Re: Accuracy vs. Relevance

Thanks for your stimulating reply.

Jonathan Thornburg ha scritto:
> I'm sorry, but this proposal simply doesn't fit with how science
> -- particularly "big science" -- is done by actual living-and-breathing
> human scientists. The GP-B project got started in 1961 (!), and
> many of the current team members have spent (plural) *decades* of
> work on it, i.e. much or all of their professional careers.

They have been working hard and having good, meaningful lives, yes.
I have genuine respect and some envy for all the people involved, but
this should not prevent me or anyone to exert constructive criticism.

>
> What do you think motivates talented people to put in that work?
> The answer, in significant part, is the promise of being (co)author
> on the *first* papers presenting the results. Putting the raw data
> out before those papers are published would expose them to the risk
> (near certainty) of being scooped by others, and indeed quite likely
> result in a Gresham's-law ("the bad drives out the good") where the
> quickest and sloppiest analyses would appear first.


They are testing a theory . If the theory truly has predictive power
(which btw I believe GR has although not to the extent commonly
claimed), it should be demonstrable in real time. The point, which I
will elaborate upon later, is that there is a significant difference
between genuine prediction and data-fitting. If they are truly testing
the theory's predictions, they should be able to churn out their raw
data, check their fit immediately and be able to claim:
"see, our model works, it all fits, we predicted it right" or "oh, our
predictive model has a problem". I know it's tough, but true
predictions are about using your model to forecast the future, not
about fitting together your model and data that you already have been
staring at for months. More below.


> Quite rightly,
> the GP-B team, their funding agencies (= mostly NSF, NASA, and DOE),
> and the scientific community as a whole, reject that.

I would use the word "understandably" rather than "rightly".

>
> Instead, the way space science -- and particle physics, and
> observational astronomy, and most other "big science" -- generally
> works, is that the original team (in this case the GP-B team) gets
> exclusive access to the data for some period, *then* the data goes
> into a public archive. A good example of this is the Hubble Space
> Telescope. Observations are generally private to the observers who
> requested them for a period of 1 year; after that the data goes into
> the HST public archive and may be accessed by anyone. For special
> cases (eg data which has to be collected over decade-long periods
> before it can be analyzed), observers can request a longer "private
> period".

Interesting! Is there a website explaining how and when an interested
third party can access the HST archive, as well as the guidelines for
the formatting and completeness of the data? Can I follow the GPB
team's work step by step based on what is stored there?

>
> There's another key point here, what do we mean by the "data"?

Indeed, this is the key issue. It's what makes the test of
predictability crucial.
See below.

> HST provides a good illustration: there are a number of rather
> sophisticated data-processing steps between the raw data ("pixel
> 123,456 of CCD#5 had 12345 counts in this exposure") and something
> of scientific interest ("star HD123-456789 had V magnitude 19.28
> +/- 0.02 on JD 2456789.123"). A major job for the HST instrument
> teams and the Space Telescope Science Institute is to maintain the
> "pipeline" of calibration data and software which does this processing.
> Doing this requires detailed knowledge about the engineering design
> and construction of the individual HST instruments.

Basically you are saying that there are dozens of adjustable parametrs
that need insider knowlege to be interpreted and may be calibrated so
as to obtain something meaningful (i.e. fitting the observer's
expectations) from the magmatic raw data. That's exactly what makes me
shake my head at "a posteriori" claims of fabulous predictive power.

There is another, distinct and subtler issue, which I refer to as the
semantic problem, i.e. the problem of mapping mathematical models into
measurement outcomes and viceversa, which might be relevant to discuss
here (cf. [1] ) .

Let's start with Wigner, who, contrasting the situation in QM and SR
with that in GR writes ([2]) "... the measurement of position, that is,
of the space coordinates, is certainly not a significant measurement
if the postulates of of the general theory are adopted: the
coordinates can be given any value one wants. ... Most of us have
struggled with the problem of how, under these premises, the general
theory of GR can make meaningful statements and predictions at all. ...
This is a point that which cannot be emphasised strongly enough and it
is the basis of a much deeper dilemma ... . It pervades the general
theory, and to some degrees we mislead both our students and ourselves
when we calculate , for instance, the mercury perihelon without
explaining how our coordinate system is fixed in space, what defines it
in such and such a way that it cannot be rotated, by a few seconds a
year, to follow the perihelion apparent motion. ... . There must be
some assumption on the nature of the coordinate system that keeps it
from following the perihelion. ... . A difference in the tacit
assumptions which fix the coordinate system is increasingly recognized
to be at the bottom of the many conflicting results arrived at in
calculations based on the general theory of relativity."

Wigner is talking about the problem of diffeomorphism invariance in GR.
Now, while for well studied cases, such as a.o. the mercury perihelion
and structurally similar situations, physicists know how to choose the
coordinates so as to obtain results that fit observations, I am not
sure that this holds in general. I am not aware of any conclusive
treatment of this issue. So, while I believe that GR has enough juice
and its practitioners are smart enough to solve the problem "a
posteriori" on a given observed system once they have enough data, I am
not sure that the same holds a "priori" in situations which have not
been examined in the light of the observed data. As long as the this
problem is not solved in a clear, general way , I wonder how GR can be
regarded as a complete predictive theory.

>
> So, if you really want to see the raw GP-B data, you can
> (a) join the GP-B team,

I would love to. I would even keep my big mouth shut for the privilege.
Alas, I can play only the cards I have got. When it was time in
Bologna, I should have perhaps honed my engineering skills instead of
wasting time at Umberto Eco's semiology lectures.

:D

> (b) wait a year or two.

I will, but by then this won't be about predictions, right?

Mach's gut,

IV

[1]
http://groups.google.com/group/sci.physics.research/browse_frm/thread/604bcbc119865d44/
[2] E. Wigner "Relativistic Invariance and Quantum Phenomena" Rev. Mod.
Phys. 29, 255 (1957)

---------------------------

"It is hard to make accurate forecasts, especially about the future."

Danish proverb

.

Relevant Pages

  • Re: Q? Gravity Probe B
    ... "15.2 The Effects of Relativity Are Clearly Visible in the Raw ... The GP-B satellite was designed as a pristine, space-borne laboratory, ... "The graph in Figure 15-3 above shows that the geodetic effect is clearly ... confirming Einstein's predictions to a precision of better than 1 percent. ...
    (sci.physics)
  • Re: The real twin paradox.
    ... the theory of relativity is wrong. ... predictions that are proved false by experiment. ... the inertial coordinate systems. ... As measured in any inertial coordinate system, ...
    (sci.physics.relativity)
  • Accuracy vs. Relevance
    ... >> theory of GR can make meaningful statements and predictions at all. ... >> coordinates so as to obtain results that fit observations, ... "events") are diffeomorphism invariant is false. ... If you set up a space-time measurement model, ...
    (sci.physics.research)
  • Re: true random number generator
    ... In fact, Bohm's formulation of>> QM makes all the same predictions, is a perfect fit the experiemtnal data,>> and does NOT claim that any such thing as "true" randomness exists. ... >> without any experimental support inasmuch as every experiment that is used>> to bolster that interpretation also supports Bohm's deterministic ... > The following hypothesis is presented for discussion:> A "true" random process is one in which future states cannot be> predicted by any physical or computational means beyond purely> statistical expectations. ...
    (sci.crypt)
  • Re: true random number generator
    ... > QM makes all the same predictions, is a perfect fit the experiemtnal data, ... Pseudorandom number generators fit this definition ... A "true" random process is one in which future states cannot be ... and sufficient computational power. ...
    (sci.crypt)