Accuracy vs. Relevance
I.Vecchi
> I anticipate your objections about how wonderful are predictions of
> GR regarding light bending or binary pulsars.
Wrong guess.
While I believe that GR is a beautiful theory with some genuine
predictive power, talk of fabulously precise predictions makes me
rather think of poll results in Turkmenistan. Data filtering may often
have more to do with career planning than with actually testing
scientific theories.
Just try to get raw data for some important experiment, like Gravity
Probe B.
According to my understanding of the scientific paradigm, the
data-processing protocol should be made available beforehand and the
raw data should ideally be posted on the net in real time, so as to
maximise falsicability and openness to public scrutiny (as well as
provide a field day to critical outsiders). Since this just does not
happen (not even close), I'll balance my admiration with some
skepticism, the basic question being "Why should I necessarily believe
something that I cannot check by myself?".
Actually the GPB team pitches GR as a theory with limited experimental
support, writing that " some of [Einstein's] most basic ideas remain
untested" and that while some experiments show "that Einstein was on
the right track ... more profound phenomena ... remain untested".
Moreover "deep theoretical problems -- some old and some new -- remain"
where "one obstacle to creative amendment ... is the paucity of
experimental evidence"([2]). No mention is made of GPS data (which
indeed are not open to public scrutiny) and it is suggested that the
famous Taylor-Hulse binary pulsar measurements conflict "with other
astrophysical data from eclipsing binary stars" ([2], cf.[3]).
> However, successful
> predictions of QM and QFT are much broader in scope and few orders
> of magnitude more accurate.
Respectively, one may add.
> So, I would place my bet on QM rather than
> on GR.
I would too, but the accuracy of its bastard offspring has little to do
with that.
Ugly QFT requires renormalisation, which provides ample opportunity
for parameter adjusting. Again, after some formidable
consensus-building, the resulting theories may have some
extrapolative/predictive power, but, unlike GR, I would not regard them
as more than effective lab phenomenologies, providing answers to a
limited set of questions defined "a priori", not unlike
renormalisation-based hydrodynamical turbulence models.
Waving my foot further, I suspect that Wilson's ultimate "microscopic
theory" , spawning different phenomenologies depending on the
observer's semiotic models (see [4]) , i.e. on the set questions she
is going to ask, may recede indefinitely as the set of questions is
enlarged.
..
Cheers,
IV
PS It's not like I doubt everything. I am positive that U-235 gets
nasty if you slam enough of it together. It's the kind of
reproducibilty that convinces everyone, from Hiroshima schoolchildren
to Islamic fundamentalists, regardless whether the blow is 19.9999 or
20.0001 kilotons.
[1]
http://groups.google.com/group/sci.physics.research/msg/32df41faf4b789f6
[2] http://einstein.stanford.edu/content/story_of_gpb/gpbsty2.html
[3] There's a nice discussion on this at sci.physics under the heading
"Long-Period Eclipsing Binary Stars and GR" at
http://groups.google.com/group/sci.physics/browse_frm/thread/e0f764ce9fcd=
5838
[4] Dosch, M=FCller and Sieroka "Quantum Field Theory, its Concepts
Viewed from a Semiotic Perspective" at
http://philsci-archive.pitt.edu/archive/00001624 .
Jonathan Thornburg -- remove -animal to reply
> Just try to get raw data for some important experiment, like Gravity
> Probe B.
> According to my understanding of the scientific paradigm, the
> data-processing protocol should be made available beforehand and the
> raw data should ideally be posted on the net in real time, so as to
> maximise falsicability and openness to public scrutiny (as well as
> provide a field day to critical outsiders). Since this just does not
> happen (not even close), I'll balance my admiration with some
> skepticism, the basic question being "Why should I necessarily believe
> something that I cannot check by myself?".
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.
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. Quite rightly,
the GP-B team, their funding agencies (= mostly NSF, NASA, and DOE),
and the scientific community as a whole, reject that.
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".
There's another key point here, what do we mean by the "data"?
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.
So, if you really want to see the raw GP-B data, you can
(a) join the GP-B team, or
(b) wait a year or two.
ciao,
--
-- "Jonathan Thornburg -- remove -animal to reply" <jth...@aei.mpg-zebra.de>
Max-Planck-Institut fuer Gravitationsphysik (Albert-Einstein-Institut),
Golm, Germany, "Old Europe" http://www.aei.mpg.de/~jthorn/home.html
"Washing one's hands of the conflict between the powerful and the
powerless means to side with the powerful, not to be neutral."
-- quote by Freire / poster by Oxfam
I.Vecchi
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
carlip...@physics.ucdavis.edu
[...]
> 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 think you are seriously underestimating the amount of data reduction
required. Look at the article in Matters of Gravity 26 (available from
http://www.phys.lsu.edu/mog/) to see what is involved.
Note that the data analysis will be "blind," in the sense that all of
the data reduction will be done *without* knowing the actual position
of the GPB guide star. This will be added in only at the last step,
precluding any chance of (deliberate or accidental) fudging.
[...]
This is certainly a major issue in quantum gravity. But in classical GR,
it is not, or at least need not be. The basic point to remember is that
*actual observations* are diffeomorphism-invariant. We do not observe,
for instance, the "coordinate value of the position of Mercury"; we observe
things like "the round trip time of a radar pulse from a fixed location
on Earth to Mercury and back, as measured by an atomic clock at that
location," or "the angle between the light arriving from Mercury and
that coming from a reference star, as measured at a particular telescope
at a time determined by a clock at the location of that telescope." Such
quantities do not depend on any choice of coordinates. To compare GR
to observation, what you do is to compute (in, say, the post-Newtonian
approximation) the predictions for such *observables*, and compare them
to the the actual observations. Better, rather than just comparing GR,
you look at a more general model (the *parametrized* post-Newtonian
approximation, for instance), and find the best fit for your free
parameters; you can then compare the result to GR, and at the same time
get a good estimate for how good the fit is.
For a simple example, take a look at Boddener and Will, Am. J. Phys. 71
(2003) 770, "Deflection of light to second order: A tool for illustrating
principles of general relativity." The authors discuss the deflection of
light by the Sun, with detailed computations in Schwarzschild, isotropic,
and harmonic coordinates. They show that, as you say, the *coordinate*
predictions differ, and give a careful explanation of the equivalence of
the *physical* predictions.
Steve Carlip
I.Vecchi
> I.Vecchi <vec...@weirdtech.com> wrote:
>
> [...]
> > 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 think you are seriously underestimating the amount of data reduction
> required. Look at the article in Matters of Gravity 26 (available from
> http://www.phys.lsu.edu/mog/) to see what is involved.
OK , no prediction.
.. .
I will argue that the distinction between GR anf QG, as you introduce
it here, is arbitrary.
> The basic point to remember is that
> *actual observations* are diffeomorphism-invariant.
No. Not in any physically meaningful way. Measurement is not
diffeomorphism invariant.
Wigner points out that GR in incomplete as long as long as no
space-time measurement model is properly defined and then he actually
sets out to define such a model . Essentially you are waving the
problem away through a linguistic trick. What you refer to as "actual
observations" are measurement outcomes or "events" in Wigner's
terminology :
"In relativity theory, the state is described by a metric which
consists of a network of points in space-time, that is, a network of
events, and the distances between these events. If we wish to translate
these general statements into something concrete, we must decide what
events are, and how we measure the distance between events" ([1]).
Your statement that "actual observations" (i.e. measurement outcomes or
"events") are diffeomorphism invariant is false. There are no local
observables in GR. If you set up a space-time measurement model, as you
actually do implicitly and Wigner explicitly, it will not be
diffeomorphism-invariant.
>We do not observe,
> for instance, the "coordinate value of the position of Mercury"; we observe
> things like "the round trip time of a radar pulse from a fixed location
> on Earth to Mercury and back, as measured by an atomic clock at that
> location," or "the angle between the light arriving from Mercury and
> that coming from a reference star, as measured at a particular telescope
> at a time determined by a clock at the location of that telescope."
Clock readings are measurement outcomes too. Essentially you are
introducing a time coordinate through your clock readings. Its values
are not diffeomorphism invariant.
> Such
> quantities do not depend on any choice of coordinates.
As I point out above, this is not true.
> To compare GR
> to observation, what you do is to compute (in, say, the post-Newtonian
> approximation) the predictions for such *observables*, and compare them
> to the the actual observations.
Expanding around the classical limit (or any other limit for that
matter) means that you are picking a privileged coordinate system to
begin with. Such a choice of coordinates in inherently meaningless in
GR and yet it plays a crucial role in the approximation you obtain.
In reality, when you resort to the post-Newtonian approximation, you
are already taking for granted " tacit assumptions which fix the
coordinate system" .
So it seems to me that when we talk about any experiment confirming GR
we are actually referring to a quantum gravity model (GR + an implicit
space-time measurement model) , where the space-time measurement model
is taken for granted.
Now, the scope of validity of such implicit space-time measurement
model is not clear "a priori".
In my previous post I suggested that this implicitly limits GR
predictive scope.
> Better, rather than just comparing GR,
> you look at a more general model (the *parametrized* post-Newtonian
> approximation, for instance), and find the best fit for your free
> parameters; you can then compare the result to GR, and at the same time
> get a good estimate for how good the fit is.
I doubt this has anything to do with the problem we are discussing. I
also wonder about the physical meaning of such free parameters and
their "fit", unless they are a way to fiddle the model to fit the
observed data .
>
> For a simple example, take a look at Boddener and Will, Am. J. Phys. 71
> (2003) 770, "Deflection of light to second order: A tool for illustrating
> principles of general relativity." The authors discuss the deflection of
> light by the Sun, with detailed computations in Schwarzschild, isotropic,
> and harmonic coordinates. They show that, as you say, the *coordinate*
> predictions differ, and give a careful explanation of the equivalence of
> the *physical* predictions.
>
Thank you for your suggested reading. However I believe that your
counter-argument misses the key point of Wigner's argument. Wigner
points to a deep problem of GR, i.e. its semantic incompletness and its
reliance on a tacitly assumed space-time measurement model. I do not
see how your argument addresses that issue.
Thanks you for your feedback. I think the issue we are discussing here
is important and not conceptually trivial. Discussing misconceptions
may help dispel them.
Best,
IV
[1] E. Wigner "Relativistic Invariance and Quantum Phenomena" Rev. Mod.
Phys. 29, 255 (1957)