyze VSMOW,
followed by SLAP, the first injection of SLAP will run about 4%o heavier
than subsequent injections.  If I want to run samples in duplicate, I
actually run them in triplicate and discard the first analysis--the first
injection simply overcomes memory [by the second injection the memory
effect is only 1% of 1%].  It is also possible to make a quantitative
memory correction if you wish.  Because of memory, analyze samples you
expect to be of similar composition in sequence.

There is one test you **must** do before analyzing unknowns.  Set up a
sequence and run the same water, say, 30-50 times.  You may see [I hope you
don't] the apparent composition become lighter with time--in fact, it may
begin to drift noticeably after 5-10 injections.  This happens because the
reference gas is fractionating [getting heavier] as protium preferentially
exits the reference inlet/bellow. I have seen this in more than one Delta
Plus, including an instrument at the factory.  Naturally, fractionation is
more likely with H2 than with larger molecules.  You have two choices if
this happens.  You will either have to refill the bellows very frequently
or make a drift correction.  This last option may make some folks
uncomfortable.  However, you will find that as the reference bellow
depletes and fractionates, it does so **very systematically** and can be
fit by a least squares quadratic equation.  If you see this behavior, you
will probably find that the correlation coefficient on the drift is almost
1--it is very predictable.  However, monitoring for drift eats up reactors
and decreases throughput.  Actually, you may have a third choice--crimp the
capillaries more.  However, if you are running other gases you will lose
sensitivity.  I think the best solution is to refill often if you
can--that's good laboratory practice--especially with H2.  However, I am
also convinced that systematic drift can be corrected without introducing
much error.

Steve Nelson
Dept. of Geology
S389 ESC
Brigham Young University
Provo. Utah  84602

voice:  1-801-378-8688
lab:    1-801-378-5124
FAX:    1-801-378-8143

CONTROL, ALT, DELETE:  the 3 most
frequently used keys on a PC.