Supercomputers and Dawn
Del Cecchi
fluid dynamics might find it of interest. Only 500 Tflops but
still.....
http://www.physorg.com/news173548668.html
del
Robert Myers
wrote:
I'll add it to my list of announcements of the conquest of the
Rayleigh-Taylor instability by the bomb labs.
Robert.
Del Cecchi
"Robert Myers" <rbmye...@gmail.com> wrote in message
news:a5720784-4837-4508...@j19g2000yqk.googlegroups.com...
Robert.
-----------------
Gee, a little snarky there aren't you? The page made no claims about
"conquest" of anything. They showed a simulation. Is the simulation
inaccurate?
del
Robert Myers
wrote:
> Gee, a little snarky there aren't you? The page made no claims about
> "conquest" of anything. They showed a simulation. Is the simulation
> inaccurate?
The nice thing about fluid mechanical simulations is that they so
often produce nice pictures. Adding processors by the tens of
thousands allows you to make the pictures more detailed and more
visually engaging.
A computer program obtainable on the open market and widely in use
will also produce very pretty pictures given enough resources under a
wide variety of fluid mechanical scenarios. It, too, uses local
differencing and would probably run well on Blue Gene. If you want
science, though, and not just pretty pictures, it might not be the way
to go.
The Rayleigh-Taylor instability is fundamental to what the labs want
to do with inertially confined fusion, and they've been working on it
for decades.
Robert.
Ken Hagan
<delcecchi...@gmail.com> wrote:
> Gee, a little snarky there aren't you?
Hah! That's not snarky. *This* is snarky...
> The page made no claims about "conquest" of anything. They showed a
> simulation. Is the simulation inaccurate?
Is there any way of knowing without doing the experiment? :)
Robert Myers
> On Tue, 17 Nov 2009 21:17:35 -0000, Del Cecchi
>
> > Gee, a little snarky there aren't you?
>
> Hah! That's not snarky. *This* is snarky...
>
> > The page made no claims about "conquest" of anything. They showed a
> > simulation. Is the simulation inaccurate?
>
> Is there any way of knowing without doing the experiment? :)
Historically, scientists and engineers have used all kinds of models,
some of them incredibly crude, as an aid in trying to gain insight
into actual physical behavior. As you suggest, the proof is always in
actually trying it out, whatever it is.
The DoD/DoE has increasingly come to rely on computer models as
substitutes for experiments that are hard to instrument, that are
unacceptably dangerous, or that can't be done, short of actual war or
violating something like the nuclear test ban treaty.
A reasonable person would ask, though, if it's possible to simulate
cracks (same article) in such detail, why is it that a patch to the
San Francisco-Oakland Bay bridge is so incompletely understood that it
comes flying off in the middle of traffic? If we understand fluid
flows so well, why is it that we always find out by accident what
parts of the shuttle's external fuel tank will come off and hit what
part of the shuttle with what kind of damage?
In the case of the work that Blue Gene was first hired to do, I'm
reasonably certain (no clearance here) that it is being used to offer
a reassurance that no computer is capable of offering. The endless
stream of pictures is little more than an elaborate bluff, or, even
worse, self-deception. The idea that we can equate (invent your own
prefix)-flops with science (or good engineering) is what's scary.
Robert.
Del Cecchi
Robert.
------------------
With respect to the nuclear stockpile, simulations are all we have and
presumably are better than nothing. I am hopeful that whatever
experiments are doable within the constraints of various treaties are
being done to make the simulations as accurate as possible.
The patch to the Bay Bridge was probably not analyzed or simulated at
all so blaming simulation technology for its failure is incorrect.
The analysis of the 35W bridge in Minneapolis that fell down made the
assumption that the gusset plates that held the beams together were
plenty strong and didn't need to be part of the analysis back then.
Unfortunately, somehow the plates were made of 1/2 inch thick steel
instead of 1 inch thick steel, and no one doing the analysis realized
that would be a problem. And it wasn't for a long time. Until one
day there was enough weight on the bridge and it fell down because a
gussett plate buckled.
So, are you saying that these simulations are a waste of time since we
don't necessarily understand the physical phenomena at a fine enough
level to properly model them? How do you feel about climate
simulations?
del
Del Cecchi
"Ken Hagan" <K.H...@thermoteknix.com> wrote in message
news:op.u3k4a...@khagan.ttx...
One would presume they are in fact doing so, since the simulations are
for the national ignition facility looking at inertial confinement
laser fusion.
Isn't how engineering is done these days? Simulate the design before
you build it, and then verify that the simulation is correct by the
performance of the resulting hardware? And the fancier the simulation
setup is the better able one is to correctly predict the behavior of
the hardware. And finding out the hardware doesn't work is a sign
that the simulation needs to be improved, perhaps because something
was left out either by design or by ignorance.
del
Robert Myers
wrote:
> So, are you saying that these simulations are a waste of time since we
> don't necessarily understand the physical phenomena at a fine enough
> level to properly model them? How do you feel about climate
> simulations?
When the nucelar test ban treaty was signed, I said to myself, "Well,
that's the end of nuclear weapons, because no one would be so insane
as to count on weapons that have gone untested for years."
I clearly underestimated how insane our defense establishment is.
During the Bush administration, the DoE, without a lot of fanfare,
increased the level of preparedness at the Nevada Test site so as to
shorten the time required to resume nuclear testing in a time of
heightened tension, so it's clear that at least some realize just how
far out on a limb we are.
I have complained loudly and repeatedly about the inappropriate level
of confidence that is being placed in climate predictions.
In general, any kind of open loop prediction that can't be subjected
to some kind of reality testing must be regarded with deep suspicion.
There's no shortage of science to be done, but the market for science
is shrinking. What's growing is the market for gigantic simulations
that claim to be able to make predictions that are demonstrably beyond
the capacity of any computer that we now possess or that we ever will
possess, barring some kind of breakthrough in fundamental
understanding. These realities have been understood for a *long*
time, and nothing about the spectacular successes of Moore's law has
changed them.
In the case of Blue Gene, I had from the very beginning a particular
concern about the kind of calculation it was limited to. The reality
is that people are building the computers they can afford to build and
using a measure of no demonstrated importance (Linpack flops) as
evidence that progress is being made. As far as I'm concerned, it's a
troubling and embarrassing con.
So far as predictions not made that matter to everyday citizens, like
whether a bridge patch will rattle itself to pieces in the wind, why
*aren't* they being made? They aren't being made because all the
resources are going into Dr. Strangelove calculations that are
unlikely to have any practical effect on anything, ever.
If I didn't think computers had the promise to advance science and the
practice of engineering, I wouldn't be here. I believe that progress
*is* being made, just not progress evidenced by visually stunning
graphics produced by computers whose only demonstrable virtue is their
size.
Robert.
Jonathan Thornburg [remove -animal to reply]
to be Robert Myers <rbmye...@gmail.com> wrote:
> In general, any kind of open loop prediction that can't be subjected
> to some kind of reality testing must be regarded with deep suspicion.
One way we can try to lower that suspicion level is by simulating
the same physical system in as many different ways as possible.
If simulations using very different numerical methods -- and even
very different ways of writing the basic equations -- give almost
identical results, then we can have a lot more confidence in those
results.
One big success story of this sort is the simulation of binary
black hole decay and coalescence via direct numerical solution of
the Einstein equations. Here (unlike some of what the bombheads do)
the basic physics equations are (we think) well understood and not
chaotic, so the difficulty is "just" in solving the equations.
Alas, the Einstein equations can be written in many different forms,
and the obvious forms have turned out to be unstable for this system.
(That is, the numerical solutions had spurious exponentially-growing
instabilities.) People first started working on this problem in the
early 1970s, but it wasn't until 2005 that a stable formulation of
the equations was found. (More accurately, until two quite different
stable formulations were found, completely independently, within a
few months of each other.)
There are now 3 very different types of numerical methods being
used to simulate this sort of system. Two use finite differencing
(but of very different sets of equations), and the third uses
pseudospectral methods (applied to still another different form of
the equations).
What's remarkable -- and I think quite relevant to Robert Myers'
comments -- is that the results from these very different simulations
all line up beautifully. (For pretty pictures, see the graphs in
http://arxiv.org/abs/0901.2437)
It's this consistency between simulations using very different
techniques that makes people tend to believe the results.
--
-- "Jonathan Thornburg [remove -animal to reply]" <jth...@astro.indiana-zebra.edu>
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
"C++ is to programming as sex is to reproduction. Better ways might
technically exist but they're not nearly as much fun." -- Nikolai Irgens
Robert Myers
<jth...@astro.indiana-zebra.edu> wrote:
>
> It's this consistency between simulations using very different
> techniques that makes people tend to believe the results.
>
The DoD has used this technique in a fairly random way, and it's my
experience with some of these exercises that has left me so deeply
skeptical.
Different groups using different codes with different numerical
methods often came up with wildly different predictions, and the
prediction that tended to win was the one that the customer wanted to
hear.
With so much of the computational capacity, extravagant graphics, and
PR machinery tied to the people who own the biggest computers (which
all increasingly report to the Secretary of Energy), there is little
chance that any kind of budget-threatening disagreement will be
exposed.
One of my fears about Blue Gene type computers is that they will kill
off spectral and pseudospectral methods forever. The recent purchase
of a Blue Gene by NOAA deepens my paranoia. This is not the place for
a deep discussion of why I think those methods are so important; it
may be simply a matter that I find it much easier to think about
what's really happening in highly nonlinear systems with such
approaches.
In the end, even in astrophysics, you need a prediction you can
check. I sure hope we get some data out of the search for
gravitational waves real soon.
Robert.
Eugene Miya
Robert Myers <rbmye...@gmail.com> wrote:
>On Nov 19, 1:22=A0pm, "Jonathan Thornburg [remove -animal to reply]"
><jth...@astro.indiana-zebra.edu> wrote:
>> It's this consistency between simulations using very different
>> techniques that makes people tend to believe the results.
I would add "with a little more than that."
>The DoD has used this technique in a fairly random way, and it's my
>experience with some of these exercises that has left me so deeply
>skeptical.
>
>Different groups using different codes with different numerical
>methods often came up with wildly different predictions, and the
>prediction that tended to win was the one that the customer wanted to hear.
What means "win?"
>With so much of the computational capacity, extravagant graphics, and
>PR machinery tied to the people who own the biggest computers (which
>all increasingly report to the Secretary of Energy), there is little
>chance that any kind of budget-threatening disagreement will be exposed.
I would not say that. I certainly know empiricists at LANL working on
Pu-aging who compare simulation graphics to "God damm Microsoft screen
savers..." That's Smith's problem.
>One of my fears about Blue Gene type computers is that they will kill
>off spectral and pseudospectral methods forever. The recent purchase
>of a Blue Gene by NOAA deepens my paranoia. This is not the place for
>a deep discussion of why I think those methods are so important; it
>may be simply a matter that I find it much easier to think about
>what's really happening in highly nonlinear systems with such
>approaches.
I seriously doubt spectral methods would be killed off. From what I
have seen for analysis, they tend to be somewhat over used as an
algorithm of choice (#5 of the list of 10) as a means of basic 1st order
blind analysis.
>In the end, even in astrophysics, you need a prediction you can
>check. I sure hope we get some data out of the search for
>gravitational waves real soon.
We have 1 data point for our given universe.
Anyways going on vacation in a week.
Back from SC'09.
--
Looking for an H-912 (container).
Robert Myers
.
>
> What means "win?"
>
Like the infamous chart based on a wildly inappropriate model that
allowed certain decisions by NASA managers to be made.
My experience has been is that the "right" answer is the one that
leads to "the budget is approved" or "the mission can go forward
without further schedule slippage."
"It's all just %#@$ screensavers" is almost as bad, and it is
ultimately rooted in the same bureaucratic survival instinct. You're
a survivor, Eugene. How could you not know?
Robert.
Jonathan Thornburg
>In the end, even in astrophysics, you need a prediction you can
>check. I sure hope we get some data out of the search for
>gravitational waves real soon.
Eugene Miya <eug...@cse.ucsc.edu> wrote:
> We have 1 data point for our given universe.
The prospects for getting more data, a.k.a. for direct detection of
gravitational waves, are seriously off-topic for this newsgroup,
but if you're interested, I reviewed them over in sci.physics.research
a few months ago:
http://groups.google.com/group/sci.physics.research/browse_thread/thread/8ef107ad92c3bce8
The short synopsis is that given upgrades that given the hardware
upgrades that are in the pipeline (funding looks ok, installation
is just beginning now) the current ground-based gravitational-wave
detectors (LIGO, Virgo, Geo) have excellent chances to see signals
with moderate signal/noise ratios by 2013-2015. If the planned LISA
space-based gravitational-wave detector is funded to fly in ~2020 as
planned (hoped for!), it will detect lots and lots of signals at very
high signal/noise ratios (allowing very precise comparisons with the
simulations).
Jonathan Thornburg
[[various things]]
Oops. Due to carless editing I neglected to remove a whole bunch of
text that's irrelevant to what I was trying to say. Please ignore
everything in my immediately-previous-message-in-this-thread prior
to the lines
Robert Myers <rbmye...@gmail.com> wrote:
>In the end, even in astrophysics, you need a prediction you can
>check. I sure hope we get some data out of the search for
>gravitational waves real soon.
Sorry for the noise... :(
--
-- "Jonathan Thornburg [remove -animal to reply]" <jth...@astro.indiana-zebra.edu>
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
Alberto Moreira (comp.arch, Jan 1999):
"Cache is, basically, a kluge generated by technological restrictions."
Donald C. Lindsay (also comp.arch, Jan 1999):
"No. It's a kluge generated by deeply fundamental restrictions,
like the speed of light."
Robert Myers
>
> >In the end, even in astrophysics, you need a prediction you can
> >check. I sure hope we get some data out of the search for
> >gravitational waves real soon.
> Eugene Miya <eug...@cse.ucsc.edu> wrote:
> > We have 1 data point for our given universe.
>
> The prospects for getting more data, a.k.a. for direct detection of
> gravitational waves, are seriously off-topic for this newsgroup,
> but if you're interested, I reviewed them over in sci.physics.research
> a few months ago:
>
> The short synopsis is that given upgrades that given the hardware
> upgrades that are in the pipeline (funding looks ok, installation
> is just beginning now) the current ground-based gravitational-wave
> detectors (LIGO, Virgo, Geo) have excellent chances to see signals
> with moderate signal/noise ratios by 2013-2015. If the planned LISA
> space-based gravitational-wave detector is funded to fly in ~2020 as
> planned (hoped for!), it will detect lots and lots of signals at very
> high signal/noise ratios (allowing very precise comparisons with the
> simulations).
>
of things.
If you think about the decades or work that preceded Kepler's Laws of
planetary motion, the time spent looking for gravitational wa doesn't
seem all that discouraging.
Science has a problem very similar to the problem that computer
architecture has: will a thousand scientists working on a problem make
the solution arrive a thousand times faster?
There is no reason to abandon a quest because it is time-consuming and
exceedingly difficult.
The problem I'm having is with bureaucrats whose mantra has come down
to more acres, more megawatts, and more pretty pictures.
Robert.
Del Cecchi
OK, say the bird of paradise flew up your nose and you are put in charge
of the nuclear stockpile at DOE. What would be your approach to
verifying that the weapons would in fact work correctly in the unlikely
and undesired circumstance that they needed to be used. Please do not
address the issue as to whether such a circumstance could ever exist.
Or any other political issue. Confine yourself to the technical issue.
Robert Myers
wrote:
>
> OK, say the bird of paradise flew up your nose and you are put in charge
> of the nuclear stockpile at DOE. What would be your approach to
> verifying that the weapons would in fact work correctly in the unlikely
> and undesired circumstance that they needed to be used. Please do not
> address the issue as to whether such a circumstance could ever exist.
> Or any other political issue. Confine yourself to the technical issue.
http://www.nytimes.com/2005/04/03/science/03nuke.html?_r=1&pagewanted=all&position
You want me to wind up another Charles Cremer?
Based on my personal experience with large fluid mechanical
simulations and the social behavior of scientists in the business, I
wouldn't trust any conclusion that couldn't be supported with
empirical evidence and calculations that could be performed on an
engineering workstation costing less than $100K. I would allow for
more expensive computers for data analysis, if necessary. I would
move as much of the basic science out of the bomb labs as I
conceivably could. I'd be spending lots of money at the Nevada test
site, in case I decided that I had no choice but to abrogate the test
ban treaty in a really big hurry. I'd know just exactly what
experiments I would perform, I'd be ready to perform them, and I'd be
doing as much contingency planning as possible to deal with the
possibility of an unwelcome suprise.
Robert.
Jonathan Thornburg [remove -animal to reply]
[[how to tackle a difficult science/engineering problem]]
> Based on my personal experience with large fluid mechanical
> simulations and the social behavior of scientists in the business, I
> wouldn't trust any conclusion that couldn't be supported with
> empirical evidence and calculations that could be performed on an
> engineering workstation costing less than $100K. I would allow for
> more expensive computers for data analysis, if necessary.
I find this strategy confusing. Why $100K, and not (say) $10K, or $1000K,
or some other nice round number? That is, what is it about computations
that take less than number-of-teraflops/terabytes/etc-that-$100K-will-buy
that makes them inherently more reliable and less prone to tell-the-boss-
-what-she-wants-to-hear problems than calculations (say) 10 times that
size? Not to mention that $100K N years from now probably will buy a
lot more computation than $100K today.
Having worked for many years in research groups doing both small-scale
and supercomputer simulations, I can well appreciate the "social behavior
of scientists" problems (not to mention "social behavior of managers of
scientists"!) that can arise, but I don't see how putting a rather
arbitrary cap on the size of the computers that can be used for
simulations (though not for "data analysis") necessarily solves
these problems.
--
-- "Jonathan Thornburg [remove -animal to reply]" <jth...@astro.indiana-zebra.edu>
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
Robert Myers
<jth...@astro.indiana-zebra.edu> wrote:
> Robert Myers <rbmyers...@gmail.com> wrote:
>
> [[how to tackle a difficult science/engineering problem]]
>
> > Based on my personal experience with large fluid mechanical
> > simulations and the social behavior of scientists in the business, I
> > wouldn't trust any conclusion that couldn't be supported with
> > empirical evidence and calculations that could be performed on an
> > engineering workstation costing less than $100K. I would allow for
> > more expensive computers for data analysis, if necessary.
>
> I find this strategy confusing. Why $100K, and not (say) $10K, or $1000K,
> or some other nice round number? That is, what is it about computations
> that take less than number-of-teraflops/terabytes/etc-that-$100K-will-buy
> that makes them inherently more reliable and less prone to tell-the-boss-
> -what-she-wants-to-hear problems than calculations (say) 10 times that
> size? Not to mention that $100K N years from now probably will buy a
> lot more computation than $100K today.
>
> Having worked for many years in research groups doing both small-scale
> and supercomputer simulations, I can well appreciate the "social behavior
> of scientists" problems (not to mention "social behavior of managers of
> scientists"!) that can arise, but I don't see how putting a rather
> arbitrary cap on the size of the computers that can be used for
> simulations (though not for "data analysis") necessarily solves
> these problems.
>
The point is to have a price at which many investigators can
participate and many calculations can be performed. If a calculation
is so large that it can be performed only on one or two computers or
only by one or two groups, it is by definition probably wrong, as far
as I'm concerned. If I speak bluntly, it's because that's the only
kind of language some people understand. I just said it, and I mean
no offense to Del: all the money being spent on one-of-a-kind
calculations and one-of-a-kind computers might just as well be put
into cargo containers and dumped into the Pacific.
If the details are so complicated that they can't really be examined
by a competent and unbiased observer with means readily available to a
competent and unbiased observer, they shouldn't be trusted. Would you
want something so important as the warheads in question to be treated
with any less skepticism?
The $100K figure is arbitrary. I figure that if some wingnut who
would be likely to belong to (say) the Union of Concerned Scientists
or IPPNW wants to badly enough, he can scrounge up access to such a
system. Maybe $1M is ok, depending on the resourcefulness of the
wingnut.
Tens of thousands of processors, megawatts of electricity, and acres
of computer center not only build empires, they can also hide
incompetence and poor judgment.
Robert.
Del Cecchi
Robert.
---------------------------------
Neither you nor the NY Times answered my question. We have a bunch of
warheads that we want some assurance that they work. From the
article, we might even be designing new weapons that we also would
like some assurance of them working.
Abrogating the test ban treaty seems unlikely. We have some data from
detonations in the past for existing designs. That data may or may
not show a problem with one of the designs in the stockpile. ( I
don't know if the guy warning is/was privy to the results of the
testing).
It appears there are two choices. Well, three if you count setting
some off. Either you can apply all the computational resources
necessary (or at least as much as you can get) to do simulations Or
you can do nothing except basic science until you come up with a way
to analyze the situation in a more (not sure of what to call it.
Straightforward? Rigorous? Intuitive?) way that is not so dependent
on simulation. In the meantime you have considerable uncertainty
unless you also simulate your ass off, to coin a phrase.
Obama just named you head of dept of energy (I think that is who is in
charge of making bombs). What do you do? Or at least, what would
you do knowing what you know now?
del
PS Do you really think that someone offed Cremer? Or was that just a
smart remark?
Bernd Paysan
> Neither you nor the NY Times answered my question. We have a bunch of
> warheads that we want some assurance that they work.
No, you don't. You want *the others* to believe that they actually
work. The whole point about having lots of nukes is to make damn'd sure
you'll never going to use them, regardless whether they work or not.
Come on, nukes have been developed before 1945. You only need three
pieces of equipment to calculate a nuke: A blackboard, a chalk, and a
Feynman, and maybe your result is slightly off (by less than the
measurement uncertainty, though). If you are going to deploy massive
supercomputers, it's not because you need them to calculate the nukes,
it's because you need them to convince *the others* that you are taking
serous efforts to keep your nukes functional.
It's a lot easier for the Russians. They know their nukes work, because
they have calculated that on a blackboard (it is probably 10 times less
accurate than the Feynman result, but still better than measurements,
and it has been completed before running out of precious chalk). They
know the Americans know that this blackboard calculation is completely
sufficient. They don't have to threaten every other idiot on the world,
because that's the job of the Americans.
--
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://www.jwdt.com/~paysan/
Robert Myers
wrote:
>
> Neither you nor the NY Times answered my question. We have a bunch of
> warheads that we want some assurance that they work. From the
> article, we might even be designing new weapons that we also would
> like some assurance of them working.
>
> Abrogating the test ban treaty seems unlikely. We have some data from
> detonations in the past for existing designs. That data may or may
> not show a problem with one of the designs in the stockpile. ( I
> don't know if the guy warning is/was privy to the results of the
> testing).
>
> It appears there are two choices. Well, three if you count setting
> some off. Either you can apply all the computational resources
> necessary (or at least as much as you can get) to do simulations Or
> you can do nothing except basic science until you come up with a way
> to analyze the situation in a more (not sure of what to call it.
> Straightforward? Rigorous? Intuitive?) way that is not so dependent
> on simulation. In the meantime you have considerable uncertainty
> unless you also simulate your ass off, to coin a phrase.
>
> Obama just named you head of dept of energy (I think that is who is in
> charge of making bombs). What do you do? Or at least, what would
> you do knowing what you know now?
>
I'm sorry if you feel that I didn't answer your question. I told you
how I would spend money if I had it to spend, and you don't seem to
like my answer.
Your set of requirements seems to presuppose that there is some
calculation that can be performed at some cost that will address the
concern(s) in the New York Times article. I don't believe that there
is any such calculation. In the end, until you take a few of those
warheads and set them off, you are just guessing.
The computational resources that were available when the warheads were
designed were a big number with a lot of zeroes behind it smaller than
what the $100K workstation I offered could provide. If you can't work
it out with those kinds of resources, then I don't think you
understand the science well enough to rely on those warheads. Period.
If it were our desire to convince someone that we mean business,
activity at the Nevada test site would count a whole lot more than
running computers, no matter how big. Perhaps the DoE isn't quite
that eager to convince anyone that we really mean business, or perhaps
the work continues frantically apace and is black.
Barring that, this country must reinvest in basic science, and there
is no reason not to use investment in basic science to support
critical defense needs.
My fundamental point is that calculations alone will do little to
alleviate uncertainty. It is my indelible prejudice that, the bigger
the calculation, the more skeptical I become.
I don't necessarily, BTW, see anything wrong with using huge
computations judiciously to hit yourself over the head with a clue
stick. If you can't reduce your understanding to something more
amenable to wide scrutiny, though, you are just kidding yourself. I
know the culture first hand.
> PS Do you really think that someone offed Cremer? Or was that just a
> smart remark?
I meant only to comment on the nature of the egos involved and how
they behave when their kingdoms are threatened.
Robert.
Robert Myers
> It's a lot easier for the Russians.
I don't believe that the Russians rely on a warhead like the W-76, at
least not to the extent that we do.
The problem that got Dick Morse agitated is real, and the kinds of
tests that people apparently designed to deal with uncertainty before
the test ban would probably not deal with the kinds of subtleties we'd
have to worry about now. Careful reading of the New York Times
article, some knowledge of fluid mechanics, a little reading in the
unclassified literature, and examination of the picture Del originally
offered should convince you that the problem is very hard.
Robert.
Jonathan Thornburg
> Based on my personal experience with large fluid mechanical
> simulations and the social behavior of scientists in the business, I
> wouldn't trust any conclusion that couldn't be supported with
> empirical evidence and calculations that could be performed on an
> more expensive computers for data analysis, if necessary.
and later
> The point is to have a price at which many investigators can
> participate and many calculations can be performed. If a calculation
> is so large that it can be performed only on one or two computers or
> only by one or two groups, it is by definition probably wrong, as far
> as I'm concerned. If I speak bluntly, it's because that's the only
> kind of language some people understand. I just said it, and I mean
> no offense to Del: all the money being spent on one-of-a-kind
> calculations and one-of-a-kind computers might just as well be put
> into cargo containers and dumped into the Pacific.
>
> If the details are so complicated that they can't really be examined
> by a competent and unbiased observer with means readily available to a
> competent and unbiased observer, they shouldn't be trusted.
I agree, scientific results that take so much effort to replicate
that nobody else in the world can manage it, are troublesome. But...
this is hardly unique to supercomputing -- you see the same issues
with any one-of-a-kind facility. So... should I distrust all the
scientific results from the LHC because nobody else has a particle
accelerator that big? Should I distrust all the scientific results
from Voyager 2's flybys of Uranus and Neptune because no other
spacecraft has (yet) made it to either planet? Should I distrust
all the unique observations from the Hubble Space Telescope because
nobody else has a comparable space telescope available?
The question of how to validate scientific/engineering results
obtained using one-of-a-kind facilities (whether supercomputers or
space telescopes) is not an easy one... but I do think you're
throwing out the baby with the bathwater by basically saying
"don't trust any results from one-of-a-kind facilities".
In particular, having multiple (competing) research groups (think
Los Alamos vs Livermore) using those one-of-a-kind facilities, as
well as careful reviews by outsiders (think Richard Garwin, or the
Jasons), can greatly enhance our confidence that things are being
done honestly.
I do agree completely that press releases & photo ops touting
"our new supercomputer has a gazillion {processors/watts/acres/etc}"
are only very minimally informative.
The example I gave in a previous posting in this thread (simulations
of binary black hole coalescence) involved half-a-dozen or so competing
research groups. The computers used typically cost something on the
order of US$ 1 million or so; lots of university researchers have access
to systems of this size. Typical simulations take a week or so on such
a system. You could _do_ these calculations on a $100K system, but
you'd rather not, because then you'd have to do one or more of:
* cut the resolution significantly (making the results much less
accurate)
* run fewer cases (making it hard to figure out what effects are real)
* wait months of each run
--
-- "Jonathan Thornburg [remove -animal to reply]" <jth...@astro.indiana-zebra.edu>
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
"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
Robert Myers
<jonat...@nitrogen.astro.indiana.edu> wrote:
> I agree, scientific results that take so much effort to replicate
> that nobody else in the world can manage it, are troublesome. But...
> this is hardly unique to supercomputing -- you see the same issues
> with any one-of-a-kind facility. So... should I distrust all the
> scientific results from the LHC because nobody else has a particle
> accelerator that big? Should I distrust all the scientific results
> from Voyager 2's flybys of Uranus and Neptune because no other
> spacecraft has (yet) made it to either planet? Should I distrust
> all the unique observations from the Hubble Space Telescope because
> nobody else has a comparable space telescope available?
>
> The question of how to validate scientific/engineering results
> obtained using one-of-a-kind facilities (whether supercomputers or
> space telescopes) is not an easy one... but I do think you're
> throwing out the baby with the bathwater by basically saying
> "don't trust any results from one-of-a-kind facilities".
>
With Hubble (and with other NASA unmanned platforms), we've had a
chance to see that big errors can slip by even in large hardware
systems, which generally have multiple layers of sign-off and review
that computations usually don't.
I've had to go deep into codes to find the subtle hack that was giving
wrong answers, and I don't think that the list of people with the
skill, patience, and insight to do that kind of work is a long one.
By comparison, the list of people who could fiddle parameters to get
the answer they wanted, working entirely without oversight, is rather
long, and the cost of mistakes is low.
It isn't nearly such an easy matter to cover your biases and screwups
with hardware. If the multiple layers of review fail (and they do),
you discover the error the hard way.
All that said, I *am* very skeptical of what the big machines are
producing. As one field theorist told me candidly, these experiments
become more and more dependent on simulation all the time, just as the
gravity wave effort is now dependent on (apparently fairly subtle)
simulation.
If any of these efforts makes a big mistake that results in
substantively misleading answers, the cost is to set back science some
number of years or billions of dollars. If the reliability of the
W-67 matters at all (and we hope that it doesn't), then mistakes
comparable to what we accept from NASA missions or other big science
experiments are simply unacceptable.
> In particular, having multiple (competing) research groups (think
> Los Alamos vs Livermore) using those one-of-a-kind facilities, as
> well as careful reviews by outsiders (think Richard Garwin, or the
> Jasons), can greatly enhance our confidence that things are being
> done honestly.
>
The Jasons? You're kidding, right? For the most part, they all
belong to the same club. You don't seriously expect them to humiliate
their own graduate students, do you? And when the Jasons get too
uppity, the DoD reviews the contract so as to send a clear message.
That's happened, too.
Los Alamos v. Livermore? That's cute. That's like the different
divisions of IBM the monopolist competing. The deal is that every lab
has its senators, and every lab is going to get its computers, its
photo ops, and its continued funding.
> I do agree completely that press releases & photo ops touting
> "our new supercomputer has a gazillion {processors/watts/acres/etc}"
> are only very minimally informative.
>
> The example I gave in a previous posting in this thread (simulations
> of binary black hole coalescence) involved half-a-dozen or so competing
> research groups. The computers used typically cost something on the
> order of US$ 1 million or so; lots of university researchers have access
> to systems of this size. Typical simulations take a week or so on such
> a system. You could _do_ these calculations on a $100K system, but
> you'd rather not, because then you'd have to do one or more of:
> * cut the resolution significantly (making the results much less
> accurate)
> * run fewer cases (making it hard to figure out what effects are real)
> * wait months of each run
>
Then why are you fighting back so hard? You sound like you know what
you're doing. You say the right kinds of things. Your machines and
calculations would all fit into the criteria I've set out.
Robert.
Peter Grandi
>> The point is to have a price at which many investigators can
>> participate and many calculations can be performed. If a
>> calculation is so large that it can be performed only on one
>> or two computers or only by one or two groups, it is by
>> definition probably wrong, as far as I'm concerned. [ ... ]
>> If the details are so complicated that they can't really be
>> examined by a competent and unbiased observer with means
>> readily available to a competent and unbiased observer, they
>> shouldn't be trusted.
> I agree, scientific results that take so much effort to
> replicate that nobody else in the world can manage it, are
> troublesome. But... this is hardly unique to supercomputing
> -- you see the same issues with any one-of-a-kind facility.
> So... should I distrust all the scientific results from the
> LHC because nobody else has a particle accelerator that big?
> Should I distrust all the scientific results from Voyager 2's
> flybys of Uranus and Neptune because no other spacecraft has
> (yet) made it to either planet? Should I distrust all the
> unique observations from the Hubble Space Telescope because
> nobody else has a comparable space telescope available?
The answer is "yes" if only because mistakes do happen. Do you
distrust all the theorems that you read in peer reviewed math
journals? All the algorithms published in prestigious numerical
analysis books? You should as well.
Decades ago there was a fashion for proving programs correct, an
idea that was fundamentally unsound and bad anyhow. But it was
based to some extent on math-worship: the idea that theorems are
somehow more reliable than code (so turning code into theorems
was the silver bullet). So some funny people went and checked
how many theorems in excellent peer reviewed journals turned out
to be "buggy", and they found lots and lots of examples.
> The question of how to validate scientific/engineering results
> obtained using one-of-a-kind facilities (whether supercomputers
> or space telescopes) is not an easy one... but I do think
> you're throwing out the baby with the bathwater by basically
> saying "don't trust any results from one-of-a-kind facilities".
But he is not quite saying "believe none of them". It is a matter
of degree: trust them a lot less than independently verified
results. And don't waste money on them. Given scarce resources,
and familiarity with how buggy computing stuff is, investing in
unverifiable results at the expense of verifiable ones seems
a waste to me too.
There are *many* good reasons why repeateability of experiment is
the cornerstone of science, and if results are not independently
verified they should be treated with much greater caution than
those who are.
Also RobertM focuses on distrusting not *all* single facility
results, but specifically the results of single computational
facility simulations that are in an area that is *known* to be
rather intractable.
Anyhow the problem is even before the LHC and similar unique
facilities; I remember reading many years agq an article that
mentioned a particular obstacle to trusting experimental physics
results, and it was written when there were multiple equivalent
facilities: that data acquisition and analysis programs were
shared by nearly all of them. Some times the same old tired codes
written by someone for their PhD work in a great hurry were passed
down the decades to everybody in the field. This situation is
particularly grave when the conclusions for several types of
critical experiments is entirely indirect arising out of data
filtering and analysis.
And given that software development is costly and takes time, do
you think that some of the precious budget for experimental
physics is going to be "wasted" anytime soon on writing again
existing, "know working" codes, just for the sake of doing it
differently?
Since that article was written computing has become more of a
monoculture; once upon a time CDC and IBM had completely different
floating point models, and memory architectures, and so on; right
now the monoculture is very strong.
If you want to understand how subtle the issues can be, consider
reading this paper on "undetected" errors in storage systems:
https://indico.desy.de/contributionDisplay.py?contribId=65&sessionId=42&confId=257
As a guy once said in an interview to Datamation, "as far as I
know the computing center I manage never from an undetected
error". Think carefully about that.
And as RobertM writes, add in not just genuinely undetected
errors, but errors that if detected and reported might harm
careers.
nm...@cam.ac.uk
Peter Grandi <pg...@0810.exp.sabi.co.UK> wrote:
>
>If you want to understand how subtle the issues can be, consider
>reading this paper on "undetected" errors in storage systems:
>
> https://indico.desy.de/contributionDisplay.py?contribId=65&sessionId=42&confId=257
Er, I would call those the obvious ones :-( There is a much more
evil category, which I know is both more common than people think
and is usually undetected, which is inconsistencies at the next
level up (or two up). That is often caused by race conditions
causing an inconsistent set of blocks to be written.
It can be missed for ages if the pattern of use is such that only
a single 'view' is observed, but a different use of the data can
give incorrect results. Surprisingly few systems have any real
checking at that level, and often don't even have a precise enough
specification to check rigorously. When I wrote my file system
scanner, aimed at a higher level, I was surprised at how many errors
at the lower level showed up. And a lot of the problem is file
systems that automatically recover from failure, because fsck is
rarely run and is usually not very thorough in any case.
I agree with the paper's points. Checksums etc. that were designed
for the odd million transactions just don't cut the mustard when
dealing with the odd million million transactions. But that's all
basic probabilty theory.
>As a guy once said in an interview to Datamation, "as far as I
>know the computing center I manage never from an undetected
>error". Think carefully about that.
Er, yes, but there is a syntax error :-) I think that you meant
to add the word "fail" after "never", but am not sure. If so,
yes, I agree with you - as the American saying goes "It's not
what we don't know that causes the trouble, it's what we know
that ain't so."
Regards,
Nick Maclaren.
Robert Myers
> Decades ago there was a fashion for proving programs correct, an
> idea that was fundamentally unsound and bad anyhow. But it was
> based to some extent on math-worship: the idea that theorems are
> somehow more reliable than code (so turning code into theorems
> was the silver bullet). So some funny people went and checked
> how many theorems in excellent peer reviewed journals turned out
> to be "buggy", and they found lots and lots of examples.
It's hard to believe that mathematics won't eventually turn to some
degree of automated correctness checking.
Everyone is agreed: there is no silver bullet (at least not that
anyone knows of), but there are lots of things we could be doing that
we don't do, in part because there is no standards body with
sufficient resources or clout.
Robert.
Jonathan Thornburg [remove -animal to reply]
[[about errors in storage systems]]
> There is a much more
> evil category, which I know is both more common than people think
> and is usually undetected, which is inconsistencies at the next
> level up (or two up). That is often caused by race conditions
> causing an inconsistent set of blocks to be written.
>
> It can be missed for ages if the pattern of use is such that only
> a single 'view' is observed, but a different use of the data can
> give incorrect results.
I once spent a number of years at an institution that used AFS to
share home directories across a few hundred mostly-Linux workstations
and some (though not all) of our larger servers. Once we had a
building-wide power failure. After everything restarted the AFS cache
on "my" workstation was left in a "peculiar" state, where one particular
file (it was a source-code file I was actively editing) displayed
(reproducibly) different contents depending on whether it was viewed
with 'vi' or 'less' or 'cat' (I think 2 of those were the same and the
3rd different, but I no longer remember which was which. Presumably
they were using different sequences of system calls to access the file.)
Our system administrators eventually wiped the AFS cache to "cure" the
problem.
Another classic example just what Nick is describing, but at a much
lower level, is the (hardware) cache corruption described in the chapter
"The Case of the Missing NAND Gate" in Tracy Kidder's "The Soul of a
New Machine".
--
-- "Jonathan Thornburg [remove -animal to reply]" <jth...@astro.indiana-zebra.edu>
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
"There's no such thing as a simple cache bug." --Rob Pike
Robert Myers
<jth...@astro.indiana-zebra.edu> wrote:
>
> Another classic example just what Nick is describing, but at a much
> lower level, is the (hardware) cache corruption described in the chapter
> "The Case of the Missing NAND Gate" in Tracy Kidder's "The Soul of a
> New Machine".
>
The ultimate classic example is pure fantasy, the novel "Fail Safe,"
where the doomsday scenario begins with a blown capacitor.
Look up "Cold War Fantasies"
http://www.amazon.com/Cold-War-Fantasies-Fiction-Foreign/dp/0742510522
on Google Books (link is too long) and start reading on page 86.
The new fantasy is that we have left that world behind us. Every time
we hear about some new wing-ding computer at one of the bomb labs, we
should be reminded that that is the world we are talking about.
I don't want to get political. My concern here is about the science
and engineering realities.
Robert.
Quadibloc
While the theory of the atom bomb is sufficiently well understood that
you don't really need more than a blackboard and chalk to be sure that
one will work, a hydrogen bomb depends on the behavior of a hot gas
under pressure under conditions that aren't easy to replicate in the
laboratory.
So to claim that a computer simulation of an H-bomb might not
accurately represent its real-world behavior doesn't strike me,
either, as an eccentric claim that indicates the concerned individual
doesn't believe Newton and Einstein knew what they were doing.
Underground nuclear tests do not put fallout into the atmosphere, and
so there is really no reason to jeopardize national security by not
resuming them, except, of course, if the test ban is to the United
States' advantage.
John Savard
Del Cecchi
"Quadibloc" <jsa...@ecn.ab.ca> wrote in message
news:8a5a11f7-f50f-4dac...@2g2000prl.googlegroups.com...
John Savard
-----------------------------------
Clearly the reasons are political, in the most general sense of the
word, not technical. The test ban treaty has not stopped north korea
from testing, nor will it stop Iran.
But the idea that Obama is going to abrogate the treaty and go back to
testing nuclear weapons underground just to make sure they still go
off strikes me as extremely unlikely, given his remarks of April 05,
2009
http://www.cnn.com/2009/WORLD/europe/04/05/czech.republic.obama/index.html
Robert Myers
wrote:
> "Quadibloc" <jsav...@ecn.ab.ca> wrote in message
>
>
> Clearly the reasons are political, in the most general sense of the
> word, not technical. The test ban treaty has not stopped north korea
> from testing, nor will it stop Iran.
>
> But the idea that Obama is going to abrogate the treaty and go back to
> testing nuclear weapons underground just to make sure they still go
> off strikes me as extremely unlikely, given his remarks of April 05,
> 2009
>
And I've never proposed that he would--at least not under normal
circumstances.
The DoE did not, so far as I know (and everything that I know is based
on unclassified information), contemplate the kind of problem that the
aging W-76 presents when it planned its final tests. The fact that
the DoE now believes that it has to do such gigantic (and presumably
fine-scale) simulations should offer a clue as to what Dick Morse et
al became worried about.
A complete discussion of the options could not omit mention of the
National Ignition Facility
http://en.wikipedia.org/wiki/National_Ignition_Facility
which conceivably could offer the possibility of meaningful
experimental evidence, and I suspect that the claimed options change
frequently, especially in light of what is happening in Iran.
The sheer size and complexity of all of this should give anyone
pause. With any normal project, no matter how complicated, in the end
you build it, and either it works or it doesn't. That won't happen in
this case, short of what I agree is extremely unlikely.
Robert.
EricP
> On Nov 29, 9:48 pm, "Del Cecchi" <delcecchioftheno...@gmail.com>
> wrote:
>> "Quadibloc" <jsav...@ecn.ab.ca> wrote in message
>>
>>
>> Clearly the reasons are political, in the most general sense of the
>> word, not technical. The test ban treaty has not stopped north korea
>> from testing, nor will it stop Iran.
>>
>> But the idea that Obama is going to abrogate the treaty and go back to
>> testing nuclear weapons underground just to make sure they still go
>> off strikes me as extremely unlikely, given his remarks of April 05,
>> 2009
>>
>> http://www.cnn.com/2009/WORLD/europe/04/05/czech.republic.obama/index...
>
> And I've never proposed that he would--at least not under normal
> circumstances.
>
> The DoE did not, so far as I know (and everything that I know is based
> on unclassified information), contemplate the kind of problem that the
> aging W-76 presents when it planned its final tests. The fact that
> the DoE now believes that it has to do such gigantic (and presumably
> fine-scale) simulations should offer a clue as to what Dick Morse et
> al became worried about.
I don't believe the simulations have much to do
with verifying the current stockpile.
The Bush admin. made loud mutterings about developing new
warheads - the Reliable Replacement Warhead for one,
nuke bunker busters for another (to rattle the Iranians I imagine).
The people below seems to know a lot on these subjects.
News Analysis: Officials Air Views on Key Stockpile Issue
http://www.armscontrol.org/act/2009_11/NewsAnalysis
New Nuclear Policies, New Weapons, New Dangers
http://www.armscontrol.org/factsheets/newnuclearweaponsissuebrief
It may also be that the simulator allows nuke engineers
to do their their jobs without blowing things up.
Otherwise they would get bored and move to other jobs.
In other words, it allows the US to retain an experienced
design workforce.
> A complete discussion of the options could not omit mention of the
> National Ignition Facility
>
> http://en.wikipedia.org/wiki/National_Ignition_Facility
>
> which conceivably could offer the possibility of meaningful
> experimental evidence, and I suspect that the claimed options change
> frequently, especially in light of what is happening in Iran.
>
> The sheer size and complexity of all of this should give anyone
> pause. With any normal project, no matter how complicated, in the end
> you build it, and either it works or it doesn't. That won't happen in
> this case, short of what I agree is extremely unlikely.
The following is also an interesting source.
The authors of articles are usually people in positions
to both know and decide. For example one of the the following
authors from 2006 is by a U.S. Assistant Secretary of Defense:
Nuclear Exchange: Does Washington Really Have (or Want) Nuclear Primacy?
http://www.foreignaffairs.com/articles/61931/peter-c-w-flory-keith-payne-pavel-podvig-alexei-arbatov-keir-a-l/nuclear-exchange-does-washington-really-have-or-?page=show
That is a rebuttle to the article
The Rise of U.S. Nuclear Primacy
http://www.ituassu.com.br/theriseofusnuclearprimacy.pdf
It is an interesting website to poke about in
on a rainy Sunday afternoon.
Eric
Robert Myers
> Robert Myers wrote:
>
> > The DoE did not, so far as I know (and everything that I know is based
> > on unclassified information), contemplate the kind of problem that the
> > aging W-76 presents when it planned its final tests. The fact that
> > the DoE now believes that it has to do such gigantic (and presumably
> > fine-scale) simulations should offer a clue as to what Dick Morse et
> > al became worried about.
>
> I don't believe the simulations have much to do
> with verifying the current stockpile.
>
<snip>
> News Analysis: Officials Air Views on Key Stockpile Issue
> http://www.armscontrol.org/act/2009_11/NewsAnalysis
"Congress in October included $223 million for the LEP in the fiscal
year 2010 energy and water development appropriations act. All of the
money is to go to refurbishment, but not replacement, of the W76
warhead, which is used on Trident submarines."
The case I built for my original contention was strong, and your
article simply confirms what I had surmised. The first Blue Gene
procurement by LLNL was a hurry up purchase, not a part of some long
term strategy that would include a loony idea like relying on a new
and untested design.
The picture that Del offered relates to an issue that would be likely
to be of special interest for the W76 design. Careful re-reading of
the New York Times article I offered would be more productive than
further speculative posting. The language of the article you cited
is, at best, misleading--probably for good reason.
Barring so many changes in the politics that I can't even imagine
them, nuclear bunker busters are a dead issue, unless there is
something *very* black underway. There will always be PhD physicists
at LANL/LLNL dreaming of reliving the go-go years of the forties and
fifties, and they will always be floating blue-sky proposals. There
may even be pocket change to keep a few lights on in the offices.
That doesn't mean that anyone is now or is ever likely to be serious,
barring a change in the strategic/political/military environment that
no one here could possibly predict.
Robert.
EricP
> On Nov 30, 2:06 am, EricP <ThatWouldBeTell...@thevillage.com> wrote:
>> Robert Myers wrote:
>
>>> The DoE did not, so far as I know (and everything that I know is based
>>> on unclassified information), contemplate the kind of problem that the
>>> aging W-76 presents when it planned its final tests. The fact that
>>> the DoE now believes that it has to do such gigantic (and presumably
>>> fine-scale) simulations should offer a clue as to what Dick Morse et
>>> al became worried about.
>> I don't believe the simulations have much to do
>> with verifying the current stockpile.
>>
>
> <snip>
>
>> News Analysis: Officials Air Views on Key Stockpile Issue
>> http://www.armscontrol.org/act/2009_11/NewsAnalysis
>
> "Congress in October included $223 million for the LEP in the fiscal
> year 2010 energy and water development appropriations act. All of the
> money is to go to refurbishment, but not replacement, of the W76
> warhead, which is used on Trident submarines."
It depends what 'refurbishment' means.
The 'The Rise of U.S. Nuclear Primacy' article states on page 4 that
"The navy, for example, is upgrading the fuse on the W-76 nuclear
warhead". Of course it could be anything really.
> The case I built for my original contention was strong, and your
> article simply confirms what I had surmised. The first Blue Gene
> procurement by LLNL was a hurry up purchase, not a part of some long
> term strategy that would include a loony idea like relying on a new
> and untested design.
>
> The picture that Del offered relates to an issue that would be likely
> to be of special interest for the W76 design. Careful re-reading of
> the New York Times article I offered would be more productive than
> further speculative posting. The language of the article you cited
> is, at best, misleading--probably for good reason.
I reread the 2009 'News Analysis' article and see nothing
overtly misleading. Perhaps you can elaborate.
The NYTimes article is from April, 2005.
It is difficult in 2009 to recover the context of what
else was going on back then, but I recall there were
serious warnings from reputable sources (Arlen Specter
and Brent Scowcroft names sprint to mind, but I may be mistaken)
that neo-cons (Richard Perle, et al) were pushing for new nuke
weapons development and abrogation of the Nuclear Test Ban Treaty
(based on it having been signed with the now non-existent
Soviet Union). The 'New Nuclear Policies' article is undated
but appears to be from about 2004 and supports that view.
There was also a certain amount of worry that neo-cons
were pushing for America to push for nuclear primacy
and a clear first strike capability while Russia was weak.
The 2006 Foreign Affairs article shows that many at
that time were mindfully of this point of view.
Within that context one can view the NYTimes article as truthful
but also as a plant to serve the Bush admin purpose - justify the
development of new nukes and build pubic support for
abrogation of NTBT, and their subsequent testing.
> Barring so many changes in the politics that I can't even imagine
> them, nuclear bunker busters are a dead issue, unless there is
> something *very* black underway. There will always be PhD physicists
> at LANL/LLNL dreaming of reliving the go-go years of the forties and
> fifties, and they will always be floating blue-sky proposals. There
> may even be pocket change to keep a few lights on in the offices.
> That doesn't mean that anyone is now or is ever likely to be serious,
> barring a change in the strategic/political/military environment that
> no one here could possibly predict.
The 2009 'News Analysis' article shows the Obama admin completely
rejects the idea of abrogating the Nuke Test Ban Treaty.
The current view that simulation is sufficient flows from
the new policy and not any technical change.
It is difficult to say whether the non-nuke testing regimes
(the Z-pinch tests for example) are sufficient to verify
the simulations or are self delusional.
Perhaps having the weapons owners doubting whether they
will actually work is a good thing.
Eric
Robert Myers
> Robert Myers wrote:
> > The case I built for my original contention was strong, and your
> > article simply confirms what I had surmised. The first Blue Gene
> > procurement by LLNL was a hurry up purchase, not a part of some long
> > term strategy that would include a loony idea like relying on a new
> > and untested design.
>
> > The picture that Del offered relates to an issue that would be likely
> > to be of special interest for the W76 design. Careful re-reading of
> > the New York Times article I offered would be more productive than
> > further speculative posting. The language of the article you cited
> > is, at best, misleading--probably for good reason.
>
> I reread the 2009 'News Analysis' article and see nothing
> overtly misleading. Perhaps you can elaborate.
The issue that is specific to the W76, talked about in surprising
detail in the New York Times article, isn't mentioned at all in the
2009 'News Analysis' article. That may be because the problem
peculiar to the W76 out to be a total non-issue, or it could be
because it really was a problem and no one particularly cares to
highlight the fact. One way or another, the near-term money is being
spent on the W76.
The 2009 'News Analysis' article appears to offer reassurances about
parts of the warhead that never were identified as likely to be
problematical. I would characterize offering such reassurances as
being misleading, at best. Whether by accident or by intent, the
'News Analysis' you offered did not lay to rest the potential W76
problem that *had* been identified.
No one sane is going to rely on an untested warhead design. "RRW
Redux" is pure sensationalism.
Robert.
EricP
> On Nov 30, 2:01 pm, EricP <ThatWouldBeTell...@thevillage.com> wrote:
>> Robert Myers wrote:
>
>
>>> The case I built for my original contention was strong, and your
>>> article simply confirms what I had surmised. The first Blue Gene
>>> procurement by LLNL was a hurry up purchase, not a part of some long
>>> term strategy that would include a loony idea like relying on a new
>>> and untested design.
>>> The picture that Del offered relates to an issue that would be likely
>>> to be of special interest for the W76 design. Careful re-reading of
>>> the New York Times article I offered would be more productive than
>>> further speculative posting. The language of the article you cited
>>> is, at best, misleading--probably for good reason.
>> I reread the 2009 'News Analysis' article and see nothing
>> overtly misleading. Perhaps you can elaborate.
>
> The issue that is specific to the W76, talked about in surprising
> detail in the New York Times article,
Yeah, surprisingly candid.
> isn't mentioned at all in the
> 2009 'News Analysis' article. That may be because the problem
> peculiar to the W76 out to be a total non-issue, or it could be
> because it really was a problem and no one particularly cares to
> highlight the fact. One way or another, the near-term money is being
> spent on the W76.
Its sounds like it might be a real issue,
and that is what RRW would address.
The near term money is being spent as they originally intended
for W-76 refurbishment, to address things like new fusing, internal
rusting, material degradation, corrosion, decay and the embrittling
of critical parts.
http://www.globalsecurity.org/wmd/systems/w76.htm
> The 2009 'News Analysis' article appears to offer reassurances about
> parts of the warhead that never were identified as likely to be
> problematical. I would characterize offering such reassurances as
> being misleading, at best. Whether by accident or by intent, the
> 'News Analysis' you offered did not lay to rest the potential W76
> problem that *had* been identified.
Its not my article so I'm not going justify its content.
However I think you are being a bit overly critical.
The issues they mention clearly were maintenance problems,
just they were not mentioned in the NYTimes article.
The refurbishing issues were being dealt with as quite
separate issues from redesign which RRW deals with.
Anyway, my point was not that simulations don't need to be backed
up by testing, just that you should also be mindful of the
powerful political forces also at work who would love to
use the W-76 flaw to justify new development and testing
even if alternatives were possible.
For example, I found this on an MIT website
http://web.mit.edu/stgs/nuclearweapons.html
"Reliability of the US Nuclear Deterrent:
Using the correlation between publicly announced tests and the known
periods of warhead development, it is estimated that there have been at
least eight full-yield detonations of the W-76 warhead during and after
development. From this, simple statistical laws set strict limits on
the just how many duds there can be in the stockpile. Using very
conservative numbers it can be said that, at the very least, 70 percent
of the W-76s would detonate as planned. If, as many analysts believe,
there are two or more warheads are already allotted for each target,
then more than 90 percent will be destroyed. This is a worst-case
scenario and the true fraction of weapons that would perform as planned
is most likely considerably greater than 70 percent."
The above analysis is naive because it ignores fratricide.
If the warheads arrive close together in time, the first can destroy
the second. If the first warhead is a dud but is big enough to take
out the second then the above stats don't apply.
If you delay the warhead arrivals and the first is a dud,
then for a period of time you have an intact adversary
who is really pissed off.
Even so, multiple warheads are possible.
But note also that Bush admin basically tore up the Outer Space Treaty
which bars weapons in space. On 31-Aug-2006 they published their
National Space Policy which essentially said they will do what
ever they want (and fuck you if you don't like it).
A summary of difference between Clinton and Bush43
Space Security Program
http://www.stimson.org/?SN=WS200610101122
> No one sane is going to rely on an untested warhead design. "RRW
> Redux" is pure sensationalism.
>
> Robert.
Then there seems to be few options:
(a) use current design with multiple warheads per target
(means they need to keep more active warheads)
(b) abrogate Comprehensive Test Ban Treaty
(c) abrogate Outer Space Treaty
(d) test on the dark side of the moon, hope no one catches you
Eric
Ken Hagan
<ThatWould...@thevillage.com> wrote:
> For example, I found this on an MIT website
> http://web.mit.edu/stgs/nuclearweapons.html
>
> "Reliability of the US Nuclear Deterrent:
> [...] From this, simple statistical laws set strict limits on
> the just how many duds there can be in the stockpile. Using very
> conservative numbers it can be said that, at the very least, 70 percent
> of the W-76s would detonate as planned. [...]"
I hate to nit-pick when we're already off-topic, but my understanding of
statistical laws is that the one thing they never do is set "strict"
limits and so you cannot say anything "at the very least". I hope that
wording came from a press officer rather than an adviser or policy maker.
nm...@cam.ac.uk
Oh, indeed! Statistical laws CAN set strict limits, but those are
strict in senses that are understood only by people with a good
grasp of probability theory.
And you are quite correct that the second sentence does not follow
from the first - quite the converse, in fact.
It's not as off-topic as all that, because exactly the same error
is made by the vast majority of 'computer scientists' when talking
about error handling. The few with Clue are like prophets crying
in the wilderness :-(
And supercomputers bring that issue into the foreground. The
correct question is not "Have they protected against failure X?"
but "What failure rate have they reduced X to?" Luckily - or maybe
by selection :-) - supercomputer architects are aware of that.
Regards,
Nick Maclaren.
Robert Myers
> On Tue, 01 Dec 2009 19:56:22 -0000, EricP
>
Leaving aside the abuse of language, the apparent abuse of logic and
statistics is arresting. Even if the claim of eight tests is
accurate, one knows nothing about the outcome of those eight tests.
We have to take it on faith that the warhead proceeded to production
based on adequate certainty arising from those eight tests, but one
does not have to be a kook to be skeptical. The pressure to keep up
in the arms race (and the propaganda war) was intense.
My claim is that the current emphasis on "mine is bigger than
yours" (as measured by linpack flops) is simply a continuation of the
propaganda war by other means. It isn't *entirely* crazy to believe
that the nation with the overwhelming advantage in computation also
has the overwhelming likely advantage in any future strategic
confrontation, but it is still a very big step. If those are the
terms on which the propaganda war is to be fought, then it is
important to keep the Top-500 list and the place of the US weapons
labs on those lists constantly in the public eye.
It also means that the US (along with its enormous propaganda
machinery) has a vested interest in not pursuing the line of thinking
that bigger (by some linear measure) isn't necessarily better.
Robert.
EricP
(Yeah it's tangentially off topic, but not terribly more so than
other past threads like climate change. And since the other msg
traffic is Skybuck, I figured at least this is not brain damaged.)
I wondered about those stats myself, but gave him the benefit of
the doubt as he appeared to not be an internet kook from his
publication resume on his web page and being part of a work group.
I suspect this was covered in more detail in his article in
Jane's Intelligence Review, July 2005.
Unfortunately the article is not online, just the abstract:
"US nuclear deterrent is secure despite doubts cast on warhead
Article Abstract:
A debate within the US weapons laboratories over the reliability of the
W-76 Trident warhead, which was giving problems, became public in April
2005 after a New York Times article reported that many nuclear weapons
experts believed up to three-quarters of the US nuclear force could be
defective. The test statistics of the weapons and the implications of
the possible defects are examined."
I'm guessing it was based on Statistical Process Control stats
on the odds of sampling and a destructive test on 8 good
warheads in a row at the very start of a production run.
But it has been so long since I worked with SPC that I've
forgotten it all.
Eric