Reliability and Probabilistic Risk Assessment and Regulatory Policies and Practices

                       UNITED STATES OF AMERICA
                     NUCLEAR REGULATORY COMMISSION
               ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
                                  ***
                MEETING:  RELIABILITY AND PROBABILISTIC
                            RISK ASSESSMENT
                                  ***
                        U.S. Nuclear Regulatory Commission
                        11545 Rockville Pike
                                           Room T-2B3
                        Rockville, Maryland
                        Monday, January 25, 1999
         The subcommittee met, pursuant to notice, at 1:02 p.m.  
     MEMBERS PRESENT:
         GEORGE APOSTOLAKIS, Chairman, ACRS
         JOHN BARTON, Member, ACRS
         MARIO FONTANA, Member, ACRS
         THOMAS KRESS, Member, ACRS
         DON MILLER, Member, ACRS
         DANA POWERS, Member, ACRS
         ROBERT SEALE, Member, ACRS
         WILLIAM SHACK, Member, ACRS
         GRAHAM WALLIS, Member, ACRS.                         P R O C E E D I N G S
                                                      [1:02 p.m.]
         DR. APOSTOLAKIS:  The meeting will now come to order.  This
     is a meeting of the ACRS Subcommittee on Reliability and Probabilistic
     Risk Assessment.  I am George Apostolakis, Chairman of the Subcommittee.
         ACRS members in attendance are John Barton -- will be in a
     few minutes -- Mario Fontana, Tom Kress, Don Miller will join us soon,
     Dana Powers, Robert Seale, William Shack, and Graham Wallis.
         We also expect Mario Bonaca, who has been selected as a new
     member.  He is expected to be appointed to the ACRS in the near future.
         ACRS invited experts are Mohsen Khatib-Rahbar and Jeff
     Kaiser.
         The purpose of this meeting is to discuss the possible use
     of frequency-consequence curves in risk-informed decision-making.
         The Subcommittee will not review proposed options to make 20
     CFR 50.59, as was previously announced in the Federal Register.  On
     December 23, 1998, the Subcommittees -- I guess there are two
     subcommittees --
         DR. KRESS:  Singular.
         DR. APOSTOLAKIS:  -- singular -- the Subcommittee will
     gather information, analyze relevant issues and facts and formulate
     proposed positions and actions as appropriate for deliberation by the
     full committee.
         Michael T. Markley is the Cognizant ACRS Staff Engineer for
     this meeting.
         The rules for participation in today's meeting have been
     announced as part of the notice of this meeting previously published in
     the Federal Register on December 30th, 1998.
         A transcript of the meeting is being kept and will be made
     available as stated in the Federal Register notice.  It is requested
     that speakers first identify themselves and speak with sufficient
     clarity and volume so that they can be readily heard.
         We have received no written comments or requests for time to
     make oral statements from members of the public.
         This issue of using the frequency-consequence curves at the
     release level was raised in an ACRS letter -- actually an attachment to
     the letter -- last July in the context of making 10 CFR 50.59
     risk-informed.  Since then we have had several discussions among
     ourselves -- I mean the members -- regarding the usefulness of such an
     approach.  There have been several concerns voiced.  We have also had
     some additional work done by Dr. Kress, so today we will have another
     opportunity to discuss the possibility of using these curves in a real
     way and see how the uncertainties that are involved can be handled.
         In the process of learning more about these curves, I
     contacted a few friends in Europe and I received information from the
     Netherlands, how they are doing their business, and they are not using
     these curves, but I was surprised to receive a paper from Switzerland,
     where they actually proposed to use frequency-consequence curves where
     the consequences, the equivalent grams of cesium released, and I read
     the paper with great interest and I noticed that it was co-authored by
     several colleagues in Switzerland, the regulatory body there, and
     several colleagues in the United States.
         In fact, these colleagues are in Rockville, so I thought it
     would be a good idea to invite the lead man from that company, Dr.
     Mohsen Khatib-Rahbar to come as an invited expert and talk to us about
     that approach and maybe address some of the concerns that the members
     have, so I am very pleased that Mohsen agreed to come.
         Usually the members desire to know what the qualifications
     are of the speakers, so Mohsen has degrees in Chemical Engineering and
     Nuclear Engineering from Cornell, a Ph.D. --
         DR. KRESS:  We all know him.
         DR. APOSTOLAKIS:  Not all.
         He worked at Brookhaven National Laboratory for about 10
     years, where he in charge of the severe accident analysis group, doing a
     lot of work for the Nuclear Regulatory Commission of course.  He spent a
     year then at the NRC as a Visiting Scientist, and then since 1990 he has
     been the President of Energy Research, Incorporated, here in Rockville,
     and he has been very active -- well, his company and he himself have
     been very active supporting -- providing technical support to regulatory
     agencies around the world including our own here, in Spain, Sweden, and
     Switzerland.
         So we are very happy that Mohsen could find the time and
     join us and enlighten us -- we really appreciate it, Mohsen.
         DR. KHATIB-RAHBAR:  Thank you.
         DR. APOSTOLAKIS:  We are also pleased that Mr. Murphy has
     agreed to come and Mr. Holahan.  This is just to exchange ideas, as I
     said.  We are not really going to make any decisions, although the
     Subcommittees of course usually do not -- not usually -- all the time do
     not make decisions, but this is really for us to understand what is
     going on about these curves.
         Yes, Dana?
         DR. POWERS:  The Subcommittee does need to make a decision
     on whether to bring this discussion forward to the full Committee or not
     either now or in the future.
         DR. APOSTOLAKIS:  Okay.  That is a good point.  Yes.
         Oh -- Mike actually put some comments here for me.  Okay. 
     We will start with Dr. Kress, who will give us some of his recent
     thoughts on the issue and then we will move on to Dr. Khatib-Rahbar.
         DR. KRESS:  Thank you, Mr. Chairman.  I hope no one notices
     my tie.
         [Laughter.]
         DR. POWERS:  Anyone that doesn't notice his tie is qualified
     under the Americans for Disabilities.
         [Laughter.]
         DR. SEALE:  I will repeat -- at least the stripes are
     running in the right direction.
         DR. KRESS:  With that out of the way, the subject I want to
     talk about is of course the proposed role of FC curves, as we call them,
     in a risk-informed regulatory system.
         I had put together what we call a white paper for the
     committee's use, just to educate ourselves on these.  I am not really
     going to talk about that white paper.  It is available if you want to
     read it.  What I am going to do is use the concepts in the white paper
     to take it a step further about, okay, given that, well how do you use
     these things and are they useful at all.
         If we are talking about their use in a risk-informed world,
     risk-informed regulatory system, you really ought to start from the
     top-down and look at what is needed for a risk-informed regulatory
     system, and what you have to have is regulatory objectives, and of
     course the major regulatory objective you might say is protect the
     health and safety of the public.  That's not good enough because you
     have to put that in terms that are useful as measurable metrics, and if
     you have such terms of measurable metrics you also have to have
     something entirely different to those and that is acceptance criteria,
     acceptance values for those metrics, so that is major attributes you
     have to have to really have a risk-informed system.
         As an example, I mentioned that in 1.174 we had the metrics
     of CDF and LERF and plant-specific values were to be provided and there
     were definitely acceptance values developed for these and this is the
     kind of regulatory objective I am talking about and the kind of
     acceptance valves I am talking about, except -- and these were good. 
     Those were very rational, good objectives.
         I personally do not think that CDF and LERF represent the
     only regulatory objectives and of course that was recognized in 1.174. 
     Also, because they asked for a lot of other things --
         DR. POWERS:  Dr. Kress, isn't this a situation -- the reason
     we look at other metrics that we have continued to chafe under the
     burden of using CDF and LERF rather than actual risk?
         DR. KRESS:  No, I don't think that's particularly -- at
     least that's not my motivation.  I actually think LERF is a very good
     connector to risk, and so that's not what bothers me.  I just think
     there are other objectives.  LERF to me is a surrogate, a pretty good
     surrogate for the prompt fatality risk -- individual risk for prompt
     fatalities.
         I just think there are other objectives that we need to deal
     with other than that one, and that that does not necessarily -- it may
     but it doesn't necessarily supersede all of the others.
         DR. POWERS:  Well, I think that maybe we are saying the same
     thing.  Arguably LERF relates closely to prompt fatality risk.
         DR. KRESS:  Very closely, yes.
         DR. POWERS:  But we recognize there are other measures that
     if you did a Level 3 PRA properly you would come forward with the
     latents, you would come forward with the property contamination, you
     would come forward with the injuries.  You would come forward with a lot
     of other measures and that we see these deficiencies in either the
     measures CDF or LERF, especially if you start thinking about doing cost
     benefit analyses for regulation.  You are not getting all the costs.
         DR. KRESS:  I think you are hitting on it.  In fact, I made
     a little attempt to list what some of these other things might be that
     we are really as an agency concerned with, and a number of these are
     real risk parameters and I may not have all of them either.  This was
     just an example of additional objectives that we may be interested in.
         Now we are interested in limiting certain risk parameters. 
     We are also interested in doing it in a particular way, and that is the
     defense-in-depth philosophy and I have also lumped the CDF under that
     because it is not a real risk metric, but it could be put up here I
     think, but the idea is in a risk-informed world there are objectives
     that you want to have acceptance criteria.  There are ways you want to
     achieve these objectives, and I list these safeguards separately. 
     Actually, it is just another set of sequences really.  They could be up
     here, but the reason I listed them separately is I think you would, when
     you approach safeguards, you do it differently.  You want to lean real
     heavily on the prevention end, as opposed to mitigation or --
         DR. POWERS:  Well, I think we lean almost exclusively on
     that --
         DR. KRESS:  Yes, so I have listed it separately but it is
     really just another set of sequences.
         DR. POWERS:  Right now it is done in the relatively
     primitive way --
         DR. KRESS:  Absolutely.
         DR. POWERS:  -- of a design basis threat and being able to
     counter that threat.
         DR. KRESS:  Absolutely.
         DR. POWERS:  And it is an area probably ripe for rethinking. 
     It seems to me that this is an inventory that when you are thinking
     about cost benefit analyses -- in a grander scheme, not in the
     regulatory scheme -- but in the grander scheme of things that you want
     all these things in addition.
         I still think the experience of the Chernobyl accident is
     telling us that injuries --
         DR. KRESS:  Injuries and --
         DR. POWERS:  -- are the various -- society costs that is
     non-negligible.
         DR. KRESS:  And I do have injuries on here, and I would add
     another one -- total number of injuries, you know, as well as individual
     risk and injury, so there may be other objectives we want to add because
     I didn't mean for this to necessarily be a comprehensive complete list,
     and in arriving at a risk-informed regulatory system there are also
     other considerations that are outside of -- different than these, things
     like caps on instantaneous CDF and things like time left on the license
     ought to be a consideration.  It's probably risk equation and I just
     listed a couple of those as examples and my point with this is you have
     to deal with all these in a risk-informed world in some way.
         Now my talk today about FC curves is not going to tell how
     to work these others in.  I am going to deal with risk parameters and
     acceptance criterias on them.
         DR. WALLIS:  Do these have to be just radiation-related?
         DR. KRESS:  Yes.
         DR. WALLIS:  I think one of the biggest problems with
     Shoreham was the risk that the public imagined or appreciated of trying
     to evacuate Long Island.
         DR. KRESS:  I think you could add things like that into it. 
     It's going to be difficult --
         DR. WALLIS:  That might be far more risky than the actual
     radiation risk.
         DR. KRESS:  Yes.  Yes, I think you could add things like
     that into it.  I am trying to keep it to pure risk numbers though right
     now.  That's so we can deal with them.
         The issue then I am dealing with today is how to incorporate
     the risk objectives into a risk-informed regulatory system.
         In the first place, we do have to have acceptance criteria,
     and that is a subject all its own, I think.  For the moment I am not
     going to say how to arrive at acceptance criteria.  I just want to say
     we've got to have them.  We have to have them before we start a
     risk-informed system and they need to be there and we have to figure out
     a way to get them.
         Then we would have to determine whether any particular plant
     status with respect to these acceptance -- what it is with respect to
     these.  You have to have a way to do that.  Well, clearly these risk
     objectives I list will come straight out of a PRA.  I mean those are
     things you get right out of a PRA, a Level 3 plant-site specific PRA.
         Well, why not just use the Level 3 plant site-specific PRA. 
     That's after all a perfectly rational way to go, and that is what I see
     proposed by some people like Bob Christie.
         Well, there are lots of reasons that I think that is not a
     very practical, viable way to go, and in the case of 1.174 they came to
     the same conclusion, that the full Level 3 PRA was just too uncertainly,
     too unwieldy, too -- just too dislocated from the things you are
     interested in in terms of the design of the system.
         DR. WALLIS:  I don't think that is true at all.  I think
     real public risk is in the Level 3 PRA.
         DR. KRESS:  Oh, there is no doubt about that.
         DR. WALLIS:  The uncertainties don't go away by saying we
     are going to use the surrogate.
         DR. KRESS:  I understand that and I am going to address that
     issue.  I think it does de-focus though, and that is the problem I have
     with it.
         So what they did in 1.174 is develop a practical,
     utilitarian, convenient surrogate for the prompt fatality.  Now it is a
     surrogate and I thoroughly believe that, and what I am saying now is
     that was for the individual risk of prompt fatalities.  We also had it
     for CDF.  What I saying is for these other risk objectives we are
     interested in preserving in a risk-informed regulatory system there is a
     need to have a consistent way to develop such convenient lower tier
     surrogates for the other regulatory objectives.
         That is what we are about today.  That is what FC curves are
     around.  We are looking for lower tier surrogates for these risk
     objectives so we don't have to use a full Level 3 PRA.
         DR. WALLIS:  You used the word -- I'm sorry to interrupt,
     Tom -- convenient.  Do you mean convenient for the Agency?
         DR. KRESS:  Convenient for the Agency --
         DR. WALLIS:  Or convenient for the people who are subject to
     risk?
         DR. KRESS:  No, convenient for the Agency -- well,
     supposedly we are going to take care of their interests.  If we don't
     take care of that interest we are not doing our job.
         DR. WALLIS:  But if your criterion is convenience for the
     Agency, that is not really the right criterion.
         DR. KRESS:  But convenient -- but not at the expense of
     giving us the risk metric.  We don't want to do that and in fact
     clearly -- clearly -- these regulatory objectives I listed do depend on
     these things.  They depend on the amount of release of different fission
     products and the timing, but they depend on the site specific parameter. 
     They are in there.  They are not -- you cannot do away with those.  They
     are part of the equation and they have to be preserved in this system
     using FC curves some way.  They have to be in there.  That is what I am
     saying.
         DR. WALLIS:  But in picking regulatory convenience as a
     criterion, you may do things that are very inconvenient for these
     populations --
         DR. KRESS:  Well, regulatory convenience may have been the
     wrong word.  I am saying something a little bit more straightforward and
     easier to do than a full Level 3 PRA, but still incorporate those risk
     parameters some way.
         DR. SHACK:  It's to make it implementable.
         DR. KRESS:  Make it implementable -- that's a better word.
         MR. HOLAHAN:  Dr. Kress?  This is Gary Holahan of the Staff.
         I wonder if an alternative to developing new surrogates
     would be to test whether the design basis requirements or a revised
     version of the design basis requirements aren't already good surrogates
     for these other objectives.
         DR. KRESS:  That could be a possibility.  I think you may
     have difficulty doing that, because the way you do that is you do a PRA
     at the level 3 and see if the resulting plant configuration that you got
     from the design basis accident actually meets the criteria.
         Now they probably do, and you can probably use that as a
     basis to say our design-basis concept hasn't been all that bad.  It's
     not -- there's no -- I don't think there's a direct way to do it,
     though.  I think it's an indirect.  But it's a thought.
         But the question then before us is how do I incorporate
     these things that are part of the risk equation and perhaps use F-C
     curves as surrogates for all of the risk objectives.  That's the subject
     I'm talking about.
         Well, I'd like to turn -- when I get to that, I'd like to
     turn to some work done by Jeff Kaiser.  Sorry he's not here, because he
     could answer the questions I have about these.  These are just a couple
     of examples of some of his work.  And what these are, and I don't know
     if you can see them or not, because they're copies of copies of copies,
     what Jeff did was basically took as an independent variable the release
     fraction or a release of fission products or a surrogate for them and
     varied -- he went to a great number of actual sites of actual plants. 
     He used the site-specific populations, meteorology, et cetera.  And he
     varied the emergency response measures over -- from bad emergency
     response like no emergency response to reasonable to really good, to see
     what the effect of the source term, which is fission product release,
     has on these risk metrics.  It's a conditional risk metric.  This is the
     prompt fatality one, and this is the latent fatality, individual risk.
         And what you see -- ignore the lines for a minute.  Those
     are not Jeff's slides.  I put these on there.  But what you see in the
     data is it scatters all over the place from site to site and depending
     on the emergency response measures you have.  A big scatter.  Look at
     those kind of things, especially up here where the release fraction is
     pretty high.  Now that's --
         DR. SHACK:  Now what did he -- he went to a specific site
     and we see the vertical distribution of dots as he varies the emergency
     response?
         DR. KRESS:  Pretty much.  Pretty much it.  And of course if
     you had a constant kind of emergency response, you'd still get a lot of
     scatter just from site to site.  But this is why Dana and I have said
     that using level 3 and atmospheric transport with the big uncertainties
     -- this by the way is not an uncertainty analysis exactly, it's a
     sensitivity analysis, and certainly would even make this worse.
         DR. APOSTOLAKIS:  Tom, I think we should be very clear as to
     what is variable here.  Is there plant-to-plant variability here?
         DR. KRESS:  This is different sites --
         DR. APOSTOLAKIS:  Different sites.  Okay.
         DR. KRESS:  And it's the effect of different emergency
     response measures.  There is no plant except the site in here.
         DR. APOSTOLAKIS:  Right.
         DR. KRESS:  Fission-product release is the independent
     variable.  It has nothing -- so, you know --
         DR. APOSTOLAKIS:  Okay.
         DR. KRESS:  If -- so --
         DR. APOSTOLAKIS:  So each vertical line represents a site,
     the points on a line.
         DR. KRESS:  Probably; yes.  That's one of the things I was
     going to ask Jeff.
         DR. APOSTOLAKIS:  Okay.  So the different vertical lines or
     imaginary lines indicate different sites.
         DR. KRESS:  Yes.
         DR. APOSTOLAKIS:  And then along the vertical axis he just
     varied arbitrarily --
         DR. KRESS:  No, no.  This axis is an arbitrary, it's the
     independent --
         DR. APOSTOLAKIS:  Yes, but I mean, if I pick one value say
     for the fraction of release.
         DR. KRESS:  Yes.
         DR. APOSTOLAKIS:  The bunch of points --
         DR. KRESS:  Represent different sites and different
     emergency response measures.
         DR. APOSTOLAKIS:  And different emergency -- and different
     sites.
         DR. KRESS:  Yes.  Yes.  If that's --
         DR. APOSTOLAKIS:  So it's not one site.
         DR. KRESS:  If that site had this release --
         DR. APOSTOLAKIS:  Yes.
         DR. KRESS:  Then you would get this with different emergency
     response measures.
         DR. APOSTOLAKIS:  Okay.  So --
         DR. KRESS:  So vertical is both of them, both site
     parameters and emergency response parameters, which are on the vertical
     line.
         DR. APOSTOLAKIS:  Okay.  Okay.  Okay.
         DR. WALLIS:  Without looking at the likelihood of the
     various responses this isn't very useful, because the values go from
     zero to 1 essentially, except 1 is off the --
         DR. KRESS:  That's right.
         DR. WALLIS:  But if the values go from zero to 1, anything
     is possible, and you haven't told me anything.
         DR. KRESS:  Yes.  This is the nature of the beast, is what I
     wanted to get across.  But what I'm -- if, however, one says all right,
     let's use best-estimate reasonable emergency response measures, then
     this scatter collapses to site-to-site variations, which is not very
     big, not as big as this.  And in fact I tried to draw a line.  Now this
     is not a regression line, it's an eyeball regression.  But that would
     kind of represent reasonable emergency response measures.
         Now you can't see the scatter of the site variation on there
     because there's no way for me to do that, but we found out with the LERF
     business in 1.174 that the variation in LERF is only about a factor of 4
     for all the sites.  And basically this is kind of what they did in
     1.174.  They took a reasonable set of emergency response for all the
     sites and come up with an acceptable LERF value to meet the safety goal. 
     And this is kind of what they did.  So what this line represents then is
     a correlation which is a kind of best-estimate correlation between
     fission-product release and the regulatory objective safety goal, risk
     metric.
         Now -- and it come right of a exercise of an atmospheric
     dispersion consequence code.  You didn't have to have a PRA or anything
     for that.  It's an atmospheric consequence dispersion code.  And what
     I'm saying is for these risk objectives that I list that are in addition
     -- are LERF plus other things, that there is such a correlation for each
     one of them that can be gotten in exactly the same way.
         Now you could adjust your assumptions on emergency response
     to whatever you think you want it to be in that particular correlation. 
     You can actually adjust that.  But I am saying that you can develop with
     such a code a correlation, a preexisting correlation for all the sites,
     for each of the risk metrics.
         Now --
         DR. WALLIS:  Tom, could you explain something about
     emergency response to me?  Is this the responsibility of the State?
         DR. KRESS:  Yes.
         DR. WALLIS:  Because you might conclude from what you've
     told us that one should regulate emergency response very carefully.
         DR. KRESS:  You might, you might.  But anyway my point is
     there is a functional correlation between fission product release and
     the risk metrics, and --
         DR. FONTANA:  When you choose -- Tom, when you choose a
     particular value at a fission product release, you're also choosing a
     particular value of the timing of release, and the only variable is
     differences in emergency response?  Is that correct?
         DR. KRESS:  That's part of the timing.  That timing is built
     into the emergency response in this case.
         DR. FONTANA:  Oh, it is?
         DR. KRESS:  Yes.  It's lumped into the emergency response in
     this case.  So if indeed I'm correct that there is a functional
     correlation between fission product release, and I'm absolutely sure
     there is, that one can use and predetermine ahead of time for all U.S.
     sites, then the frequency consequence curves are useful, and they are
     useful in this way.
         I showed in the white paper -- now there may be some
     discrepancies in how I've used CDF here, so kind of take that with a
     little grain of salt.  But the integral part is correct.  If you want to
     look at what I call a risk consequence metric, that's those lists of
     objectives I had, and --
         And find out how one calculates it from an FC curve, but it
     is an interval of the full FC curve.  And here I have put it in terms of
     release fraction.  Of course, you don't want a release fraction because
     that doesn't give you the power of the reactor as a function, you would
     want actual curies, but I just did it in release fraction because -- for
     convenience in this case.
         But what you see is it does have the derivative of the
     frequency consequence curve in it.  It does have this functional
     correlation in it for that particular consequence, so you have to -- it
     has that in it, and it has CDF in it some way, and it may be -- it may
     be built into the derivative or it may be separate, I am not certain
     now, but it is in there.
         So, what I am saying is, if you are applied with appropriate
     FC curves that come straight out of the PRA, it is a level 2 PRA, and
     have developed ahead of time these appropriate functional relationships,
     then it is an easy task to develop software that will digitize this.  In
     fact, Bill Shack sent me some digitized values of FC curves.  You can
     digitize them, the slope of the FC curve, and numerically integrate and
     directly determine from the FC curve that way the status with respect to
     that risk objective.
         In my mind, this is the way you have to use FC curves.  So I
     have a bit of a difficulty with I think Mohsen's talk where he is going
     to come up with actual lines for the FC curve.  I think you have to do
     it -- you look at the curve and it is the whole curve that matters, not
     just different parts of it.
         DR. WALLIS:  Tom, this is an integral.
         DR. KRESS:  It is an integral.
         DR. WALLIS:  So it is very forgiving about the shape of the
     curve in terms of the result.
         DR. KRESS:  That's correct.
         DR. WALLIS:  It may well be that if you know CDF from LERF,
     that different curves that look different only change this number by a
     factor of 2.
         DR. KRESS:  Could be.  I am not sure yet.  But I am talking
     now technical rigor.  And there may be other ways to get around this, I
     am sure, but I want to be sure we know what we are doing.  So the
     summary of my talk is that I don't think CDF and LERF are the complete
     regulatory objectives, and I suggested others that would be on the list.
         The first thing you have to do is develop acceptance
     criterias for these, and there is a way to do that, and that is the
     subject of another talk.  I didn't discuss that.  But you do have to
     have these acceptance criteria or else you don't have a way to operate. 
     And these acceptance criterias might have to incorporate the uncertainty
     some way.
         FC curves appear to me to be a way to provide a systematic,
     consistent, surrogate means for determining a plant's status with
     respect to all of these risk-related regulatory objectives, including
     LERF, and I shouldn't have put CDF on there, because they are clearly
     not CDF, but including LERF.
         DR. FONTANA:  Tom, why are you calling it a surrogate?
         DR. KRESS:  It is a surrogate.
         DR. FONTANA:  Because you can take your allowable --
     apparently one FC curve is an allowable criterion that you don't want to
     exceed.
         DR. KRESS:  There is no unique FC curve, though, that will
     meet your regulatory objectives.
         DR. FONTANA:  I know.
         DR. KRESS:  Okay.
         DR. FONTANA:  But, presumably, you have --
         DR. KRESS:  The surrogate is the integral.
         DR. FONTANA:  And then to calculate the FC curve for a
     particular -- you go to a PRA, level 3 PRA.
         DR. KRESS:  Level 2.
         DR. APOSTOLAKIS:  Two, level 2.
         DR. KRESS:  That's why I am avoiding level 3.  That's my
     whole objective is to get rid of level 3.
         DR. FONTANA:  Okay.  You have got the consequences in terms
     of --
         DR. KRESS:  This functional correlation which was developed
     from what would have been the back end of a level 3, but you do that
     ahead of time.
         DR. FONTANA:  It is necessary for --
         DR. KRESS:  Yeah, that's what makes it a surrogate.  Yeah. 
     You have to have that functional correlation.
         Okay.  And I think one can deal with some of these
     objectives such as total deaths, total injuries, you might better deal
     with those with site-specific regulations that are separate from
     acceptance criteria, I don't know.  My approach would be to look for
     ways to do that.  So that was just a side comment, that last bullet,
     just, you know, for what it is worth.
         That's what I wanted to say about FC curves.
         DR. WALLIS:  You split the site characteristics from the FC
     curves.  If you just said level 3 --
         DR. KRESS:  They are in there automatically.
         DR. WALLIS:  And so a reactor in New York City and one in
     Idaho would have completely different allowable FC curves but the same
     risk, perhaps, if you took the calculation into account.
         DR. KRESS:  You could always go back to level 3 and be
     site-specific and have different FC curves to meet the same objective,
     yes.
         DR. APOSTOLAKIS:  Thank you, Tom.  Yes.
         DR. KHATIB-RAHBAR:  Let me make a comment on what Tom has
     said.  First of all, if you are using early fatality as a surrogate, I
     totally agree with Tom.  If you are going to use latent fatalities, the
     site-specific differences make very little difference.
         DR. KRESS:  Yes, that's right.
         DR. KHATIB-RAHBAR:  What makes a difference is really in the
     early fatalities, where evacuation makes a difference and inhalation
     dose often dominates.
         DR. KRESS:  Absolutely.
         DR. KHATIB-RAHBAR:  When you go to latent fatalities, when
     you integrate it out to thousands of miles, the uncertainties arriving
     from, you know, using a straight time plume model, et cetera, --
         DR. KRESS:  And the same thing about total injuries.
         DR. KHATIB-RAHBAR:  Right.  Right.  I mean which are
     integrated in the facts when you are talking about the early fatalities.
         DR. KRESS:  Site variations really show up in prompt
     fatalities more than even anywhere --
         DR. KHATIB-RAHBAR:  Absolutely.  Absolutely.  Absolutely. 
     In fact, we did a study at Brookhaven about five, six years later after
     Jeff Kaiser's study, which is a nuclear engineering -- nuclear
     technology, where we demonstrated, considering all the uncertainties,
     the uncertainties in the weather, variability dominates.  That's why, as
     I will discuss later on, I hope that you can be convinced that you do
     not go to the level 3 PRA, which sort of supports what you are saying.
         DR. KRESS:  You really can bound it or make a best guess for
     it.
         DR. KHATIB-RAHBAR:  Absolutely.  Absolutely.
         DR. KRESS:  And what is what I am saying.
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. APOSTOLAKIS:  You mentioned a paper.  Would you send us
     a copy?
         DR. KHATIB-RAHBAR:  I will be happy, too.  Yes.
         DR. APOSTOLAKIS:  Thank you.  Okay.  Let's go on with
     Mohsen's presentation.  You have a comment, Joe?  Go ahead.
         MR. MURPHY:  Joe Murphy from the staff.  Let me just add one
     thought for the early fatalities.  What we found in 1150, that the
     number of early fatalities is almost entirely driven by the assumption
     about evacuation.  Basically, if you assume 100 percent of the
     population evacuates, you will have no early fatalities.  If you assume
     nobody does, you will have essentially the close-in population of one to
     two miles, you will have an extremely high percentage of that that are
     under the plume, will have early fatality.  So it is almost directly
     proportional to the assumption on an evacuation.
         DR. KRESS:  I agree with that, and I have an additional
     comment on that I would like to make.  And that is we are fond of saying
     emergency response is just a defense-in-depth concept, but when it is
     that important, and it is very important to the LERF value you end up
     with, to me, it is not really an emergency -- it is not really
     defense-in-depth, it is part, it is a strong part of your acceptance
     criteria.  It couldn't be characterized as defense-in-depth in that
     case.
         DR. APOSTOLAKIS:  Okay.  Let's go on with Mohsen's
     presentation and then I am sure we will have discussion, a free for all. 
     It is in the front there someplace.
         DR. KHATIB-RAHBAR:  Okay.  Good afternoon.  I guess, before
     I start, I would like to bring to your attention one important issue
     here.  As George indicated, my company works with a number of regulatory
     authorities, including the U.S. Nuclear Regulatory Commission,
     therefore, as I have told George, my presence here should not be
     construed in any way as me representing the U.S. NRC or any of the other
     sponsors in Europe, and, in particular, my views about the Swiss
     approach are solely my own and do not represent -- I am not a
     representative of the Swiss Nuclear Safety Inspectorate here today, I am
     representing myself.
         So, if there are any comments, any questions, they should be
     directed at myself rather than these organizations.
         DR. APOSTOLAKIS:  Even so, we thought it would be worthwhile
     to have you here.
         DR. KHATIB-RAHBAR:  I appreciate it.  I am happy to be here.
         Let me first give you an outline of what I will discuss
     here.  I guess this -- the precursor for this was a paper that
     Chakraborty sent to Dr. Apostolakis about a paper we gave at the IEA
     back in 1995 which had the proposed Swiss safety criteria in it.  And I
     guess for the first time we decided to use the frequency consequence
     curves in a manner which is a little different than what was used in
     NUREG-1150, or had been perceived a little bit by others.
         In the subsequent year, I received some communication from
     George that he needed some specific issues that he wanted to discuss at
     this meeting.  So I was away last week in Europe, so before I left on
     Friday the following -- previous week, I prepared this presentation,
     which really consists of two parts, George.
         The first part deals primarily with the issues that you had
     raised in your e-mail to me, and the second part deals with the Swiss
     approach.  So, depending on your interest, I can skip one or concentrate
     on the other.  I don't know how much time you want to devote to this. 
     But I have prepared something equivalent to about 45 minutes worth of --
         DR. APOSTOLAKIS:  I believe you should go ahead with the
     full presentation.  I appreciate that you -- the fact that you put extra
     effort to respond to my questions.
         DR. KHATIB-RAHBAR:  First of all, let me say -- the first
     question, I guess, or issue that, George, you raised, was -- What do
     these curves represent?  If you look at a typical frequency consequence
     curve, forget about the consequence measure, what it is, in this case it
     is cesium-137, but it doesn't matter.  These curves represent really
     several regions.  If you don't mind, I will stand because I can --
         DR. POWERS:  You need a microphone.
         DR. APOSTOLAKIS:  You need a mike.  Theron.
         DR. POWERS:  While you find a microphone --
         DR. KHATIB-RAHBAR:  I can stand closer to the microphone,
     that's okay.
         DR. POWERS:  It strikes me that it does matter that you have
     plotted 137 across the bottom because --
         DR. KHATIB-RAHBAR:  Yeah, we can address that, yeah.
         DR. POWERS:  -- when people --
         DR. APOSTOLAKIS:  Can we wait for just a second then, I am
     sorry, for him to put a mike on so the recorder can --
         DR. POWERS:  Well, I am just going to give a speech, I
     wasn't going to ask him a question.
         DR. APOSTOLAKIS:  Well, if he interrupted you, though, and
     the recorder wants to take everything down.
         DR. KHATIB-RAHBAR:  Okay.  I can sit here.  That's fine. 
     Don't worry.
         DR. APOSTOLAKIS:  Okay.  Go ahead now.  Yeah.
         DR. POWERS:  Well, I mean the fact is that if we look at a
     reasonable consequence code, it will typically include 66 isotopes, I
     think.
         DR. KHATIB-RAHBAR:  Fifty-four, yes.
         DR. POWERS:  Some, like that.
         DR. KHATIB-RAHBAR:  Right.
         DR. POWERS:  For some sites I think we have done 96 or
     something like that.  It strikes me that we have to then wrestle with
     the idea that we have not got one frequency consequence curve, or two,
     or three, but 54.
         DR. KHATIB-RAHBAR:  Yeah, I will address that issue a little
     bit later, if you don't mind.  Let me first address what I want to say,
     then I will come back to your point, if you don't mind.
         What I want to show here, this curve really represents what
     I call three regions.  The fat region, and the fairly sharp region here,
     and the intermediate region.  What does fat region represent is really
     that the -- most of the current lightwater reactors have a curve which
     look like this.  By and large, most of the severe accidents are
     mitigated by the existing design.  This is what they call the
     defense-in-depth region.  That means most accidents, most of the severe
     accidents are mitigated by the containment.  You are getting very, very
     low releases for most of the accidents.  I don't want to point this at
     you.
         MR. MARKLEY:  If you would like to stand.
         DR. KHATIB-RAHBAR:  This is too complicated.
         DR. APOSTOLAKIS:  We thought you could handle it, Mohsen.
         DR. KHATIB-RAHBAR:  You would be surprised, George.
         Okay.  And then in this region, the uncertainties on the
     releases are -- the consequences are enormous.  As you can see,
     essentially, if you look at this curve, this horizontal region shows,
     essentially, the width of the uncertainty band.  It goes over several
     decades.  You are talking about as much as six or seven decades.  But on
     the other hand, if you look at the upper bound releases in this region,
     given the uncertainties that exist, they are not that significant,
     consequentially.  Okay.
         However, in this region, this is where, even though the
     frequencies for these accidents in this region are very small, but the
     consequences are enormous.  So, in terms of a risk, even though the
     overall frequencies are low, you get oftentimes a large fraction of the
     total risk of severe accidents coming from this region.  These are
     primarily events which involve bypass of the containment, steam
     generator tube rupture events, event V, V sequences, et cetera.  Thank
     you.
         Now, this is what I call in this region, what I call the
     vulnerable region, the containment vulnerable region.  So region B is
     somewhere in between.  These are events which typically result from
     early containment failures, steam explosions, direct containment
     heating, et cetera.  So if you are going to be talking about what we are
     going to do, what we are going to be preventing, we have to consider the
     fact that as you go up in consequence, the frequencies become very low,
     the uncertainties and frequencies become very significant, but the
     uncertainties and consequences become relatively narrower.
         On the other hand, the upper bound consequences are very
     significant.  So if one wants to deal -- talk about severe accidents and
     consequences of nuclear reactor accidents, really very large in terms of
     a public mind, the issue is really mostly what we are dealing in this
     region here.  Okay.  Society as a whole, when you are talking about
     contamination area, you are talking about huge land contamination, we
     are talking about releases of this magnitude.
         Another way put, this is for a typical study we have done
     for a European power plant, which is a -- this is boiler water reactor
     with a Mark 3 containment.  This is a real curve.  In this case, the
     release is percent of core inventory.  As you can see, this shows, this
     region of the curve, most of the releases are due to areas where
     containment is relatively intact.  Release is occurring because of
     design basis leakage from the containment, and for this particular case,
     early containment venting with suppression pool being a factor.
         DR. WALLIS:  So the TMI is somewhere way down in there.
         DR. KHATIB-RAHBAR:  TMI Is somewhere down here.  It is
     actually, if you go --
         DR. WALLIS:  It is negligible.
         DR. KHATIB-RAHBAR:  It is very negligible, exactly.  In
     fact, I wish I had brought up the figure which I had compared the
     releases between TMI with these plants.  It is somewhere down in this
     line.
         MR. KAISER:  So those are individual --
         DR. KHATIB-RAHBAR:  These are individual contributions to
     this curve.
         DR. APOSTOLAKIS:  You have to speak to the microphone, Jeff.
         DR. KHATIB-RAHBAR:  That is correct.  This shows, for this
     particular plant, we had, I believe, 20-odd release categories.
         DR. WALLIS:  And Chernobyl is right on the righthand.
         DR. KHATIB-RAHBAR:  Chernobyl is actually down here.
         DR. WALLIS:  Right.
         DR. KHATIB-RAHBAR:  Somewhere up here.  Exactly.
         MR. KAISER:  Aren't you comparing a complementary cumulative
     distribution function with individual accidents on the same plot?  Isn't
     that --
         DR. KHATIB-RAHBAR:  It is.  But it is, this is a CCDF.
         MR. KAISER:  Yes.
         DR. KHATIB-RAHBAR:  Complementary cumulative distribution. 
     All this shows, these are the contributions to this, which is really,
     this is the integral curve essentially.  And this shows that -- where
     the different contributions to that part of the curve is coming from. 
     In this end, actually, the CCDF and the frequency are really one and the
     same thing.
         DR. APOSTOLAKIS:  And just for the record, they are not
     really CCDFs, they are not -- they are not complementary cumulative
     distribution functions, I don't think.  They are just curves showing the
     behavior of the frequency of exceedance, right?
         DR. KHATIB-RAHBAR:  Why do you say they are not?
         DR. APOSTOLAKIS:  Well, they are normalized, it doesn't go
     to one, does it?
         DR. KHATIB-RAHBAR:  Well, this is -- I can normalize for the
     core damage frequency, then it goes to one, yes.
         DR. APOSTOLAKIS:  Well, then it is a different curve.  But
     it is a distribution really of the frequency.
         DR. KHATIB-RAHBAR:  That is correct.  The distribution of
     the frequency, absolutely.  Which I can normalize it with frequency, I
     can get a relative frequency in a sense, rather than a probability, per
     se, depending on which school of thought you come from.
         So, well, all I am trying to show is in this region, the
     bypass events and things dominate, and this is the region where you have
     some chance of containment failure.
         DR. WALLIS:  Well, they must be probabilities because they
     haven't happened.
         DR. KHATIB-RAHBAR:  So, to restate, in the region A, even
     though the uncertainties in releases are very large, nevertheless, the
     contribution to risk, even the upper bound level, are not significant. 
     If you look at the overall risk, the contribution coming from, you know,
     releases resulting from leakage from the containment, overall,
     risk-wise, it is significant, but in terms of the consequence-wise, it
     is an insignificant contribution.
         DR. KRESS:  But when you talk about risk, you are talking
     about what?
         DR. KHATIB-RAHBAR:  Product of the two.
         DR. KRESS:  Prompt fatalities?
         DR. KHATIB-RAHBAR:  No, I am generally talking about latent
     fatalities.
         DR. KRESS:  Latent fatalities.
         DR. KHATIB-RAHBAR:  Latent, generally.  What I will be
     addressing here is mostly latent.
         DR. KRESS:  But it could be possible that they might be
     significant with respect to, say, land contamination or --
         DR. KHATIB-RAHBAR:  Latent fatalities are non-contamination,
     Tom.  In terms out they are correlated --
         DR. KRESS:  They are one to one, almost, yeah.
         DR. KHATIB-RAHBAR:  One to one ration, just about.
         DR. KRESS:  How about total injuries, for example?
         DR. KHATIB-RAHBAR:  Total injuries, again, latent total
     injuries integrated over, let's say, 50 years, they are also very
     similar, but early fatalities are not.  What is affected by inhalation
     dose will not be.  Noble cases dominate inhalation.  Okay.  But
     actinides and cesium, et cetera, will dominate the latent injuries, et
     cetera.
         DR. KRESS:  Cesium, strontium, and actinides.
         DR. KHATIB-RAHBAR:  Strontium-90 and barium-140, exactly.
         DR. WALLIS:  It would be very useful if you also showed the
     probability curve and not the integral, and then the probability times
     consequence, which would really indicate -- illustrate your conclusion
     here.
         DR. KHATIB-RAHBAR:  Yes.  Actually, I should have brought
     one in.  I have a curve like that, but since the focus is mostly on the
     FC curves, I did not want to --
         DR. WALLIS:  Your FC curves are a bit misleading.  They are
     log scale, you have to look at the extremes to figure things out and so
     on.
         DR. KHATIB-RAHBAR:  That's correct.  That's why I tried to
     put the other figure to show you, really, you need to plot frequency
     versus consequence if you want --
         DR. WALLIS:  And frequency times consequence.
         DR. KHATIB-RAHBAR:  And times the consequence, exactly,
     which is really the risk.
         DR. APOSTOLAKIS:  So just as a point of clarification, when
     you say that the uncertainties in release are very large, you mean on
     the abscissa.
         DR. KHATIB-RAHBAR:  That is correct.  I will show you.
         DR. APOSTOLAKIS:  So that is really the aleatory part.
         DR. KHATIB-RAHBAR:  Precisely.
         DR. APOSTOLAKIS:  That's what --
         DR. KHATIB-RAHBAR:  Precisely.
         DR. APOSTOLAKIS:  Okay.  Okay.  That's clear.
         DR. KHATIB-RAHBAR:  Yeah, these are the uncertainties in
     this region here.  Okay.
         On the other hand, the region V and C, the uncertainties of
     releases are much smaller, however, the contribution to risk are more
     significant because of consequences, you know, the releases are a lot
     more significant.  You are talking about kilogram quantities rather than
     microgram quantities.
         Now what are these uncertainties, because the question was
     asked also what are the uncertainties and what confidence do we have in
     their quantification?  This is the part of that curve which I was
     showing you with the horizontal line essentially.
         What you see here, these are a number of discrete release
     conditions.  That means a number of ways you can fail your containment. 
     On this x axis here you see release quantity as a percentage of total
     core inventory, initial core inventory.  These bands, the ones which are
     relatively narrow, oftentimes, as you can see here, there are a number
     of them, RC1, RC3, RC7, and a few others.  These represent the early
     containment failure and containment bypass scenarios.
         I have listed a few things.  This is, for example, for this
     plant this is the uncertainty in release.  In this case I believe this
     is for cesium iodide or cesium hydroxide -- they're very similar -- for
     a bypass event in a boiling water reactor.  This goes from about 2
     percent to as much as nearly 40 percent of the entire core inventory. 
     So about an order of magnitude.
         But the upper bound is very significant.  As you go down in
     terms of the level of release, the uncertainties become greater, which
     is also explainable, because you're talking about smaller numbers,
     you're talking about the agglomeration processes for aerosols, which are
     very difficult to quantify.  Leakage through very small parts is very
     difficult to quantify.  So the uncertainties grow as you go later and
     later and later in containment failure time basically.  Okay?  So the
     uncertainties become horrendous when you're talking about conditions
     where containment does not fail or if it fails very late.
         DR. WALLIS:  Essentially they go to zero on the left axis.
         DR. KHATIB-RAHBAR:  Exactly, because the tail end goes to
     zero.  But again if you look at the upper bound of these releases, they
     are manageable, they are not that significant.  So when we are talking
     about putting criteria on, requirements on, I think for these we need to
     look at the upper bounds.  We are really not going to try to regulate
     the tail of this distribution here.
         DR. APOSTOLAKIS:  Now the RC states are what --
         DR. KHATIB-RAHBAR:  These are release categories.  They are
     discrete ways you can fail the containment.
         DR. APOSTOLAKIS:  Are these the accident progression bins of
     the 1150 study?  In other words, it concludes the whole accident
     sequence up to and including containment failure.
         DR. KHATIB-RAHBAR:  Precisely.  Yes.
         DR. APOSTOLAKIS:  And then you are telling us what the
     consequences of these -- okay.
         DR. KHATIB-RAHBAR:  Associated with those release
     categories.  Exactly.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  Given a way you can fail the
     containment.
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  In this case, given that you can have a
     bypass, this is the uncertainty in the release.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  What are the uncertainties and what
     confidence do we have in their quantification.  Let me continue.  In
     general the subjective uncertainties are quantified using expert
     judgment, because there is no other way.  If you're talking about
     probabilities, you are talking about expert judgment.  And to the extent
     possible, models -- some of these could be incomplete.  But by and large
     there is a model for just about everything which I have discussed here. 
     But the degree of completion of the model depends on the process.  And
     of course to the extent possible, one uses test data.
         Even though it's difficult to demonstrate in my view,
     however, in my personal view, the extent of uncertainties are probably
     overstated, simply because, as it relates to release of volatile fission
     products, cesium, tellurium, et cetera, the upper bounds often involve
     nearly 100 percent release.  I mean, you cannot get any worse than that
     even if you tried.
         So because of that I think the releases are -- by and large
     the uncertainties are relatively overstated.  Again, I cannot prove it,
     because it's opinion.  But on the other hand, if you look at both
     NUREG-1150s and the studies that we have done -- I think these are the
     only two types of studies which have included the uncertainties -- by
     and large I think the uncertainties are pretty much overstated.
         DR. APOSTOLAKIS:  Now I looked at a few of the papers you
     sent me -- not in detail I must say -- but I did look, and you do state
     what you say here also in the papers that, you know, you have to use
     expert judgement, and I agree.  And then there is a quantification of
     the parameter uncertainties.
         DR. KHATIB-RAHBAR:  Yes.
         DR. APOSTOLAKIS:  Now it seems to me that in level 2 PRAs
     what is really important is the model uncertainties, the assumptions we
     make about how the phenomena will progress, and I am not sure that your
     earlier paper for nuclear science and engineering addressed those.
         DR. KHATIB-RAHBAR:  No, it did not.
         DR. APOSTOLAKIS:  It did not.
         DR. KHATIB-RAHBAR:  That paper only focused on parameters.
         DR. APOSTOLAKIS:  On parameters.  Now your second bullet
     applies -- I mean, does the second bullet address that concern a little
     bit by saying look, when in doubt, we assume that the whole thing is
     released?
         DR. KHATIB-RAHBAR:  Just about.  Yes.
         DR. APOSTOLAKIS:  So I don't have to worry about model
     uncertainty in this context?
         DR. KHATIB-RAHBAR:  Exactly.  In this context, in the
     uncertainties that we have quantified, in a sense the model uncertainty
     has been included.
         Give you an example.  In the paper you refer to, which is
     based on an old suite of computer codes developed by the NRC, rebay
     position phenomena, which is a process when volatile fission products
     sit on the surface, because they have decay heat, they can heat up, and
     because as the temperature increases, they can be volatilized again, and
     they can be released.  In that particular model, this process was not
     physically modeled, okay?  So even if we vary the parameters over five
     orders of magnitude, since the model was not there, we could not see the
     effect.  But it turns out what that process does, it actually narrows
     the uncertainties, because it increases the lower-bound releases.  The
     upper-bound releases are close to 100 percent already.
         So what that rebay position process does, it actually brings
     up the releases which were lower in quantity.  So in effect some of
     these unmodeled processes tend to narrow the uncertainties rather than
     necessarily widen then.  Okay?  So from that perspective in the numbers
     we have here, we have allowed for rebay position processes, processes
     which are not easy to model or were not at least at that time, and in a
     sense what uncertainties we represent here do cover some of these
     unmodeled phenomena.
         However, given these uncertainties, narrowing the
     uncertainties in my view are not expected within the reasonable time and
     resources that we have available.  Therefore, any use of risk curves
     should encourage ways of circumventing, okay, the potential impact of
     uncertainty.
         DR. APOSTOLAKIS:  If we go back to your slide 2 --
         DR. KHATIB-RAHBAR:  The outline?
         DR. APOSTOLAKIS:  No.
         DR. KHATIB-RAHBAR:  Slide 3?
         DR. APOSTOLAKIS:  Oh, I'm sorry.  I'm sorry, 4.  Yes, number
     4.  FC curves and uncertainties.
         DR. KHATIB-RAHBAR:  Yes, the actual curve?
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  Um-hum.
         DR. APOSTOLAKIS:  So what are we to understand from your
     discussion a few minutes ago regarding the bands that you are showing
     there, 95th percentile, 5th percentile?  Are these realistic?  Are these
     conservative, optimistic, the fifth percentile curve is not so well
     understood, but the 95th is pretty good?  What is your message?
         DR. KHATIB-RAHBAR:  The message is that this uncertainty
     band in my view is very wide.
         DR. APOSTOLAKIS:  Which --
         DR. KHATIB-RAHBAR:  The upper bound, the 95th percentile.
         DR. APOSTOLAKIS:  Yes?
         DR. KHATIB-RAHBAR:  Okay, in my view is relatively
     conservative.
         DR. WALLIS:  It looks suspiciously like a factor of 2, if
     you look at the axis and the --
         DR. KHATIB-RAHBAR:  About here?
         DR. WALLIS:  Yes.
         DR. KHATIB-RAHBAR:  It's about an order of magnitude.
         DR. WALLIS:  Goes through 2.
         DR. APOSTOLAKIS:  No, from the mean.
         DR. WALLIS:  .2 and .5.
         DR. KHATIB-RAHBAR:  Yes, exactly.  It's about a factor of 2. 
     Yes.  Yes.  Right.  Just for this case it's a case of cesium-137, okay? 
     But, yes, what I'm saying is the upper bounds are relatively
     conservative.  I mean, I'm talking about releasing, you know, most of
     these actions close to 100 percent of cesium in the vessel.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  Experimentally, I mean, FIPA's
     experiments have measured releases of those magnitudes.
         DR. APOSTOLAKIS:  No, but there are two different issues
     here.  I think you have addressed the question of the uncertainty in the
     horizontal axis there.
         DR. KHATIB-RAHBAR:  But you do release quantity, right.
         DR. APOSTOLAKIS:  You release quantity.  And you told us
     that when in doubt, we assume the whole thing comes out.
         DR. KHATIB-RAHBAR:  Correct.
         DR. APOSTOLAKIS:  So that would tend to make your curves
     flatter; right?
         DR. KHATIB-RAHBAR:  Exactly.
         DR. APOSTOLAKIS:  Flatter.
         DR. KHATIB-RAHBAR:  Right.  That's why they're larger than
     they probably should be.
         DR. APOSTOLAKIS:  Okay.  Now, if I select, though, say 10 to
     the 100th grams of cesium --
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  And I go vertically, now that's a
     different kind of uncertainty.
         DR. KHATIB-RAHBAR:  This is the uncertainty in the frequency
     --
         DR. APOSTOLAKIS:  In the frequency.
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  And what -- the argument you just gave us
     --
         DR. KHATIB-RAHBAR:  Was only on --
         DR. APOSTOLAKIS:  Does not necessarily apply.
         DR. KHATIB-RAHBAR:  No.  That's only for the consequence.
         DR. APOSTOLAKIS:  So how do you -- so you have another
     argument then supporting your assertion that the 95th percentile curve
     is conservative with respect to the frequency now?
         DR. KHATIB-RAHBAR:  With respect -- no.  With respect to the
     consequence -- I'm only focusing right now on the consequence.
         DR. APOSTOLAKIS:  Okay.  So what you're saying --
         DR. KHATIB-RAHBAR:  Only on the consequence.
         DR. APOSTOLAKIS:  Is that the curves are flatter than they
     ought to be.
         DR. KHATIB-RAHBAR:  Right.  Right.
         DR. APOSTOLAKIS:  But you are not saying anything about --
         DR. KHATIB-RAHBAR:  The width of this --
         DR. APOSTOLAKIS:  The width --
         DR. KHATIB-RAHBAR:  No.  No.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  Because the width actually is not very
     meaningful.  This shows a total uncertainty core damage frequency here. 
     It is meaningful from that standpoint.
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  But these are really exceedence
     frequencies.  They are not the actual frequencies.  Okay?
         DR. APOSTOLAKIS:  But this was one of the concerns, because
     in a regulatory framework, that uncertainty also in fact if you look at
     what Tom presented that would play a role --
         DR. KHATIB-RAHBAR:  Sure.
         DR. APOSTOLAKIS:  The uncertainty in the frequency curve.
         DR. KHATIB-RAHBAR:  Sure.
         DR. APOSTOLAKIS:  So that's where it seems to me --
         DR. KHATIB-RAHBAR:  Let me talk about the uncertainty --
         DR. APOSTOLAKIS:  Well, on the other hand -- no, I would
     certainly let you continue, but I'm trying to understand a little better
     what's going on.
         The question that I'm not sure anybody has addressed is if I
     am very conservative on the aleatory part, in other words, I make the
     curves flatter than they ought to be, how would that affect my
     uncertainties in the epistemic part, which is the frequency of the
     releases?  I don't think anybody has any thought --
         DR. KRESS:  I don't think the two are related.
         DR. APOSTOLAKIS:  No, but if I'm very conservative this way,
     I mean, horizontally --
         DR. KRESS:  Oh, you're talking about making an acceptance
     criteria of some sort.
         DR. APOSTOLAKIS:  Yes.  Yes.  Yes.  Yes.  How does that
     affect my acceptance?  Because if I know -- if I look at the curves in
     Region C, right, where for example Mohsen has 10 to the -- let's say 6
     10 to the 4th, okay, somewhere in there?  And I know from the argument
     Mohsen just gave me that I really should not be there, that maybe that's
     an artifact of the calculations that I have assumed more stuff is being
     released under conditions that really is -- that is really justified --
     so the fact then that I have uncertainty about the frequency of that
     release is -- what does that do to my acceptance criteria, since I know
     that I'm already conservative?
         DR. KRESS:  Well, you see the problem you're having is the
     problem I've wrestled with pretty hard, and that is these acceptance
     criteria which were generally risked things, prompt fatalities or land
     interdiction or something are almost universally expressed as an
     expected value, a mean.
         DR. APOSTOLAKIS:  Yes.
         DR. KRESS:  What we're looking at here is basically a hidden
     probability -- it's a cumulative distribution function, which is a
     one-to-one correlation.  So when you're looking for an expected value,
     it's an integral, and when you're talking about a metric that's an
     integral of this, you don't know how to deal with it.
         DR. APOSTOLAKIS:  It's a double integral.
         DR. KRESS:  It's a double.
         DR. APOSTOLAKIS:  It's an integral horizontally and
     vertically.
         DR. KRESS:  Well, it's not, it's a single --
         DR. APOSTOLAKIS:  Sure, because you have to eliminate all
     the uncertainties.
         DR. KRESS:  Yes, when you're talking about uncertainties it
     may be a double.
         DR. APOSTOLAKIS:  That's right.
         DR. FONTANA:  Excuse me, let's see if I understand this.  I
     came down from Mars and I got a job with Energy Research, and I have to
     do some calculations.  And I say I calculate a reactor accident, and I
     get a certain probability of a certain hole occurring or some failure. 
     Okay.  I get a certain probability, and I look across this curve, and
     that tells me that the release for that particular accident expands
     let's say 10 to the 1 to 10 to the minus 4, 10 to the 4, is a big
     uncertainty in release.
         Okay, now let poke in another department.  And now I want to
     say I can now accept a release of let's say 10 to the minus 3 -- 10 to
     the plus 3 of cesium.  And that means that the acceptable probabilities
     are that vertical band.
         DR. KHATIB-RAHBAR:  The frequency exceeds --
         DR. FONTANA:  Am I understanding this right?
         DR. KHATIB-RAHBAR:  Yes, the frequencies, all it says -- how
     do you read this curve, really, that's what it boils down to.  This
     shows the frequency of having a release greater than 1,000 grams --
         DR. FONTANA:  Yes.
         DR. KHATIB-RAHBAR:  Of in this case cesium-137 is of the
     order of 10 to the minus 6 per reactor year.
         DR. FONTANA:  Yes.
         DR. KHATIB-RAHBAR:  Frequency of exceedence.
         DR. FONTANA:  That is working from what is allowable to
     whether the plant is meeting that.  In other words, if I am doing an
     accident analysis, I'm going to get a probability of a certain failure,
     I'm going to calculate some releases.
         DR. KHATIB-RAHBAR:  Right.
         DR. FONTANA:  If I pick a probability and look across this
     curve, it looks like I've got a wide band of --
         DR. KHATIB-RAHBAR:  Of releases.  Sure.
         DR. FONTANA:  Of releases.
         DR. KHATIB-RAHBAR:  Sure.  This curve shows you the
     uncertainties both in release quantity and the frequency both.
         DR. FONTANA:  Okay.
         DR. KHATIB-RAHBAR:  Right.  Absolutely.
         DR. FONTANA:  I just wanted to see if I understood.
         DR. KHATIB-RAHBAR:  This is one -- essentially a good way to
     show both uncertainties.  In the release frequencies this is the total
     core damage frequency uncertainty, okay?  And this is for a given, you
     know, depending where you are at in this region, this is the uncertainty
     in the quantity of release.  All it shows you, as you're going to lower
     and lower frequencies, the frequencies are becoming more uncertain, and
     that makes sense.  Why?  Because these are mostly, let's say, steam
     generator tube rupture events, initiating events, but that depends --
         DR. WALLIS:  I think you're saying that CDF is pretty in
     good shape, it's within a factor of 2.  LERF is awfully uncertainty.
         DR. KHATIB-RAHBAR:  Precisely.
         DR. WALLIS:  How you get that steep cliff.
         DR. KHATIB-RAHBAR:  Exactly.  Exactly.
         DR. WALLIS:  Depends on lots of uncertainties.
         DR. KHATIB-RAHBAR:  Frequency points.  Absolutely.
         DR. WALLIS:  You can put it anywhere between 10 to the minus
     8 and 10 to the minus 6.
         DR. FONTANA:  I've got another thought, too, and I know you
     can't draw a curve like this, but if you drew it on linear terms rather
     than log terms, you'd get a curve that would look like this --
         DR. KHATIB-RAHBAR:  Um-hum.
         DR. FONTANA:  And it would probably make it a little easier
     to see conceptually.
         DR. APOSTOLAKIS:  I'm sure that would help the people who
     will read the transcript.
         Joe Murphy had that comment before.  Yes, there are several
     comments.  Joe has been trying to say something for awhile.
         MR. MURPHY:  I just want to point out that what we're trying
     to do -- I agree with what Mohsen has said, but understand that we were
     comparing on the horizontal, you're comparing the uncertainty in an
     integral to the uncertainty in the scalar.  If you want to compare them
     on an equal basis, you have to differentiate the curve.
         DR. KHATIB-RAHBAR:  Yes, you've got to put just frequency
     versus consequence, as was indicated earlier.
         MR. MURPHY:  And it is good information there, but the
     broadness of that horizontal line, recognize just the broadness of the
     integral and not of the differential.  The differential would be much
     smaller.
         DR. KHATIB-RAHBAR:  This is what I can show you by that,
     just showing you this is the width, it's just right there.
         DR. APOSTOLAKIS:  Yes, yes.  Yes, yes, yes.
         I think what you're saying -- yes, let's go back to the
     other -- I think what you're saying, Mohsen, is really that you feel
     that the 95-5th percentile curve is reasonably conservative --
         DR. KHATIB-RAHBAR:  In terms of consequence.
         DR. APOSTOLAKIS:  But the mean in the 5th percentile a more
     realistic analysis would move down.  That's really what you're saying. 
     It wouldn't be so flat.  It would be steeper.
         DR. KHATIB-RAHBAR:  Yes.  This curve would be steeper.
         DR. APOSTOLAKIS:  The mean of the 5th percentile.
         DR. KHATIB-RAHBAR:  It will be steeper.
         DR. APOSTOLAKIS:  Now we had --
         DR. KHATIB-RAHBAR:  Well, the 95th itself could also be
     steeper.
         DR. APOSTOLAKIS:  Yes, but maybe, since it's an upper bound
     --
         DR. KHATIB-RAHBAR:  In a sense the mean, George, the mean --
         DR. APOSTOLAKIS:  I understand.
         DR. KHATIB-RAHBAR:  In these types of things --
         DR. APOSTOLAKIS:  Is driven --
         DR. KHATIB-RAHBAR:  Is very close-driven by the upper bound.
         DR. APOSTOLAKIS:  You're right.  You're right.
         DR. KHATIB-RAHBAR:  It makes no difference.
         DR. APOSTOLAKIS:  You're right.  Well, it would -- anyway. 
     But --
         DR. KHATIB-RAHBAR:  Because he was talking about difference
     between zero to 1.  I mean, the mean --
         DR. APOSTOLAKIS:  I got the message; yes.
         DR. KHATIB-RAHBAR:  Is going to be much closer to 1 than to
     zero.
         DR. APOSTOLAKIS:  Yes.  So even the 95th you are saying --
         DR. KHATIB-RAHBAR:  Precisely.
         DR. APOSTOLAKIS:  Would go down.  That's already
     conservatism built in.
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. APOSTOLAKIS:  Jeff?
         MR. KAISER:  Yes, I guess I'm a bit puzzled about the
     interpretation in the horizontal of that as an uncertainty in the
     magnitude of release.  If you had a set of individual accidents and it
     happened that, let's say, between about 1 and 10 to the 3 curies you had
     no accidents at all, wouldn't that lead to a curve that looks something
     like this?
         DR. APOSTOLAKIS:  Maybe your previous transparency explains
     that, the one you just put up there with the release categories.
         DR. KHATIB-RAHBAR:  This one here?
         DR. APOSTOLAKIS:  Yes.  Maybe we can put the -- is that what
     you mean, Jeff?
         MR. KAISER:  Yes.
         DR. APOSTOLAKIS:  So where would you have no accident --
     releases.
         MR. KAISER:  If you happened to have none -- you go back to
     the other slide we were just looking at.
         DR. APOSTOLAKIS:  I'm sorry.
         DR. KHATIB-RAHBAR:  Which one do you want?  This one?
         DR. APOSTOLAKIS:  You mean the flat part.
         MR. KAISER:  Yes.  The flat part could be --
         DR. APOSTOLAKIS:  Sure.
         MR. KAISER:  Reproduced if there were no accidents at all.
         DR. APOSTOLAKIS:  Yes.
         MR. KAISER:  In that interval with the slant.  So in other
     words it may have nothing to do with uncertainty on the magnitude of
     release.
         DR. APOSTOLAKIS:  It may not.  You're right.  You're right.
         The other things -- I try to differentiate --
         DR. KHATIB-RAHBAR:  Let me make sure I understand that
     point.  What you were talking -- the uncertainty is this.  The head and
     the tip -- I mean the bottom and the tip of this thing right here. 
     Okay?  In the quantity of release.
         DR. WALLIS:  Your artist has drawn one which actually goes
     down to the left, which is impossible.
         DR. KHATIB-RAHBAR:  What is impossible?
         DR. WALLIS:  Because you're having negative releases around
     the middle of that --
         DR. APOSTOLAKIS:  Well, that may be an --
         DR. KHATIB-RAHBAR:  Down here?
         DR. WALLIS:  No, in the middle, underneath --
         DR. APOSTOLAKIS:  Well, the curve goes up a little bit --
         DR. WALLIS:  Underneath the 95th --
         DR. APOSTOLAKIS:  Where it says "mean."
         DR. WALLIS:  It goes down to the left.
         DR. APOSTOLAKIS:  But that's an artifact of --
         DR. KHATIB-RAHBAR:  Yes, this is an artifact of the graph.
         DR. APOSTOLAKIS:  No, but let's not forget, as I forgot when
     I was differentiating by hand, that this is a logarithmic scale.  So the
     fact that it's flat there does not mean that there are no accidents,
     because -- if the frequency were flat, the frequency itself, then that
     means you are not contributing anything, right?
         So the fact -- if I take the curves that Mohsen is showing
     as real, then there are accidents all over the place, because the fact
     that the logarithm is flat doesn't mean anything.
         DR. KHATIB-RAHBAR:  As this one shows.
         DR. APOSTOLAKIS:  As this one shows.  That's right.
         DR. KHATIB-RAHBAR:  This is the whole idea.  As this one
     shows, there are accidents all over --
         DR. APOSTOLAKIS:  But Jeff's point was good in general.  If
     there are no accidents there, then you will see the frequency itself
     being a lot, not the logarithm.
         DR. WALLIS:  No, the frequency would be zero if there were
     no accident.
         DR. APOSTOLAKIS:  No, because it's cumulative.  The density
     will be zero.
         MR. KAISER:  The density will be zero, but this would be
     flat.
         DR. APOSTOLAKIS:  Yes.
         MR. KAISER:  But you'd still have the 95th, the 5th and the
     mean, separated.
         DR. APOSTOLAKIS:  Yes.
         MR. KAISER:  But you would not be entitled to take the, you
     know, the gap between two extremes as an uncertainty on magnitude of
     release.
         DR. APOSTOLAKIS:  Right.
         MR. KAISER:  It doesn't necessarily mean that.  So I think
     we should be careful not to assume --
         DR. APOSTOLAKIS:  Yes.
         MR. KAISER:  That it does mean that.
         DR. APOSTOLAKIS:  Yes.  Actually the cumulative in general
     is not the best way to communicate these results.  It's the PDF that --
     that's why PDFs were invented.  But --
         DR. KHATIB-RAHBAR:  Yes, but in a sense, though, the
     cumulative in a sense is also nice that it shows you the flatness of the
     curve is --
         DR. APOSTOLAKIS:  Sure.  But it's logarithmic, and that's,
     you know --
         MR. KAISER:  We could talk about some more numbers.  Right.
         DR. WALLIS:  It must be flat at both ends, because it is
     logarithmic.
         DR. KHATIB-RAHBAR:  Sure.
         DR. WALLIS:  You can't get away from it.
         DR. KHATIB-RAHBAR:  But it's how flat it is in the middle
     that's what matters, really.
         DR. WALLIS:  It's how curved it is that matters.
         DR. APOSTOLAKIS:  Dr. Shack has a question.
         DR. SHACK:  As a way of characterizing risk, in my
     simpleminded view I look at CDF and LERF as a way of pinning the
     horizontal part of that curve and then going out somewhere and saying
     I'm going to force this much of the curve to be so low at that LERF
     value, and then are we -- I guess the question is how much more
     information do I get -- when I've pinned the flat part and I've pinned
     that part at the LERF, how much variation can I get in the curve?  Am I
     really getting any more information out of the whole curve than I am by
     pinning those two points?
         DR. KHATIB-RAHBAR:  Yes, you are, because it is telling you
     in a given plant if you're looking at whether the likelihood that you
     can in a sense for these fission products, conditional probability or
     the probability of releasing fission products is really a spectrum of
     conditions that can lead to release of fission products.  On the one
     hand you have those which are very large releases, bypass events, et
     cetera, but you can have a number of other scenarios where your
     containment can fail, you know, under different conditions.  So it is
     giving you the spectrum of conditions that your containment can fail,
     and it's showing the effect of mitigation, how it's spread out.  It is
     really not either the containment fails or it doesn't fail.  There is
     something in between.  That's all it's telling you.
         But in terms of overall magnitude of release, you are
     absolutely right, what is really dominating is that LERF -- releases
     part of it.
         DR. APOSTOLAKIS:  Okay.  Let's go.
         MR. KAISER:  Could I ask something?
         DR. APOSTOLAKIS:  Sure.
         MR. KAISER:  Because if it so happened that we had accident
     sequences in which we could define exactly the amount of cesium
     released, so there's no uncertainty on those amounts, but we had some
     uncertainty on the frequencies.
         Wouldn't the -- you have taken it away.  Could you put it
     back?
         DR. KHATIB-RAHBAR:  Need it again?
         MR. KAISER:  Thanks.  I mean one that would still look
     exactly like this even if there was no uncertainty on the magnitude of
     release.
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  The curve itself would look exactly like
     this, right -- one of these curves.
         MR. KAISER:  So it proves my other point.  How can you
     interpret it this way, as an uncertainty on magnitude of release?  I
     don't think you can.
         DR. KRESS:  You can't.
         MR. KAISER:  No.
         DR. APOSTOLAKIS:  No.  All these uncertainties like how much
     is it released if I have a release category and so on have to be
     converted to uncertainties and frequencies when you produce these
     curves.
         MR. KAISER:  Exactly.
         DR. APOSTOLAKIS:  Because the release or the consequence
     here is the independent variable.
         DR. KRESS:  Yes, it has no uncertainties --
         DR. APOSTOLAKIS:  It is the independent variable.  You enter
     the figure that way.
         MR. KAISER:  Precisely.
         DR. APOSTOLAKIS:  You say 10 grams --
         MR. KAISER:  Right.
         DR. APOSTOLAKIS:  -- now tell me what you know about it and
     what you know is you go up and say it will be released with such and
     such a frequency and the frequency is uncertain.
         Now when you produced it in the PRA, it doesn't mean you did
     it that way.  You did it in fact the way that Mohsen showed us.  You
     have the release categories and you have uncertainty on this, and then
     there has to be some process there that converts it to this, because I
     have seen curves where they talk about both the uncertainty on release
     and frequency, which is of course wrong.
         DR. KRESS:  And of course you cannot talk about the
     frequency of a specific release because these are cumulative
     complementaries.
         DR. APOSTOLAKIS:  Sure -- or greater.
         DR. KRESS:  It is greater.  It makes it even harder to deal
     with the curve.
         DR. POWERS:  George, I guess I understand your
     interpretation which says that I am going to use release as an
     independent variable until you asked questions about this figure.
         You could turn the figure by 90 degrees and alternatively
     say I will take the frequency as the independent variable and ask
     questions about the uncertainty in release.
         DR. APOSTOLAKIS:  No.  I don't think you can do that because
     of the way that these curves are produced.
         DR. POWERS:  Well, what I know is that the frequency of any
     given is an uncertainty and having specified that any given sequence
     that the release one calculates to be associated with that sequence is
     an uncertainty, so in fact what you are plotting here is the product of
     two rather uncertain quantities which are not independent of each other.
         DR. APOSTOLAKIS:  We are becoming way too technical now, but
     the way the curves, the three curves that Mohsen is showing there are
     produced, say, from the simulations is such that you cannot really do
     what you said, fix a frequency and say, ah, with this frequency here is
     my uncertainty in the consequences, because if you put back -- those two
     are the most popular transparencies I guess.
         Can I go up there?
         DR. WALLIS:  I think you could, George, but you would end up
     with different percentile curves.
         DR. APOSTOLAKIS:  Yes.
         DR. WALLIS:  You could do it, but you would end up with
     different percentile curves.
         DR. APOSTOLAKIS:  You would have a different curve, a
     different figure.
         DR. POWERS:  I think Graham is probably right but having
     been in the business of producing such curves, I think the way they get
     produced in fact allows you to switch back and forth.
         DR. APOSTOLAKIS:  For example, here, in this figure, the PRA
     itself will give me the RCJ and it will give me the frequency of the
     RCJ, right, for each one?
         DR. WALLIS:  With uncertainty.
         DR. APOSTOLAKIS:  Yes, with a curve.  It is a curve.  Then I
     would, to generate the frequency-consequence curve I would enter this
     figure, say, at 10 to the minus 3 percent of core inventory, okay, or
     greater if you want, and I have to go vertically now and see how I
     intersect the horizontal bars that Mohsen has up there and every time I
     intersect I get the amount and I get the corresponding frequency from
     the probability distribution of the frequency of RC 10, RC 11 and so on,
     and then I have to take some sort of a convolution to come up with a
     frequency.
         In other words, I am converting a figure where the release
     is not an independent variable to one where it is.
         DR. WALLIS:  Oh, but George, really you have got an ellipse
     there, you have got an ellipse with uncertainties in both axes and you
     can take your convolution, cut it horizontally or vertically and
     integrate.  It doesn't matter.
         DR. APOSTOLAKIS:  Yes, but then you can produce curves like
     the ones Dana wants, but it will be different curves.  I can do it that
     way.  Gareth?
         MR. PARRY:  I am not sure this is going to help, but --
         DR. APOSTOLAKIS:  Unlike the previous comments, eh?
         [Laughter.]
         MR. PARRY:  But I suspect that what these bounds here that
     are called uncertainty bounds are, they are a combination of -- and you
     raised the term so I am not afraid to use it -- of both aleatory and
     epistemic elements, which means -- because these are results from a
     particular type of accident and the underlying conditions in that
     accident could be variable so that this may have some frequency --
         DR. APOSTOLAKIS:  Which is this now?
         MR. PARRY:  Let's take RC1.  RC1 could have an element
     embodied in that, what is called an uncertainty band, of what is really
     a frequency of an accident, so this is an uncertainty and embodied
     within a frequency distribution, I think.
         DR. KHATIB-RAHBAR:  No.  No, not at all.  In fact, if I were
     to put this graph up and take away the uncertainty in frequency, I would
     find the pinch point right -- everything collapsed --
         We may be talking about different frequencies.
         DR. APOSTOLAKIS:  You are saying it is called epistemic. 
     Good.
         DR. KHATIB-RAHBAR:  For this particular case, yes -- no, for
     the frequency.
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  If I take away the frequency uncertainty
     these two will collapse right at the V Sequence point.  All it shows is
     the uncertainty V Sequence which is Event V Bypass in this region.  If
     you don't have a frequency uncertainty, okay -- on a vertical axis these
     two will collapse.
         MR. PARRY:  So your V Sequence is so precisely defined that
     all the parameters associated with it are --
         DR. KHATIB-RAHBAR:  If you take away the frequency
     uncertainty.
         MR. PARRY:  Which frequency uncertainty?  I am not talking
     about the initiating event.  I am talking about other things that
     underlie that which could be -- I don't exactly where the --
         DR. KHATIB-RAHBAR:  Stochastic type issues.
         MR. PARRY:  -- where the leak is, for example.
         DR. KHATIB-RAHBAR:  Stochastic type issues.
         MR. PARRY:  Which is what I am saying right, and those
     issues then should really be treated as aleatory and not epistemic and
     they complicate the things.
         DR. APOSTOLAKIS:  We have got the same problem with the
     Army's -- what is it? -- incineration of chemical munitions, right?
         Imagine that you create these curves with Monte Carlo
     simulation or Latin hypercubes.  A single simulation, in other words a
     single set of input variables, no epistemic uncertainty -- you know, you
     selected the variables -- will result in one curve that will show the
     aleatory uncertainty on the cesium released.  If there was no aleatory
     uncertainty you would get one point, so get a curve.  The uncertainty
     horizontally, as Mohsen said earlier, because of the uncertainty, you
     know, the aleatory part.
         The fact that you have epistemic creates a family of curves. 
     They are not unsure.
         The problem with that is that when you say 95th percentile,
     this curve that Mohsen is showing connects all the 95th percent
     percentiles of the various -- of the spaghetti curves, okay?  This is
     not what you would get by using 95th percentiles for all the parameters,
     right?
         MR. PARRY:  Right.
         DR. APOSTOLAKIS:  Okay, so it is a pointwise derivation but
     this is really becoming too detailed now.  It forced me to introduce
     additional technical terms like spaghetti that have not been explained.
         I think --
         DR. KRESS:  Fettucini.
         DR. APOSTOLAKIS:  Why don't we agree that the uncertainties
     we are talking about here are really between the 5th and 95th and the
     message that Mohsen sent us is that in his belief the 5th percentile
     curve should be steeper than it is, okay? -- and he also claimed that
     the 95th percentile curve should also be steeper, therefore driving the
     mean, but perhaps not as much.  I think that is the message he is
     sending, because otherwise we will spend the whole afternoon discussing
     the Monte Carlo stuff, okay?
         DR. KHATIB-RAHBAR:  Other questions?
         DR. APOSTOLAKIS:  Let's go on.
         DR. KHATIB-RAHBAR:  I don't remember what I was going to say
     next.
         DR. APOSTOLAKIS:  You said you had two parts in your
     presentation.
         DR. KHATIB-RAHBAR:  I am not even through the first part
     yet.
         DR. APOSTOLAKIS:  Okay. When you are through with the first
     part, maybe we can take a short break.
         DR. KHATIB-RAHBAR:  Okay.
         The next question is what to use as the consequence measure. 
     In my mind the problem of nuclear accidents really is the catastrophic
     potential for contaminating a very large area of land for a very long
     time -- the Chernobyl accident is given a point in time.
         DR. APOSTOLAKIS:  So it is not risk?
         DR. KHATIB-RAHBAR:  The risk is of ground contamination
     depending what risk measures --
         DR. APOSTOLAKIS:  Well, I know, but I mean --
         DR. KHATIB-RAHBAR:  It's the risk of heart attacks.  Land
     contamination is of concern because of health effects.
         DR. WALLIS:  It is not the land.  It's the land if used by
     humans.
         DR. KHATIB-RAHBAR:  By humans, absolutely.  Especially in
     Europe where land is relatively scarce, the population in terms of
     density is fairly high, so land contamination, useful land contamination
     becomes more significant.
         Prompt fatalities could often be mitigated by offsite
     emergency counter-measures.  Again, in the case of Switzerland, for
     example, the offsite emergency counter-measure often involves
     sheltering.  All Swiss homes are equipped with a basement which was
     designed as a fallout shelter, so during an accident most people
     vertically evacuate.  They go down to the basements.
         As a result, if you do away with the cloud passage,
     typically the early fatalities are relatively small.
         DR. KRESS:  How viable is that option in this country?
         DR. KHATIB-RAHBAR:  In the U.S.?  That is another issue. 
     Actually we looked at it a number of years ago for the NRC.  It is
     viable.  It is relatively viable.
         DR. APOSTOLAKIS:  So let me understand your first bullet.  I
     am still trying to digest it.  If it were up to you, you would revise
     the safety policy statement of the Commission --
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. APOSTOLAKIS:  -- contamination --
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. KRESS:  And you would --
         DR. KHATIB-RAHBAR:  I will come to that later.
         DR. KRESS:  And the mean value of the acceptance values on
     prompt fatality would be without evacuation but sheltering instead? 
     That's an interesting thought.
         DR. KHATIB-RAHBAR:  Because you would actually eliminate one
     large uncertainty by doing it that way.
         DR. WALLIS:  There is also the catastrophic potential of
     wiping out a source of useful energy because of the reaction to the
     accident and that is a catastrophe too.
         DR. KHATIB-RAHBAR:  Yes.  I haven't looked at that yet.
         DR. SEALE:  Few people do.
         DR. POWERS:  Your last point is an endorsement of hormesis.
         [Laughter.]
         DR. KRESS:  That is supposed to be uninhabited.
         DR. APOSTOLAKIS:  Uninhabited, yes.  A very unfortunate
     place to have --
         DR. KHATIB-RAHBAR:  This source is actually incorrect. 
     Coming on the plane on Saturday I looked at this and I came to my office
     trying to dig out this source to bring you a copy.  I realized this is
     not the correct paper, where this thing came from.  I can dig it out for
     you.
         DR. APOSTOLAKIS:  Yes, if you could.  Okay, thanks.
         DR. KHATIB-RAHBAR:  The issue is if you take one-half a gram
     of Cesium-137, you spread it uniformly over one square kilometer, it
     makes the area fully unhabitable for essentially a very long time. 
     Chernobyl is a point in time.
         DR. KRESS:  That is a release fraction of what, 10 to the
     minus 20?
         DR. KHATIB-RAHBAR:  A very, very small quantity.  Very, very
     small quantity.
         DR. KRESS:  That end of the curve could be important way
     down there on the release fractions.
         DR. KHATIB-RAHBAR:  Well, this is very ideal.  I mean it's
     taking half again a gram and spreading it uniformly is the point.
         DR. KRESS:  It is indicative though of what small releases
     can do to you --
         DR. KHATIB-RAHBAR:  But the release fractions could be much
     greater if you went to backup.  This issue, a few years ago when we were
     debating this, it came up.  This does not mean half a gram released from
     the reactor.
         DR. FONTANA:  It says there was 165 kilograms of cesium.
         DR. KHATIB-RAHBAR:  That's correct.  You have got a total of
     165 kilograms in the reactor typically, for a lot of power reactors and
     when you are talking about half a gram, this is what is on the ground if
     you uniformly spread that.
         DR. APOSTOLAKIS:  What do you mean by totally uninhabitable?
         DR. KHATIB-RAHBAR:  In other words, you cannot use that land
     for a long time.
         DR. APOSTOLAKIS:  You cannot drink the water, underground
     water, you cannot send your cows there.
         DR. WALLIS:  I think it is worse for grazing animals.
         DR. POWERS:  I think that water problems arise with cesium
     contamination One is that it does get preferentially uptaken into
     plants.  The other one is a really interesting one that does arise in
     the Chernobyl accident, where it goes onto pine needles and one
     subsequently has forest fires.  It migrates.  The hot spot moves around
     on you.  It's kind of fun.  Moths pick the stuff up and they transmit it
     and then tend to go from tree to tree and things like that, so you have
     some interesting transmission with cesium.
         But I think deep groundwater is by and large very poorly
     effected in some soil types.  Now, there are -- I can find soil types
     where it is egregiously effected, but deep groundwater is slow to effect
     because of the high absorption potential of cesium.  By the time cesium
     gets to the ground water aquifer, it has decayed several half-lives.
         DR. APOSTOLAKIS:  So the half-life he has there, what is the
     half-life?
         DR. POWERS:  It is 12 or 13 years.
         DR. KHATIB-RAHBAR:  Yes.  A typical cesium-137 is about a 30
     year half-life, this is in days, 10,000 days, 11,000 days, 30 years.
         If you look at this table, this comes up from, I guess, one
     of these original -- 54 dominant radionuclides.  These are just a few
     typical ones for a typical pressure water reactor.
         This is the inventory in becquerel, what is the fuel for a
     typical PWR which has been running equilibrium cycle timeline.
         DR. WALLIS:  It is a very interesting calculation to spread
     that over the world and to look at grazing animals, and look at the
     standards for becquerels in meat, and you come up with a horrific
     conclusion about how you can't eat these animals.  It is not too
     difficult a calculation.
         DR. KHATIB-RAHBAR:  First of all, let's focus on noble
     gases.  In terms of the overall inventory, as you can see, the total
     inventory for a typical reactor is dominated by noble gases.  But noble
     gases oftentimes have a very, very short half-life.  For example, you
     keep the containments closed for about 12 hours, you can decay roughly
     half of your noble gases.  That's why, when they talk about early
     fatalities, what really dominates early fatalities are noble gases.  And
     if you can do away with noble gases, you have done away with a large
     portion of your early fatalities.  This dominates the inhalation dose. 
     Okay.
         The same for iodine.
         DR. KRESS:  Isn't there a pretty sizable fraction due to the
     I-131, though?
         DR. KHATIB-RAHBAR:  Yes, I am coming to iodine-131.
         DR. KRESS:  Yes, but for prompt fatalities, too.
         DR. KHATIB-RAHBAR:  Yes, I am coming to the next item,
     iodine group.  The iodine group, you know, this is typically about five
     radionuclide -- radioisotopes.  Iodine-131, you know, if you look at it
     in terms of the overall Becquerel, 134 dominates, but it doesn't have a
     substantial half-life, it decays right away.  So the one which really is
     significant is iodine-131, which has an eight-day half-life, you know,
     or one-day half-life here.
         DR. APOSTOLAKIS:  That is what Farmer used, right?
         DR. KHATIB-RAHBAR:  Yes.
         DR. POWERS:  Understand, however, you can get some
     misleading things out of this.  For instance, 132 half-life really
     doesn't count, it is the tellurium-132 half-life that counts here.
         DR. KHATIB-RAHBAR:  Yes.
         DR. POWERS:  Okay.  So the fact that 132 goes away very
     quickly, but it is only being generated --
         DR. KHATIB-RAHBAR:  I am just looking at first order type
     effects, yes.
         DR. POWERS:  That is the first order effect.
         DR. KHATIB-RAHBAR:  Yes, but tellurium, again, is not a
     significant concern.
         DR. POWERS:  It is because it produces the 132 iodine that
     you have to worry about it.
         DR. KHATIB-RAHBAR:  Tellurium --
         DR. KRESS:  It decays.
         DR. KHATIB-RAHBAR:  It decays, yes, it decays to iodine-132,
     that is correct.  But, again, if you look at it, really, what dominates
     again on this first order, this is not complete, this is only a few of
     them, cesium sticks its ugly face out.  Okay.  That is really what
     dominates.
         And there are other things, for example, strontium-90,
     barium-140, these are significant risk factors in terms -- because
     strontium goes into the bone marrow.  Barium-140 is also the same.  So
     these tend to be very significant.
         Quantity of release, if you look at the mass quantity of
     release, you have for a typical PWR about 170-odd kilograms of cesium,
     but you have tons of things like plutonium and other things.  Okay.
         So, again, this gives another credence why cesium is more
     significant.  Again, if you can deal with the early counter-measures
     with iodine issue, I think the iodine issue totally goes away, because
     that is really -- the concern with iodine is primarily it gets into the
     milk and goes into the thyroid.  And if you can, for example, dump the
     milk for several days and come up with some counter-measures for iodine,
     you can do away with the effects of iodine altogether.
         It still does hang on, there is a contribution from iodine. 
     In fact, the calculations that we did, there's about a 10 percent
     contribution from iodine, latent fatalities.  It is not altogether
     eliminated, but its significance becomes much smaller.
         DR. APOSTOLAKIS:  So, let me understand this, what you are
     saying is that early fatalities are dominated by iodine?
         DR. KHATIB-RAHBAR:  They are dominated by things like noble
     gases and iodines, right.
         DR. APOSTOLAKIS:  Yes.  Okay.  But we can do something about
     them?
         DR. KHATIB-RAHBAR:  To mitigate their impact by emergency
     actions or different counter-measures that we can take.
         DR. APOSTOLAKIS:  Well, in essence, what you can say then is
     that you are using the release of iodine and noble gases as a surrogate
     goal for early fatalities, but you are confident you can do something
     about it.
         DR. KHATIB-RAHBAR:  Yes.
         DR. APOSTOLAKIS:  Which is the same thing for cesium for
     latent.  So you are not really using only the cesium frequency
     consequence curve, you are using both.
         DR. KHATIB-RAHBAR:  We are -- in the curves that we have,
     the concept of equivalence which is developed is on the basis that we
     have looked at 54 risk dominant radionuclides.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  And we have determined the latent
     effects of different radionuclides in this family of 54 radionuclides,
     and we have said, if you release 100 percent of strontium-90, how much
     would that be equivalent to in terms of mass if you convert it back to
     cesium?
         DR. APOSTOLAKIS:  But this only for latent?
         DR. KHATIB-RAHBAR:  This is only for latent.
         DR. APOSTOLAKIS:  So what I am saying is if I read your
     paper with the Swiss, without talking to you, I get the impression that
     you are only using cesium.
         DR. KHATIB-RAHBAR:  No, it says cesium equivalent.
         DR. APOSTOLAKIS:  The equivalent, equivalent.  But this is
     for latent.  But now, after talking to you, I also know that you can
     write another paper for iodine and noble gases for the early releases,
     but you simply don't worry too much about it because you are confident
     that you can do some, you can manage it.
         But, in essence, you are using both.  You are not neglecting
     early fatalities.
         DR. KHATIB-RAHBAR:  No, we are not neglecting it, but we
     focus, as I said.
         DR. APOSTOLAKIS:  You are focused on latent.
         DR. KHATIB-RAHBAR:  The issue, cesium issue is latent
     effects.  The issue, is my view, is not early effects.
         DR. KRESS:  Not only that, George, with the possible
     acceptance of noble gases, you can probably correlate the release of the
     volatile fission products.  So that if you pegged the cesium, you
     probably also know what the iodine is.
         DR. KHATIB-RAHBAR:  Yes.
         DR. KRESS:  So that if you just use cesium, you can
     incorporate it as the iodine part of it, actually, because they are
     correlated.  Now, there is a little problem with noble gases doing that
     because it is not a good -- not as firm a correlation, but there is even
     a correlation there.
         DR. APOSTOLAKIS:  So you are -- I mean they are not
     independent.
         DR. KRESS:  They are not independent, that's right.
         DR. KHATIB-RAHBAR:  Yes.  That is a very good point Dr.
     Kress brings up.  The cesium, iodine, nobles gases, even tellurium, to
     some extent, are essentially correlated by roughly correlation
     coefficient close to one.
         DR. KRESS:  Yes, it is close to one.
         DR. KHATIB-RAHBAR:  Close to one.  The reason that the
     release to environment for noble gases is overly correlated, in the
     sense, because noble gases do not retain, they just escape.
         DR. APOSTOLAKIS:  So let me understand this.
         DR. KRESS:  That's why it is a little problem with that.
         DR. KHATIB-RAHBAR:  Precisely.
         DR. APOSTOLAKIS:  Are you claiming then that all I have to
     worry about is cesium equivalent?
         DR. KRESS:  I think you could probably do it with just
     cesium.  Maybe you need two of them, maybe you need cesium and something
     for -- or maybe the actinides.
         DR. KHATIB-RAHBAR:  Yeah.  What we have done --
         DR. SEALE:  On cesium, he is smart to know what the
     correlation is.
         DR. APOSTOLAKIS:  Yeah.  But he says equivalent, so, I
     assume he is smart enough to come up with that.
         DR. KHATIB-RAHBAR:  Yeah, the equivalence has not been
     developed.  Let me give you the weights, this is for particular
     calculations we had done.  The iodine contribution was 10 percent,
     cesium is not 1, it is .6, 60 percent, tellurium is about 1 percent,
     Dana.  Strontium, barium is a factor of roughly 2, 1.7.
         DR. APOSTOLAKIS:  Percent of what?
         DR. KHATIB-RAHBAR:  This is just the weights.  The
     contribution essentially to latent fatalities, if you look at it.  If I
     were to derive the cesium equivalents, as I said, I would take all the
     inventory for these 54 radionuclides released into the environment and
     try to see how many latent fatalities they would cost over a period of
     50 years, I forget, integrated out.  And then you find out how many --
     how much in terms of mass of cesium would they correspond to, per
     radionuclides, and I can weigh it according to dose.  Okay.  So the
     weighting factor that I get the contribution to the essential latent
     fatalities, there is some from iodine.  As I said, it is roughly about
     10 percent.
         DR. KRESS:  You just multiply the cesium fatalities by --
         DR. KHATIB-RAHBAR:  Precisely.
         DR. KRESS:  Divide it by .6, basically.
         DR. KHATIB-RAHBAR:  Right.
         DR. KRESS:  Then you cover the others.  And that's the way
     you would handle it, sort of.
         DR. WALLIS:  Would you tell me how many fatalities you come
     up with when you make this calculation?
         DR. KHATIB-RAHBAR:  I don't have the number, it is
     thousands, thousands.
         DR. WALLIS:  It is thousands?
         DR. KHATIB-RAHBAR:  Thousands, yes.
         DR. WALLIS:  It is not hundreds of thousands?
         DR. KHATIB-RAHBAR:  Hundreds of thousands, I forget.
         DR. KRESS:  It depends on the release.
         DR. WALLIS:  It is a large number of people.
         DR. KHATIB-RAHBAR:  Sure.  When you integrate it over a 50
     year period to thousands of miles, of course.
         DR. WALLIS:  I think you need to tell us these numbers,
     because that tells us how important it is.
         DR. KHATIB-RAHBAR:  What numbers?
         DR. WALLIS:  Are we dealing with a thousand people or
     100,000 people?
         DR. KHATIB-RAHBAR:  Oh, in terms of numbers, those numbers
     are readily available, I mean 1150 gives you those numbers, that is why
     I didn't want to things that are already available.  I mean 1150 has
     those numbers, you can look in the appendices.
         DR. KRESS:  You don't want to look at just the deaths in
     absence of the probability.
         DR. KHATIB-RAHBAR:  Frequencies.
         DR. WALLIS:  Now, George put some words in your mouth, he
     said you were confident that you could handle these sort of immediate
     fatalities with emergency measures.
         DR. KHATIB-RAHBAR:  By and large.
         DR. WALLIS:  Now, I am not confident that the emergency
     measures will work.  You are relying upon a whole lot of societal things
     to happen, people to respond to something they have never seen before,
     with a lot of uncertainty about what is going on.  I think that is not
     something you can rely on.
         DR. KRESS:  You might tell somebody to go down to their
     basement and they will say, I am getting out of here.
         DR. WALLIS:  If you think about Three Mile Island, there was
     a lot of confusion about shall we evacuate, shall we not evacuate?  Is
     it serious, is it not?  What is going on?
         DR. KRESS:  A lot of uncertainty.
         DR. KHATIB-RAHBAR:  The whole concept of emergency actions
     is controversial, you are absolutely right.  But what I personally --
     personally promote is sheltering concept, because I think sheltering is
     cheap.
         DR. KRESS:  It is the least uncertain.
         DR. KHATIB-RAHBAR:  It is least uncertain, and you can
     demonstrate that.
         DR. WALLIS:  Or it could be handled if there were proper
     plans and everybody did what they are supposed to do.
         DR. KHATIB-RAHBAR:  Yeah, it just doesn't work that way,
     unfortunately.  When you are talking about relocations of a large number
     of people, especially in the high population areas, that is not easy.
         DR. WALLIS:  It is difficult enough without any nuclear
     accident.
         DR. KHATIB-RAHBAR:  Now, what to use as a consequence
     measure, to continue.  Off-site consequences, fatalities, land
     contamination, et cetera, require level 3 PRA.  This involves even
     additional and greater uncertainties.  In calculating those types of
     weights, those uncertainties are also inherent in there.  Let me not
     mislead you.  Those are there, as you pointed out.
         But you are not -- since you know this is uncertainty, an
     uncertainty that you are going to add on top of other uncertainties, and
     the quantification of these uncertainties are even more difficult -- I
     mean there are studies done, I think some that NRC has recently
     sponsored, some European commission has sponsored, and some we have done
     a number of years ago, and, by and large, the uncertainties and risk are
     going to be dominated, if you want to do a level 3 PRA, by off-site
     consequences.  Weather variability itself is significant.  Population
     variability is very significant.  Emergency action factors are very
     significant.
         So that is a complication I would rather one would not deal
     with.  You have not totally gotten away with it, with the concept that
     we have introduced, but we are not requiring additional calculations and
     additional burdens on somebody to demonstrate.
         DR. WALLIS:  It is interesting, all the things that really
     matter are not in something as simple as CDF.
         DR. KHATIB-RAHBAR:  Release fractions alone, again, are not
     adequate.
         DR. APOSTOLAKIS:  They matter from the uncertainty point of
     view, because if you don't have core damage, I mean --
         DR. FONTANA:  As far as weather goes, the wind direction,
     for example, really isn't an uncertainty, because, in principle, you
     could do the calculation for that.
         DR. KHATIB-RAHBAR:  But rain is, for example.  Again, given
     lot hot spots are driven by rain, how many rains do you put in your
     weather bits, and that is very --
         DR. FONTANA:  That is --
         DR. POWERS:  But I think Mario might argue, maybe you would
     argue, that if an accident were to occur, you would know exactly what
     the wind direction is, and you would know what the -- whether it was
     raining or not.  You would still have the other uncertainties, what the
     magnitude of source term was.
         DR. KHATIB-RAHBAR:  You know that for 20 kilometers.  When
     you are talking these numbers into 2,000 kilometers, that is
     nonsensical, absolutely nonsensical.
         DR. WALLIS:  It makes all the difference in the world.  The
     Chernobyl plume didn't have to go to Scotland.
         DR. KHATIB-RAHBAR:  Actually, they measured high
     concentration of, for example, plutonium near Sweden for Chernobyl.  But
     if you had to use existing models, they would not predict a single -- in
     grams of plutonium, or micrograms of plutonium in Sweden.
         DR. POWERS:  When you say use existing models, models of
     what?
         DR. KHATIB-RAHBAR:  By using the straight line plume models,
     MACCS, CRAC.
         DR. POWERS:  Now, I think if you used -- the trouble with
     those models is that they are really fairly localized models.
         DR. KHATIB-RAHBAR:  Yeah, but they are integrated up to
     1,000 kilometers, though.  They are applied, unfortunately, to very
     large areas.
         DR. POWERS:  When you apply things like AIRREG, do you get
     plutonium in Sweden?
         DR. KHATIB-RAHBAR:  Sure.  You have to update the weather --
     all the way during its trajectory and follow the mountains and all the
     pathways.  I mean if you again look at the AIRTEC calculations, they are
     subject to other uncertainties which, you know, is a ball game in
     itself.  I mean there is -- you cannot avoid that.  I mean that is
     there, but that is hidden somewhere else.  I mean you have to put in
     eddy diffusivity parameters into those codes.
         DR. WALLIS:  I think you want to make sure you don't have a
     hot plume, too, so that it doesn't go up.
         DR. KHATIB-RAHBAR:  It can rise, exactly.  There are
     ebulliency effects you have to consider.  There are many, many factors
     in here which are not straightforward to calculate.  So all I am saying
     is, why burden it with that, if possible?
         DR. KRESS:  If you can incorporate all those uncertainties
     appropriately in the acceptance criteria.
         DR. WALLIS:  So I think what you are saying is that the real
     consequences are dominated by things that are difficult to figure out,
     so let's forget them.
         DR. KHATIB-RAHBAR:  No, I am not forgetting them altogether. 
     I am saying let's not burden the process with going through those types
     of exercises.
         DR. KRESS:  You have to incorporate them somehow in your
     acceptance criteria.
         DR. KHATIB-RAHBAR:  Yes.
         DR. KRESS:  And you have to do it in a bounding sense.
         DR. WALLIS:  But when you are telling the story to the
     people who are going to be effected, you have got to level with them and
     say there are these things which are major actors like emergency
     response, really major actors in determining how many people are going
     to be hurt.
         DR. KHATIB-RAHBAR:  Right.  I mean that's --
         DR. SEALE:  The details of emergency response is
     site-specific.  And I think at this point we are trying to find out what
     the general principles are rather than emphasizing the fact that you get
     a different answer for Palo Verde whether or not the Santa Ana winds are
     blowing towards Southern California.  And I think that is the reason
     that you don't go into these other things at this point, you do have to
     have a site-specific element in it, but I think it is premature.
         DR. KRESS:  That is exactly what I said.  For example, if
     you will excuse me a minute.
         DR. KHATIB-RAHBAR:  Please.
         DR. KRESS:  Now, that Jeff is here, I love to show his
     curves, but when I showed this curve from his work, I intended this line
     that I drew on there to represent reasonable emergency response
     measures.  You could very well, if you are developing an acceptance
     criteria, you could very well go this line, which is basically no
     evacuation and no emergency response at all.  And, of course, that makes
     your acceptance criteria very strict, but you have covered everything
     then.  And I say that is where you incorporate these uncertainties,
     somehow, in this correlation function of your acceptance criteria.
         DR. APOSTOLAKIS:  I suggest that we let Mohsen finish the
     first part and perhaps when we come back, --
         DR. KHATIB-RAHBAR:  Let me finish the first part.
         DR. APOSTOLAKIS:  -- we will have an opportunity to --
         DR. KHATIB-RAHBAR:  Yes, let me finish with this, and then I
     think we should take a break.
         Release fractions alone are not adequate because they do not
     account for reactor size.
         The impact of short-lived radionuclides can be mitigated by
     emergency response measures, so if you look at an attribute which
     actually satisfies nearly all of these, I think release activity,
     Becquerel, or equivalent mass equivalent of cesium-137, in our view
     appears to be a reasonable measure of the consequences.  So that would
     be the response.
         DR. APOSTOLAKIS:  And that would be only one?
         DR. KHATIB-RAHBAR:  No, as I said, I would -- in fact, in
     some of the more recent studies they use release activity, in terms of
     Becquerel, which includes everything.  But the question, it's much more
     difficult to come up with acceptance criteria for Becquerel, perhaps, as
     it will be if you have to use a release quantity.
         DR. KRESS:  But in principle it's about the same.
         DR. KHATIB-RAHBAR:  It's the same thing, really.  But some
     people take easier to the mass quantity rather than Becquerel quantity.
         DR. KRESS:  I would have used curies, myself.
         DR. KHATIB-RAHBAR:  Oh, curies.  Yes.
         DR. POWERS:  I spend a decreasing fraction of my time
     associating with people of the chemical persuasion, and they grow
     uncomfortable with focusing on a single kind of radionuclide, for a lot
     of reasons.  It seems to me that if I were to generalize for them what
     you've said, what you've said is that you really only have to understand
     the particulates coming out of the plant, that we can handle the gases.
         DR. KHATIB-RAHBAR:  No, not necessarily, because cesium is
     not necessarily particulate matter.  I mean, I think -- I mean --
         DR. POWERS:  When is cesium not --
         DR. KHATIB-RAHBAR:  I'm talking about the being volatile,
     particulates in the sense as you look at TIID 14844 where there are
     refractories of that nature, no.  Yes, particulate, it's an aerosol
     matter.  That's correct.  But it's volatile.  It's a volatile aerosol. 
     Volatile aerosol.
         DR. POWERS:  When is it a volatile?
         DR. KHATIB-RAHBAR:  Cesium is a volatile species.  I mean,
     you can evaporate it at a few hundred degrees.
         DR. POWERS:  Yes, and a few hundred degrees is located
     where?
         DR. KHATIB-RAHBAR:  In the atmosphere it goes and condenses
     into aerosols.
         DR. POWERS:  I think it's an aerosol shortly after emerging
     from a core.
         DR. KHATIB-RAHBAR:  Right.  It condenses.  It approaches
     temperatures more than --
         DR. POWERS:  It bears no resemblance to a volatile species
     once it moves away from a hot -- and that if -- if there were other
     radionuclides that behaved as badly as cesium, and there are certainly
     people that have advocated that we've omitted one, it wouldn't change
     your point at all if you said "particulate" instead of "cesium."
         DR. KHATIB-RAHBAR:  Okay.  Excluding noble gases,
     essentially.
         DR. POWERS:  What you're excluding is both noble gases and
     the gaseous forms of iodine, whether it's I2 or methyl iodide.
         DR. KHATIB-RAHBAR:  Yes, but as I said, don't -- in the
     analysis we have done, we have credited iodines.  We have not recognized
     whether iodine is as a gaseous form or it's a particulate form per se. 
     We have like a 2-percent fraction being in a gaseous form.  So there is
     some contribution from gaseous iodine in there as well.  But it does
     credit iodine.  It does have some contribution from gaseous iodines.
         All I'm saying is that we should not include overwhelmingly
     iodine and/or noble gases.  We should dominate early fatalities.  We
     should focus on latent fatalities and contribution from iodine, whether
     it's gaseous or particulate, is smaller than what it would contribute to
     early fatalities.
         I don't know what I said on particulates per se.
         DR. POWERS:  Well, I guess I'm a little confused with what
     you've said.  You have a line here that says reactivity of cesium-137
     appears to be a reasonable measure of consequences, yet in your own
     studies you've added in iodine.
         DR. KHATIB-RAHBAR:  Cesium equivalent.  Cesium-137
     equivalent is what we're using always.  Did I have just cesium-137.
         DR. POWERS:  You have mass equivalent.  You have release
     activity or mass equivalent.
         DR. KHATIB-RAHBAR:  Mass equivalent.  That's -- mass
     equivalent, sure.
         DR. POWERS:  Okay.  So the equivalent applies to both mass
     and activity.  I do understand that.
         DR. KHATIB-RAHBAR:  Absolutely.  Absolutely.
         DR. POWERS:  Yes.
         DR. KRESS:  When you look at the safety goals, prompt and
     latent fatalities, which I view as something like acceptance criteria,
     prompt fatalities dominate.  Have we chosen the wrong safety goals? 
     Because we could just show -- forget the latent fatalities.  The prompt
     fatality safety goal will dominate almost all sides.  Maybe once in a
     while they don't.  It depends on --
         DR. POWERS:  That's only because of the way you use your
     consequence -- your transport modeling.  If you didn't do that, if you
     didn't narrow the plume down artificially, then you would find that the
     latents dominated.
         DR. KRESS:  Maybe.  It depends on population distribution.
         DR. POWERS:  It clearly does, but the thing that you can
     affect when you integrate overall sites, the thing most at your command
     to affect is how narrow you make that plume.  And at least what we find
     from the uncertainty studies that the NRC has sponsored is we may be
     artificially -- may have artificially narrowed that plume in ways we
     shouldn't have, and we're deriving a whole lot, a bunch, out of
     intuition we gain from those artificially narrowed plumes.
         DR. KRESS:  You could very well have a good point.
         DR. WALLIS:  Well, the message I'm getting is that the
     latents dominate.  When you're talking about hundreds of thousands of
     latents, it's very difficult to get 100,000 prompt fatalities.  You have
     to work pretty hard to do that.
         DR. POWERS:  I think he was talking about in the case of --
         DR. KRESS:  I was talking about the safety goals which are
     individual risk.  You're talking about totals.
         DR. WALLIS:  That's silly.
         DR. KRESS:  Well, that's right, that may be another
     silliness we ought to --
         DR. WALLIS:  That's the lunatic who stands around the plant
     during an accident.
         DR. KRESS:  No, it's not him.
         DR. WALLIS:  Says come get me.
         DR. KRESS:  No, it's an average over the population.  But
     that may be another thing about acceptance criteria you need to look at. 
     Total deaths may be a better thing to look at.
         DR. POWERS:  It is certainly my understanding that when we
     look at prongs for accidents that progress slowly enough that the
     evacuation plans can be implemented, that they are dominated by the
     inefficiency of those evacuation plans.  Undoubtedly there is a certain
     number of lunatics, but mostly, you know, the conventional wisdom, 10
     percent just never get the word, you know.  I think that's more of a
     concern than lunatics.
         DR. APOSTOLAKIS:  Okay.  Maybe we can take a break now.  By
     the way, does the staff plan to say anything or just participate in the
     discussions?  Do you want some time for presentation, Gary?
         MR. HOLAHAN:  I don't believe we do; no.
         DR. APOSTOLAKIS:  Okay.  So just general discussion.
         And, Jeff, I understand you will do the same.
         MR. KAISER:  I don't have a presentation.
         DR. APOSTOLAKIS:  Just participate.  Okay.  So we'll
     continue with Mohsen's presentation at 10 minutes after three.
         [Recess.]
         DR. APOSTOLAKIS:  Well, let's reconvene.  We are on the
     record now.
         Before we start maybe I should say a few words about our
     other invited expert who joined us a little earlier.  Jeff Kaiser is a
     Vice President at SAIC.  He is managing the process industry programs. 
     He has a Ph.D. in Elementary Particle Physics from the UK.
         The reason why he is here is because in his past he has done
     a lot of work on Level 2 and Level 3 PRAs.  In fact, Dr. Kress used some
     of your curves -- or not curves, points, I guess, figures.
         DR. KRESS:  Figures.
         DR. POWERS:  Scattered ions.
         DR. APOSTOLAKIS:  Scattered ions.  Okay -- and Jeff has also
     worked on the development of atmospheric dispersion and consequence
     models, both for nuclear and the chemical industries, so thank you very
     much for coming.  Jeff, we appreciate your willingness to come.
         So -- we go back to Mohsen now, to his second part.
         DR. KHATIB-RAHBAR:  Okay.  There are a number of other
     issues that were raised in communication with Dr. Apostolakis.  One was
     are the accidents outside of the core considered.
         In the studies we have done, we have looked at shutdown and
     other modes of operations but in terms of frequency-consequence curves,
     we have never, at least I have never seen any shutdown studies that have
     gone beyond calculating the fuel damage frequency -- in other words,
     releases have never been computed, to my knowledge.
         Therefore, the answer to this is no.  They have not been
     considered.
         Should they be included?  Should, in other words,
     frequency-consequence curves include other than full power operational
     modes or operational modes, shutdowns, refuelling, et cetera?  And the
     answer to that is yes, a qualified yes.  By qualified I mean there are a
     number of other issues one has to deal with when you are going to
     non-power operational modes, but I think by and large the uncertainties
     associated with those other issues are not in my view any more
     significant than what we are dealing with currently.
         There are other issues, for example, issues such as air
     ingress can cause revolatilization of otherwise non-volatile fission
     products, lithium being one of them, but again I think that they could
     be dealt with.  They are not any more difficult at least in my mind than
     full power operational mode.
         DR. KRESS:  Let me ask that question a little differently. 
     Are PRAs configured in such a way that you can actually do a shutdown
     risk that is -- evaluation?
         DR. KHATIB-RAHBAR:  They have done them, yes.  They have
     been --
         DR. KRESS:  They look at any sequence or any plant
     configuration --
         DR. KHATIB-RAHBAR:  Yes.  I'll give you an example.  We have
     look at one kind of plant in Europe.  They have looked at 35 different
     configurations -- all the configurations they could think of.
         DR. KRESS:  Do you ascribe a certain amount of time to which
     those configurations are in --
         DR. KHATIB-RAHBAR:  They are all in different timeframes,
     exactly.
         DR. KRESS:  And is there variation of that timeframe for --
         DR. KHATIB-RAHBAR:  Yes, yes.
         DR. KRESS:  -- for Monte Carlo on that?
         DR. KHATIB-RAHBAR:  No, no.  It's not Monte Carlo on that
     but what you have to do is -- that's what I mean it's a qualified yes --
     a number of issues have to be considered.  How do you calculate these
     releases?  How do you integrate them, over what period of time, et
     cetera, but I don't think -- You know, I have not done it but I do not
     think they are insurmountable.  That's something that you can deal with.
         DR. KRESS:  Well, one shutdown's -- well, something like
     another one but not exactly, and you have to have that variation in
     there some way.
         DR. KHATIB-RAHBAR:  Absolutely.  Every plant looks
     different.  It's very procedural driven.
         DR. KRESS:  And you can't do it for past shutdowns.  You
     have to do it for future shutdowns so it almost has to be a Monte Carlo
     simulation of some kind, it seems to me like -- if you are going to do a
     proper shutdown risk assessment.
         DR. KHATIB-RAHBAR:  Monte Carlo simulation -- maybe I
     don't --
         DR. KRESS:  Monte Carlo simulation of plant configuration
     states and times at the states.
         DR. KHATIB-RAHBAR:  But those are known.  You know the time
     you have different shutdown states, because you do them very often. 
     They are not uncertain in a sense.
         DR. KRESS:  Yes, but it matters -- it matters whether the
     state is -- say, this piece of equipment is out now and another one is
     out now, whether they overlap or whether the one precedes the other.  Is
     that the sort of thing you think you can handle without --
         DR. KHATIB-RAHBAR:  They are all considered, absolutely.
         DR. KRESS:  -- without a Monte Carlo?
         DR. KHATIB-RAHBAR:  Without a Monte Carlo they are
     considered.  Maybe I am missing, maybe I am not answering your
     question -- I don't understand the question properly, but I think those
     are configural issues.
         DR. KRESS:  I just think it is very difficult to use the PRA
     event tree methodology as is to do a shutdown risk assessment, that's
     all -- that the technology has to be improved.
         DR. POWERS:  But I got the impression the point was that the
     challenges that you bring up, which I think are very real, are simply
     because the order that you do things and what-not is not known in
     advance.
         DR. KRESS:  That's right.  That's part of it.
         DR. POWERS:  The shutdown evolution is not done often enough
     that you can anticipate 99 percent of the time I am going to do this
     first and this second.  There are some things that you can and some
     things that you can't but I thought the speaker's point was none of
     these things are insurmountable and defy the current technologies.
         You can -- we do Monte Carlos all the time with PRAs
     nowadays.  It is a relatively routine thing to do.
         DR. KRESS:  But my point was I thought you needed to do a
     Monte Carlo to do it properly.
         DR. POWERS:  You think it's essential -- arguable.  We have
     had speakers come before this group and within this group that have
     argued that there are major challenges with the treatment of human
     performance and success criteria in shutdown PRAs.  I don't think that
     they would argue with the proposition that these are not insurmountable
     things.  They just said they had to be done.
         DR. KHATIB-RAHBAR:  Exactly.  They have to be dealt with,
     absolutely.
         DR. POWERS:  And there may be some substantial
     methodological developments that have to occur here to make is
     relatively routine to do shutdown PRAs, routine in the same sense that
     it is routine now to do an operational PRA.
         I think that the "Monte Carloness" that might be required
     that you talked about becomes feasible simply because our computing
     power is going up so much.
         DR. KRESS:  I think it is definitely feasible.  I just don't
     think that PRAs are configured in a way to do it.
         DR. POWERS:  I don't think that people -- well, let's say up
     until October I didn't think people had looked at that.  Now we are
     evidence that people have looked at doing Monte Carloness on PRAs for
     other purposes, but they have implemented it, so it seems to be doable.
         This question of success criteria I think does come up,
     because I think that the success criteria for shutdown accidents are
     different than they are for real accidents when the containment barrier
     is not in place.
         I think they are identical if the containment barrier is in
     place, but I think they are not when the containment is open.
         DR. WALLIS:  Now Chernobyl did not occur in a power
     operational mode.  It wasn't really a shutdown either -- tinkering mode. 
     That maybe is the most dangerous where something, an unusual mode is
     entered.
         DR. KHATIB-RAHBAR:  Yes, there are different modes and of
     course there are methodological issues which do come into play but the
     issue is you have to balance your uncertainties.  Are the uncertainties
     going to be driven by lack of data or are the uncertainties going to be
     driven by modelling these types of complex overlaps that you refer to,
     et cetera.
         But what I am focusing here again is mostly on phenomena,
     things like success criteria that Dana put his finger on.
         DR. KRESS:  Or on your first bullet -- part of the
     motivation for that question was spent fuel pools.
         DR. KHATIB-RAHBAR:  Yes.
         DR. KRESS:  And things like resins and waste products that
     release small amounts of radioactivity which normally aren't in PRAs but
     might be incorporated into FC curves.  They could be and I think part of
     that question is are those, is it feasible to incorporate those and have
     they ever been incorporated in FC curves?
         DR. KHATIB-RAHBAR:  To my knowledge they have not been
     incorporated.  I have not seen one, but I think it is feasible and it
     can be done.
         DR. KRESS:  Particularly the spent fuel pools.
         DR. KHATIB-RAHBAR:  Absolutely.  The spent fuel pool has
     been looked at in a large number of studies and that is some in fact we
     know most about in at least in lightwater reactor business -- at least
     in Europe I know they have looked at them.  In the U.S.  I don't know,
     except for a couple I know of, but I don't know that many have been
     done.
         Here again the phenomena of the uncertainties -- there are a
     few issues which do come into play but I contend they are not any more
     difficult than the issues we already are supposed to deal with.  By and
     large the containment is open oftentimes.  We don't have the same
     containment loading challenges that we have to deal with typically. 
     Oftentimes the reactor pressure vessel head is in fact removed, but
     there are other processes.  The air factor is in there, natural
     convection processes come into play, et cetera, that you have to deal
     with, but I think given the uncertainties that we are currently putting
     on these risk curves, these will not be any larger than what we are
     currently dealing with.
         Now let's come to the core of what I am here for.  It's a
     paper which we presented several years ago at an agency-IEA meeting
     which dealt with the proposed Swiss risk criteria, safety criteria.
         These criteria is composed of two parts, one for existing
     power plants and then for the newer generation plants -- if there ever
     will be a new power plant built in Europe.
         The core damage frequency including all external events --
     this again at that time we did not think of shutdown modes or other
     modes of operation.  This was only power operational mode but including
     all external events is to be less than equal to 10 to the minus 5 per
     reactor year.
         Frequency release is greater than 10 kilogram equivalent of
     Cesium-137, gram equivalent again -- I would like to underline that.  It
     should be much less than 10 to the minus 6 per reactor year.
         This was put into a exceedence frequency consequence curve
     of the nature we have been discussing this afternoon, which is of this
     format.  This is the total core damage frequency of 10 to the minus 5.
         I'll give you a little background how we selected these
     breakpoints in the curve.  That might be of interest to you.  This
     number here, this release corresponds to approximately the maximum
     release you would get for an equivalent Swiss PRA or BWR during a severe
     accident when filter containment venting system was actuated.  All the
     Swiss plants are required to have -- have been backfitted with a filter
     containment venting system, and for a boiling water reactor you have an
     additional filter on top of the suppression pool.  These are typically
     Venturi type filters.
         So this corresponds to releases which are by design
     required.  In other words, with these filter containment venting systems
     filters are supposed to perform to certain criteria, and this is
     essentially like a design criteria roughly.
         DR. KRESS:  I have a little trouble following a -- this is
     what I would call an acceptance criteria that is derived on the basis of
     what you could possibly get out of the reactor.  That seemed to be the
     wrong direction to approach an acceptance criteria.
         I don't care what you can get out.  I want to know what is
     acceptable for some other reason, and perhaps the criteria that went
     into the general design basis for the filter had that incorporated in it
     so it had to be this good because this is what we accept, and then I
     would say okay, that is all right, because you got it by a round-about
     way.
         But it ought to be stated in such a way that this is what is
     acceptable for some reason or other.
         DR. KHATIB-RAHBAR:  I understand where you are coming from. 
     I agree with you, but I think the discussions on how to arrive at
     this -- how we looked at what is acceptable from a societal point of
     view -- you know, what about an engineering point of view?  Can we
     design a system in which we have a high degree of confidence that give
     you releases of decontamination factors of that order, et cetera, et
     cetera.  I agree -- that should not have been the basis for selecting
     this type of a criteria but I am just telling you how this whole thing
     evolved.
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  Whether I agree with it or not is
     another issue.
         DR. APOSTOLAKIS:  But the 10 to the minus 5 though in the
     frequency does what Tom says.
         DR. KRESS:  Yes.
         DR. APOSTOLAKIS:  That is what society in Switzerland felt
     was acceptable.
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  But I guess what Tom is saying is this
     number --
         DR. SHACK:  That's an ALARA concept there.
         DR. WALLIS:  Right.  I also sense it's a flat curve.  You
     are saying that all these accidents are at the maximum because there is
     nothing below that to make it keep going up.
         DR. KHATIB-RAHBAR:  All it says is that most of the
     accidents are --
         DR. WALLIS:  Are at that point.
         DR. KHATIB-RAHBAR:  -- of this point, exactly.
         DR. WALLIS:  Most of the accidents, right.
         DR. KHATIB-RAHBAR:  In fact, yes.  I mean the previous
     curves showed that most of the severe accidents are mitigated by the
     containment.
         DR. KRESS:  Look -- point of question.  Suppose -- now this
     is an acceptance curve.  Now suppose an actual reactor plant
     configuration comes in so that it started at below 10 to the minus 5 and
     sort of paralleled that curve, but went through the line up there about
     10 to the minus 1 and then dipped back down.  Is that an unacceptable --
     I mean how does one decide?
         DR. KHATIB-RAHBAR:  No, that's a good point.  No, I know
     where you are coming from.  I agree, but again you have to recognize
     that all of these criteria are developed with something in mind.
         If you go back to the U.S. NRC safety goals, look at
     WASH-1400.  It is basically what the technology allowed and you could
     calculate it in WASH-1400 in 1985.
         DR. KRESS:  There is a basis of truth in that, yes.
         DR. KHATIB-RAHBAR:  I mean I don't know if that is exactly
     what happened but if you go and take WASH-1400 and put the numbers on a
     curve, that is what you come up with, and I think you, I agree with you
     in an ideal world you should not keep that in mind but you look at
     something which is achievable.
         If I set the safety criteria at the level where I could not
     demonstrate it and I could not achieve it, it's worthless.  In fact,
     this part of the curve --
         DR. KRESS:  It's not worthless.  It is not worthless if it
     is a real true acceptance criteria.  You would shut everything down --
     you don't do it.
         DR. KHATIB-RAHBAR:  I mean demonstrating a criteria at this
     level is impossible.
         DR. KRESS:  Oh, I understand that.  That is a different
     issue, yes.  I agree with that, because we are having that same trouble
     in 50.59, yes.  You can't go beyond your tools for deciding these
     things.
         DR. KHATIB-RAHBAR:  Right, and then this level here, which
     is 10 kilograms release, this corresponds to approximately 6 percent
     inventory for a large power reactor.  This again is typically what you
     would get for early containment failures, steam explosions, direct
     containment heating, type failure.
         We said any releases greater than this should have a
     frequency much less than 10 to the minus 6.
         DR. KRESS:  Based on what?
         DR. KHATIB-RAHBAR:  The idea was that --
         DR. KRESS:  That is a large release and 10 to the minus 6 is
     a magic number for large releases?
         DR. KHATIB-RAHBAR:  This was based again on the same
     philosophy.  No question about it.
         DR. KRESS:  Okay.
         DR. KHATIB-RAHBAR:  Okay?  Somewhat ad hoc.
         DR. KRESS:  I have a bit of a problem with that kind of
     acceptance criteria, although, you know, it's very practical.
         DR. KHATIB-RAHBAR:  But again you have got to look at the
     history.  If you look at the IAAs, the INSAC standards, it somehow
     follows the INSAC.
         DR. KRESS:  Yes, I know there is a history behind that.
         DR. KHATIB-RAHBAR:  Behind this whole process, so the idea
     was not to really re-invent the numbers here, but try to say how we --
         DR. KRESS:  To say that they should be consistent.
         DR. KHATIB-RAHBAR:  Exactly.
         DR. KRESS:  Be consistent with values of --
         DR. KHATIB-RAHBAR:  Precisely.
         DR. KRESS:  -- that are deemed acceptable for some other
     reason.
         DR. KHATIB-RAHBAR:  Precisely.
         DR. KRESS:  I understand.
         DR. KHATIB-RAHBAR:  We did not want to re-invent the wheel
     here, so this has been set at 10 to the minus 5, an acceptable number --
         DR. MILLER:  10 to the minus 6 --
         DR. KHATIB-RAHBAR:  10 to the minus 6 is being a large
     release.  In a sense we define a large release, okay?  This corresponds
     to something on the order of 10 percent cesium or iodine.
         DR. KRESS:  Now probably this could not be at all consistent
     with the U.S. safety goals.  It may be more severe because it looks like
     it is.
         DR. KHATIB-RAHBAR:  It is more severe.
         DR. KRESS:  Yes, it looks like it is much more severe.
         DR. KHATIB-RAHBAR:  It is a lot more limiting.
         DR. KRESS:  Yes, a lot more limiting.
         DR. KHATIB-RAHBAR:  This in fact -- I will show that later. 
     It is very limiting.
         DR. KRESS:  So it is like almost redefining the safety
     goals.
         DR. KHATIB-RAHBAR:  Much more restrictive safety goals.
         DR. KRESS:  Did you intend this for new plants or --
         DR. APOSTOLAKIS:  Let me understand what that means.  If you
     calculated a LERF at an acceptable level --
         DR. KRESS:  It would be a lot more than 10 to the minus 5.
         DR. WALLIS:  Where is LERF for this curve?
         DR. APOSTOLAKIS:  LERF is an average value.  It is not here,
     right?  It's the average of the larger release -- but, no, it could be
     here?
         DR. KHATIB-RAHBAR:  Basically this I would say -- it's about
     here.
         DR. KRESS:  But LERF is basically a conditional early
     containment failure, as we defined it, but if you wanted to know what
     the prompt fatality value, individual risk of prompt fatality was from
     this curve, which this LERF is supposed to be surrogate for, it -- you
     would have to integrate this curve with this functional relationship for
     the prompt fatality versus fission product release that Jeff's curve has
     inherited.
         It's a convoluted interval between the two.  So it's not
     obvious just looking at it just to know that functional relationship is
     such that this level, 10 to the minus 5 going down at this point of
     fission products, you know that's going to be well below 10 to the minus
     5.
         DR. KHATIB-RAHBAR:  So in other words, about 10 percent
     inventory of cesium with iodine for a large power reactor would
     approximately correspond to one early fatality, roughly.
         DR. WALLIS:  So where is LERF?  Do you have --
         DR. KHATIB-RAHBAR:  So if you take that as being a
     definition of LERF, I don't know what it is in ACRS's mind, this is
     about it.
         DR. APOSTOLAKIS:  Well, but Tom, based on what Mohsen just
     said, I'm not sure that your conclusion was correct.  Or maybe -- he
     says, Mohsen says, that LERF is about where you have the second knee. 
     So it's 10 to the minus 6 for a release of 10 to the 4th or greater. 
     Like that is indeed correct.  In the U.S. goals, I am not saying
     anything about the straight line that he has beyond 10 to the 4th.  We
     don't have that.  We don't have regulations there.
         DR. KHATIB-RAHBAR:  You don't go exactly.
         DR. APOSTOLAKIS:  Right.
         DR. KHATIB-RAHBAR:  This is as far as you go.
         DR. APOSTOLAKIS:  For that part we don't.  We only have that
     point where the knee is.  Now your formulation earlier showed that the
     whole curve affects --
         DR. SHACK:  The question is if you bring that knee low
     enough, does that force everything good enough?
         DR. APOSTOLAKIS:  But wait a minute.  Wait a minute.
         DR. KRESS:  Yes, it definitely could.  Yes.
         DR. APOSTOLAKIS:  No.  I think they're regulating more, but
     I'm not sure whether they are more stringent, the criteria are more
     stringent.
         DR. KRESS:  I guarantee you that's more stringent than 10 to
     the minus 5 LERF.
         DR. KHATIB-RAHBAR:  It's much more stringent.
         DR. APOSTOLAKIS:  Than 10 to the minus 5 what?
         DR. KHATIB-RAHBAR:  LERF.
         DR. APOSTOLAKIS:  Oh, LERF.  Well, yes, it's 10 to the minus
     6.
         DR. KRESS:  All right, that's also much more -- that's also
     more constringent than prompt fatality individual risks go.
         DR. APOSTOLAKIS:  But it is conceivable, though, that in an
     American plant you could have that little piece of the last segment of
     the straight line, in the real assessment now, not the criterion,
     having, you know, larger scope or smaller.
         DR. KHATIB-RAHBAR:  Farther.
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  I'll have some examples for --
         DR. APOSTOLAKIS:  And then that would affect the fatalities,
     would it not?
         DR. KRESS:  Yes.  It's an integral.
         DR. APOSTOLAKIS:  And we are not regulating that.
         DR. KRESS:  Anything that affects the area under the curve.
         DR. APOSTOLAKIS:  Are we regulating that part, Gary?
         MR. HOLAHAN:  Well, I think -- no, nature is regulating the
     total amount of --
         DR. APOSTOLAKIS:  That's correct.
         MR. HOLAHAN:  Cesium, and that's what establishes the end
     point.
         DR. APOSTOLAKIS:  So 10 to the 5th is a total.
         MR. HOLAHAN:  It's total.  Well, you know, more or less. 
     Yes.
         DR. WALLIS:  But the point is LERF is not obvious from an
     American curve.
         DR. KRESS:  No.
         DR. WALLIS:  There isn't a kink like this in it.
         DR. APOSTOLAKIS:  Well, don't use a curve at all.
         DR. WALLIS:  I always have problems knowing when I see an
     F-C curve how to relate it to this LERF which we talk about.
         DR. KRESS:  It's related to the slope of the curve, its
     actual position on there, and convoluted with the health consequences. 
     Just look at it and say it.
         DR. WALLIS:  It's complicated, especially since --
         DR. KRESS:  There is no unique function of this that gives
     you a LERF value.
         DR. WALLIS:  Most of these curves get vertical --
         DR. KRESS:  Yes, it makes it even more --
         DR. WALLIS:  And you don't really know where you are at all.
         DR. KRESS:  It's hard.  But the nice thing about that
     vertical is it's going down to low frequencies in a hurry.
         DR. KHATIB-RAHBAR:  And then this by and large is also an
     extrapolation as well.  That so far is --
         MR. KAISER:  Well, this curve will have to get vertical
     somewhere.
         DR. KRESS:  Yes, it has to.  If it's a cube of the -- it has
     to go to zero.
         DR. WALLIS:  Are you telling us LERF is not very real?
         DR. KRESS:  I mean, there's no way around it.  You're right. 
     All those curves will go to --
         DR. WALLIS:  Are you telling us LERF is not very real?
         DR. APOSTOLAKIS:  The question is how quickly do they go.
         DR. WALLIS:  Is it that part of the curve is not very real?
         DR. KHATIB-RAHBAR:  This part of the curve is not very real.
         DR. APOSTOLAKIS:  And in our regulation we don't specify how
     quickly.
         DR. WALLIS:  And LERF is probably not very real, either.
         DR. KRESS:  It's inherent in our LERF criteria.
         DR. KHATIB-RAHBAR:  I need to know what LERF is in your mind
     before I can say it is or it's not.
         DR. WALLIS:  I don't know.  That's what I'm trying to figure
     out.
         MR. HOLAHAN:  I'm not sure I agree with Dr. Kress, which is
     a dangerous thing to do, I know, but I think LERF is pretty -- the LERF
     definition's pretty close to the knee of that curve.
         DR. KHATIB-RAHBAR:  It is about here?
         MR. HOLAHAN:  Yes.
         DR. KHATIB-RAHBAR:  That's what I thought it was.
         MR. HOLAHAN:  Yes.
         DR. SHACK:  It is the frequency of a given large release.
         MR. HOLAHAN:  Yes.
         DR. SHACK:  It depends on how you define the large release.
         MR. HOLAHAN:  Yes.
         DR. SHACK:  But once you've done that, then you've pinned
     that point.
         MR. HOLAHAN:  And whether it's 6 or 10 or 20 percent of the
     core, it's somewhere near that knee.
         DR. KHATIB-RAHBAR:  Now let me tell you how we arrive at
     this one.  This actually -- this point here is a lot more uncertain. 
     The idea here was that we knew -- demonstrating any numbers -- numbers
     of 10 to the minus 6 are fairly difficult to demonstrate, let alone
     demonstrating numbers of 10 to the minus 8 level.  It's impossible. 
     Okay?  In other words, this is the area where incompleteness of the PRAs
     come into play, other uncertainties come into play, et cetera.
         But this was important to put in here to demonstrate that we
     want to get releases which are fairly large, as low as practically
     possible.  So the meaning in this part of the curve is nothing more than
     qualitative indication that we like to make sure that the releases
     corresponding to anything greater than 10 grams of cesium equivalent is
     as low as practicable.
         DR. APOSTOLAKIS:  So if you did a PRA for a Swiss reactor,
     and I know they have PRAs for all their units, and your F-C curve
     happened to cross that straight line there and go above it, and then you
     realize that, you know, it would take $2 billion to bring it down, you
     would say it's okay.
         DR. KHATIB-RAHBAR:  Yes.  I'll come to that.
         DR. APOSTOLAKIS:  So these are not criteria then.  These are
     --
         DR. KHATIB-RAHBAR:  These are not speed limits.  These are
     not -- these are not --
         DR. APOSTOLAKIS:  So it's proposed safety goals.
         DR. KHATIB-RAHBAR:  These are not going to be interpreted as
     speed limits.
         DR. APOSTOLAKIS:  They are goals.
         DR. KHATIB-RAHBAR:  These are just goals.  These are
     aspiratory targets that you want to meet.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  But how you meet them is another issue. 
     I'll come to that a bit later.
         DR. APOSTOLAKIS:  And this pink line is intended to be used
     with a mean value?
         DR. KHATIB-RAHBAR:  I'll address that to you in a second.
         MR. KAISER:  Can I come back?
         DR. APOSTOLAKIS:  Sure.  Yes.
         MR. KAISER:  Mohsen has said that this criterion is
     considerably stricter than, say, the safety goals are.
         DR. APOSTOLAKIS:  Well, he didn't quite say it that way.
         DR. KHATIB-RAHBAR:  It could be -- yes, it could be more
     strict for some, yes.
         MR. KAISER:  So my question is, was that done with careful
     thought?  You know, you've decided that the criteria that exist
     previously are too lax, and that something more strict is needed.
         DR. KHATIB-RAHBAR:  No, at the time when we started doing
     this, there was really no definition for larger early release.  If you
     go back to U.S. NRC requirements --
         DR. APOSTOLAKIS:  If they're still discussing it, what's
     large early release?
         DR. KHATIB-RAHBAR:  An interpretation of large early release
     has always been subjective in nature.  At the time NUREG-1150 was done,
     this was interpreted as one early fatality.  In fact, the question was
     we'll be back here even without the bad case.  Then later on, people
     started talking about in terms of equivalent release quantity.  Ten
     percent may be a good number.  And the reason we came up with 10
     percent, because that was somewhere lower than bypass scenarios.  It was
     something we could live with.  But there was never a clear indication
     what the large early release criteria in the U.S. was, nor was this for
     European or REA standard.  So when we started talking about this in
     1985, there were no criteria for large early release.  Okay?
         So we came up with something that we thought we could
     support.  In other words, there was some basis for it, you know, the
     basis for this part of the curve, as I told you, is essentially a design
     standard preventing system, and this is the large early release
     criteria.  We said anything greater than this will be total catastrophe. 
     Okay?  Radiologically.  This is bad, as is.  This is fairly bad.
         DR. KRESS:  All right.  Pegging the two points determines
     the slope of that flat part.  What determines the slope of the part
     beyond --
         DR. KHATIB-RAHBAR:  As I said, this was relatively
     arbitrary.
         DR. KRESS:  Just an arbitrary slope.
         DR. KHATIB-RAHBAR:  This was very arbitrary.  This can go as
     high as the entire core inventory if you draw down in frequency.
         DR. APOSTOLAKIS:  You have a risk-aversion factor of two
     there.
         DR. KHATIB-RAHBAR:  Right.  All this part of the curve shows
     --
         DR. APOSTOLAKIS:  Minus 2.  Which is really pretty
     stringent.  In fact, there was a study years, years ago down at UCLA
     that showed that no industry could survive, could be accepted if you had
     an aversion factor of minus 2.
         DR. KRESS:  Yes.  That's one reason I asked, is that risk
     aversion.
         DR. APOSTOLAKIS:  Yes.  Yes.  It's really difficult to --
         DR. KRESS:  Is there some valid reason for choosing that
     slope?
         DR. KHATIB-RAHBAR:  Yes, the idea is that you -- basically I
     think that for future designs you want to prevent -- in particular for
     future designs you want to prevent releases of this nature by design,
     recognizing you could not demonstrate this number on probabilistic
     levels.  So you have to do it deterministically by eliminating by
     design.
         DR. KRESS:  Where is your view -- you're a pretty good PRA
     guy -- where is your view of where that level is that PRA gets -- you're
     really isn't the noise of PRA?  Is it on this frequency curve part?  Is
     it like 10 to the minus 7, do you think?
         DR. KHATIB-RAHBAR:  I would have a difficulty with numbers
     below 10 to the minus 6, to be honest with you.
         DR. KRESS:  See, this I think is a significant point that
     I've tried to make in relations to the 50.59 problem, where, you know,
     if you tried to use PRAs to define what's risk-significant or not, if
     you start getting below those numbers, you're in the noise where PRA are
     basically useless to you.  And that's why I asked you --
         DR. KHATIB-RAHBAR:  Many things matter in this --
         DR. KRESS:  Because I was considering it like 10 to the
     minus 6 to 10 to the minus 7 also.
         DR. KHATIB-RAHBAR:  In fact, that's one reason why this
     line's for one thing, a thick line.
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  These have to be interpreted not as one
     speed limit that you can exceed.  Furthermore, they are also to reflect
     the fact that in particular when you go down in frequency --
         DR. APOSTOLAKIS:  You just scored some points here.
         MR. KAISER:  Can I go back to my question about --
         DR. APOSTOLAKIS:  Sure.
         MR. KAISER:  How this compares with other criteria that are
     out there?  I spent some time looking at the chemical industry, and as
     you know, there are requirements in law in places like the Netherlands
     and the U.K. and other places, and one of the big difficulties is in
     trying to understand whether the criteria that are proposed in these
     different places are really the same or whether there are enormous
     differences between them, because they're couched in very different
     terms.
         And I think one warning or one caution we should bear in
     mind is that it's nice to throw up these curves here, but what if you do
     a curve like that and then you find that it's four orders of magnitude
     more stringent than some other criterion that people have been pretty
     happy with for a long time like an individual risk, for example?  Why
     should you suddenly ratchet these criteria by four orders of magnitude?
         DR. KRESS:  That's really a good question.
         MR. KAISER:  Yes.
         DR. APOSTOLAKIS:  Well, and that question has been of
     concern to this committee.  Core damage frequency, for example, a goal
     of 10 to the minus 4 reactor year, is more stringent than what you would
     derive by working backwards from the individual risk criterion we have
     now and some reasonable numbers for containment performance.
         MR. KAISER:  Right.
         DR. APOSTOLAKIS:  So then this is new policy.  That's what
     you're saying.  It's not just a surrogate anymore.
         MR. KAISER:  Yes, it might be --
         DR. APOSTOLAKIS:  It's new policy.
         MR. KAISER:  It might be, yes.  If you're not careful, you
     might still --
         DR. APOSTOLAKIS:  So you have to be --
         MR. KAISER:  Ratchet everything.
         DR. APOSTOLAKIS:  That's correct.
         MR. KAISER:  And of course the consequences of that to an
     industry can be very large.
         DR. APOSTOLAKIS:  Now the Swiss --
         MR. KAISER:  Yes.
         DR. APOSTOLAKIS:  The way I understand it do not have goals
     at the risk level, individual versus societal.  Do they?
         DR. KHATIB-RAHBAR:  No.  That's correct.  No.
         DR. APOSTOLAKIS:  So this is their criterion.
         DR. KHATIB-RAHBAR:  These are equivalents.  Right.
         DR. APOSTOLAKIS:  This is it.
         DR. KHATIB-RAHBAR:  This is it.
         DR. APOSTOLAKIS:  This is the highest level goal they have.
         DR. SHACK:  Proposed.
         DR. APOSTOLAKIS:  Yes, proposed.
         MR. KAISER:  Proposed.
         DR. KHATIB-RAHBAR:  Now let me address that issue.  This is
     a very good point.  The criteria which have been -- release criteria
     which have been used by different organizations in the U.S. or
     elsewhere, which is safety-health criteria, early-fatality criteria, we
     debated that extensively, and it was decided that there is no interest
     to -- societally to look at the health issues because especially when we
     talk more latent fatalities, land contamination is what matters, okay? 
     They were talking about the fact at Chernobyl, the fact that you have
     done the contamination as much as you could possibly do, so the
     contamination levels are at the level that you cannot go and occupy the
     land in that area or use it for anything for many, many years to come.
         So it was recognized if there was an accident in
     Switzerland, and all the powerplants are within about a 50-kilometer
     area, they're located in one area, if you stand on a hill, there's an
     accident in one place, that entire area around Zurich is going to be
     totally uninhabitable for many, many years to come.  So talking about
     the health criteria or the fatality criteria in my view is not very
     meaningful -- land contamination area which can be so severe for a very
     small country, it doesn't matter anymore --
         DR. FONTANA:  It's defined as uninhabitable because you
     can't grow crops on it or because you can't stand on it?
         DR. KHATIB-RAHBAR:  All of the above.
         DR. KRESS:  You get sick and die.
         DR. KHATIB-RAHBAR:  All of the above.  Essentially all of
     the above.
         DR. FONTANA:  And it kills you somewhere before you normally
     would die.
         DR. KRESS:  Unless you're our age.
         DR. KHATIB-RAHBAR:  It is something socially that you're
     going to -- I mean right now we can probably argue for many years how
     many people will die as a result of Chernobyl, but we know how much land
     we cannot use today.  Society understands that much easier, I think.
         DR. KRESS:  Mohsen, I had on one of my curves a list of
     potential -- I call them regulatory objectives, that included land
     interdiction, injuries, death.  There's no way with that many objectives
     in mind, there's no way you cannot have a set of consistent criteria, I
     think.  Some of them are going to dominate over others.  And in my mind
     and a couple of the other ACRS Members who will remain unnamed, the only
     common metric that we can come up with among those things, believe it or
     not, is dollars.
         And if we set a dollar limit on what we're willing to accept
     for nuclear power, then you could put that same dollar limit on every
     one of those metrics and come up with a value for it, and one of them
     may dominate over -- the value of the acceptance criteria you get for
     that may be the one that sets the stage and dominates, but it seems to
     me like that's the only real consistent way because you have to have a
     common metric if you're going to compare all of these acceptance.  Is
     that --
         DR. KHATIB-RAHBAR:  I understand.  Yes.
         DR. KRESS:  So, you know --
         DR. KHATIB-RAHBAR:  You could turn actually land -- dollars.
         DR. KRESS:  And what you're saying about land interdiction
     for many years.
         DR. KHATIB-RAHBAR:  It's the same thing.  It's the same
     thing.
         DR. KRESS:  It's going to cost a lot of money.
         DR. KHATIB-RAHBAR:  Precisely.  That's all economic --
         DR. KRESS:  That's probably why it would be the one that
     would dominate.
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. APOSTOLAKIS:  Well, you know, that's why --
         DR. POWERS:  I wonder if I could come back to your previous
     viewgraph --
         DR. KHATIB-RAHBAR:  Certainly.
         DR. POWERS:  And even one that you used before then.  But
     maybe I'll just recall --
         DR. KHATIB-RAHBAR:  Which one?  This one?
         DR. POWERS:  Let's start with this one.
         DR. KHATIB-RAHBAR:  Okay.
         DR. POWERS:  You've -- let's turn to your second knee.
         DR. KHATIB-RAHBAR:  This one.  Okay?
         DR. POWERS:  Out there.  If you look at a previous viewgraph
     you used in which we got releases up to that magnitude, they seemed to
     all be associated with some sort of a catastrophic failure or a bypass,
     if you get that.
         DR. KHATIB-RAHBAR:  Um-hum.
         DR. POWERS:  Could you impose the same effect of this curve
     by saying I shall reduce the frequency of those accidents that result in
     early containment failure or bypass below whatever number that you
     picked here?  That would have the same effect?
         DR. KHATIB-RAHBAR:  In other words, this is a risk factor
     you're talking about.
         DR. POWERS:  Yes.
         DR. KHATIB-RAHBAR:  You either reduce the consequence or the
     frequency or both.
         DR. POWERS:  Um-hum.
         DR. KHATIB-RAHBAR:  Absolutely.  You can do either one to
     satisfy the same end.
         DR. POWERS:  There is no concern on your part that there are
     accidents other than early containment failure or bypass that could give
     you numbers like this.
         DR. KHATIB-RAHBAR:  There could be some shutdown sequences. 
     Some of them could lead to that.
         DR. POWERS:  Let's leave out shutdown.  Let's -- that's only
     -- that's only --
         DR. KHATIB-RAHBAR:  This curve does not recognize particular
     accidents --
         DR. POWERS:  I understand that, but what I'm asking is,
     could we replace this curve by having essentially a containment failure
     criterion, that is, if you take bypass as a failure of one part of the
     containment, you just simply don't allow --
         DR. KHATIB-RAHBAR:  Um-hum.
         DR. POWERS:  The frequency of containment failure to exceed
     some limit.
         DR. KHATIB-RAHBAR:  You can, but there's a problem with
     that, because when I'm putting -- venting in there I'm already failing
     containment.  I don't like to -- do not particularly like the
     containment failure criteria, because what matters is what you release
     into the environment.  Those are subsidiary criteria you can derive from
     this, no question about it, but because venting is one mode of
     containment failure, leakage is one mode of containment failure, I would
     prefer, because there is always a conditional probability of 1 that you
     have failed your containment, by not failing it, you have a certain
     release, design-basis release.
         DR. APOSTOLAKIS:  So if we go back to your viewgraph with
     the release categories --
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  If you can pull that up.  If you do what
     Dana just suggested, then you are just dealing with the frequency of the
     categories on the vertical axis, and you are ignoring your bars.  Is
     that correct?
         DR. KHATIB-RAHBAR:  I don't know.
         DR. APOSTOLAKIS:  Let's go back to that.
         DR. KHATIB-RAHBAR:  Maybe Dana can explain that himself.  I
     don't understand it.
         This one, George?
         DR. APOSTOLAKIS:  Yes.  Yes.  See, on the right --
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  That's why I asked earlier.  Are these the
     accident progression bins?  On the right, the RC1, 2, 3.
         DR. KHATIB-RAHBAR:  Yes.  Right.  Right.
         DR. APOSTOLAKIS:  These are the accident progression bins. 
     You have -- in some of them you have failed the containment.  Right?
         DR. KHATIB-RAHBAR:  Yes.  These are all the ones with
     containment -- yes.
         DR. APOSTOLAKIS:  Okay.  Now if I do what Dana said, and
     deal only with the containment frequency -- failure frequency -- then
     I'm saying I will take all the RC from 1 to 21 and I will make sure that
     their frequency is less than a number.
         DR. KHATIB-RAHBAR:  Is that what you mean, Dana?
         DR. APOSTOLAKIS:  Well, that's a containment.
         DR. POWERS:  Yes.
         DR. APOSTOLAKIS:  So the information then I'm not using is
     the information you have in the figure.
         DR. KHATIB-RAHBAR:  On the quantity released.
         DR. APOSTOLAKIS:  The quantity of release.
         DR. KHATIB-RAHBAR:  Yes.
         DR. APOSTOLAKIS:  That's the difference.
         DR. KHATIB-RAHBAR:  Right.  But you could do that.  I mean,
     there is nothing which prevents you.  But I said the reason I personally
     do not like it, because you're restricting it.
         DR. APOSTOLAKIS:  So by using the curve that you showed us,
     you are using all the information --
         DR. KHATIB-RAHBAR:  Precisely.  It's all there.  The
     conditional probability could be derived from that if you wanted to.  We
     could extract that information for you.
         DR. POWERS:  Okay.  What we conclude from this discussion is
     that by restricting the frequency of -- the conditional frequency of
     containment failure we're being somewhat more conservative than when we
     use the curve.
         Now I have learned something very important.
         DR. WALLIS:  Well, surely you could have a containment
     failure with no core damage.  You could have a LOCA that let's steam out
     but doesn't let any fission products out.
         DR. KRESS:  I don't know of any reactors that would do that. 
     The design criteria is such that they won't do that.
         DR. SHACK:  It still seems to me that if you do that or you
     pin the LERF, the real question is whether that curve can have some
     funny shape out at the high end so that, you know, I mean, by pinning
     two points on a curve, you clearly don't define the whole curve.
         DR. KRESS:  You don't define the curve.
         DR. SHACK:  But in reality are all the curves of such shape
     that by pinning those two points, you've essentially done everything
     that you've needed to do.  I mean, is it really -- are there -- if you
     pin that point, are there changes that you would allow people to make
     that would suddenly tail that end of the curve way out?  In other words
     --
         DR. FONTANA:  A good example, Bill, was -- let's say for --
     well, 10 to the 5th is all the cesium there is, okay?  So that puts
     that.  You're talking about PRAs, you're going to be talking about 10 to
     the 8th, minus 8th, hyphen 10, doesn't make any difference.  So there's
     no point fooling around below that point.
         DR. APOSTOLAKIS:
         DR. KHATIB-RAHBAR:  You could do that; sure.
         DR. APOSTOLAKIS:  Well, that's a containment --
         If you don't like 10 to the minus 4 and 10 to the minus 6
     and you want to reduce that, then you have got to put a curve on the
     other part of that curve.  You can't have a straight line anymore
     because you are reducing that allowance unnecessarily I think.
         Your two point argument is for a straight line but if you
     reduce that point too far you are going to have to put a curve on it.
         DR. SHACK:  No, I'm saying all the curves have a certain
     shape and so when you pin them by two points you have essentially fixed
     the curve.
         DR. KRESS:  You know, the reason all these curves have a
     shape is because they are cumulative -- the complementary distribution
     function -- and all it says is most probability density functions for
     fission product release look a lot alike for the plants.
         DR. SHACK:  But what am I gaining for looking at the whole
     curve over the two points?
         DR. APOSTOLAKIS:  In other words, Mohsen, if I take the CDF
     and the LERF, am I capturing 99 percent of the information you have
     there so I don't have to worry about the actual shape of the curve, of
     the criteria?
         DR. KHATIB-RAHBAR:  I think you are, but the way we applied
     this curve, and I will show it later on, is important because we want
     to --
         DR. APOSTOLAKIS:  The management part.
         DR. KHATIB-RAHBAR:  -- to focus on when you start putting
     certain actions to try to meet this criteria where are you getting the
     benefits?
         When you only focus on large early releases, you only focus
     on one part of the problem, not the entire problem.
         DR. KRESS:  I think that is a good point.
         DR. APOSTOLAKIS:  So maybe we should let him go on and --
         DR. KRESS:  Well, I wanted to make one other point.
         When you fix this curve, you in essence have fixed the PDF
     and what you are saying -- I am putting a regulation, a speed limit on
     PDF.  I think that is a difficult thing to do because PDF is sort of
     inherent in the nature of the beast.
         My point earlier was if you do have a set of criteria, set
     of objectives that I listed on mine, what one could do is say, all
     right, let's take the curve that you get from a plant -- is it
     acceptable with respect to all of those, and what one would have to do
     is do this integral with respect to each one of these functional
     relationships that I got from Kaiser's exercise, and then convert that
     integral, which is a risk consequence, convert that into a dollar cost
     and you add up the dollar cost of all of them and then you have an
     acceptance level on the total dollar cost, and if you exceed that you
     have exceeded your speed limit and if you don't, you're acceptable.
         That was my concept that I was trying to get across and I am
     not sure I did.
         You only do that -- I mean you can't do that with individual
     little curves.
         You have to do it for all of them.  You have to have all of
     these functional relationships but the only common way that --
         DR. SHACK:  But you are doing that to set the acceptance
     value then.
         DR. KRESS:  I set the acceptance value independent of all
     that.  This is to see if you meet the acceptance value and it is the sum
     of the costs of all of them, and then my acceptance value is a dollar
     cost, and it may be the one part of this when one of these things
     dominate that dollar cost and the others add a little bit but -- and
     then it probably would be the land interdiction would dominate the
     dollar cost mostly, and if you -- except if you set the curve so that it
     has to meet that one, probably the other ones don't add a whole lot but
     they add some, and it would be a total dollar cost would be my
     acceptance limit.
         This was my concept on how to do a truly risk-informed
     regulatory system incorporating all of the NRC's regulatory objectives,
     and then you have to worry about how you do it, how you do it with
     defense-in-depth and how you incorporate other things, so there are
     different beasts you have to worry about.
         DR. APOSTOLAKIS:  Let me suggest something because some of
     this stuff is interesting and I want to discuss it in detail but why
     don't we limit our questions to questions of clarification and let
     Mohsen finish, and then open up the discussion on more general subjects
     because what he just said, for example, that the curves are important
     when you consider risk management is something that we should see first
     before we talk about the value of the curves and their equivalents with
     LERF and CDF, so I suggest we limit our questions to those of
     clarification and let you finish, and then we open it up to anything
     that --
         DR. KHATIB-RAHBAR:  It is up to you.
         DR. APOSTOLAKIS:  Yes.  Okay.
         DR. KHATIB-RAHBAR:  For existing operating reactors, it is
     proposed that to conform to the criteria at the mean level of exceedence
     frequency through cost benefit backfits, how will utilities --
         DR. WALLIS:  Clarification.  When you say cost benefit, that
     means that you have to put a price on an incremental CDF or LERF or
     whatever you are measuring, got to put a dollar price, otherwise you
     can't do a cost benefit analysis.
         DR. KHATIB-RAHBAR:  That is a totally --
         DR. WALLIS:  So you must put a price on these safety
     measures.  Thank you.
         DR. KHATIB-RAHBAR:  That has not yet been actually -- we
     debated that extensively.  We have not yet addressed that issue.
         DR. WALLIS:  Otherwise it is just invoking cost benefit --
         DR. KHATIB-RAHBAR:  Absolutely.
         DR. WALLIS:  -- as a phantom.
         DR. KHATIB-RAHBAR:  Absolutely.  That in itself is a very
     complex issue.
         DR. SHACK:  But it gives you a way to look at relative
     measures, even if you don't know exactly what dollar to put on it if one
     gives you more bang for the buck.
         DR. WALLIS:  I don't buy anything when I don't know exactly
     what dollar.  There's a pricetag.  You don't buy a car at the fuzzy
     price --
         DR. FONTANA:  Yes, you do.
         DR. WALLIS:  When you write your check, you don't write a
     fuzzy check.
         DR. APOSTOLAKIS:  You don't use thick lines.
         DR. KHATIB-RAHBAR:  One major emphasis of this whole
     criteria development was that to try to encourage utilities to reduce
     the uncertainties, to really circumvent the uncertainties through
     procedure and/or low cost modifications.  In other words, instead of
     spending millions of dollars trying to say what the uncertainties are,
     you have to do things that do not get you in a situation where these
     uncertainties become important.
         DR. WALLIS:  Aha.
         DR. KHATIB-RAHBAR:  That is an important thing to keep in
     mind.  I will address that a little bit later -- what I mean and how
     that could be achieved.
         So in terms of a procedural thing, this is how this whole
     thing is perceived.  If you come to the top here, if the conditional
     complementary distribution function at 95 percent level, it exceeds the
     proposed safety criteria --
         DR. KRESS:  Let me ask you about that.  How did you arrive
     at this 95 percentile number?
         DR. KHATIB-RAHBAR:  The idea here was to try to relate
     something which is relatively conservative --
         DR. KRESS:  You are operating only on the uncertainty with
     this?  There's two degrees -- you could operate on the mean value and
     the uncertainty.
         These CCDS tend to be, the PDF its derived from tend to be
     log-normal and the 95 percentile means the best you can do was about
     four times the mean, which seems like that is almost negligible in PRA
     space.
         I am concerned about using a 95 percentile of an existing
     curve as a -- I don't consider that a different, really a different
     level of safety -- hardly -- in PRA space.
         DR. KHATIB-RAHBAR:  All I am saying is the mean is a good
     indicator of uncertainty really for these things.  When you are talking
     about numbers between zero to one, your mean is fairly high.  You are
     close to the upper bound of the uncertainty.  For all the fission
     product releases you are going from 10 to the minus 5 to one so I don't
     care what kind of distribution you put on that, your mean is going to be
     fairly high.  It's a very good indicator of the uncertainty that you
     have in your process.
         All this thing is saying is that I am putting in a slightly
     higher confidence on the number I am using.  That's all.
         DR. APOSTOLAKIS:  Well, actually, what it says is later.  It
     says that if both your mean and the 95th are above --
         DR. KHATIB-RAHBAR:  Then you are okay.
         DR. APOSTOLAKIS:  -- then -- no, above.
         DR. KHATIB-RAHBAR:  Above -- oh, I'm sorry.
         DR. APOSTOLAKIS:  Then you consider backfits and so on.
         DR. KHATIB-RAHBAR:  Other things.
         DR. APOSTOLAKIS:  Otherwise you consider cheaper things,
     procedural changes --
         DR. KHATIB-RAHBAR:  The one idea here is to use this
     process --
         DR. APOSTOLAKIS:  That is what he is saying.
         DR. KHATIB-RAHBAR:  -- to encourage either reducing the
     uncertainties through modifications, procedural changes, et cetera, so
     that you can get away from this whole large uncertainty issue.
         DR. KRESS:  Well, you know, what I was looking at with this
     is I was viewing it as equivalent to kind of what they do in 1.174 where
     they have these regions and if you are above, say, on CDF, you are above
     10 to the minus 4, you kind of start worrying about it, or above 10 to
     the minus 5 you worry about it.  Once you approach 10 to the minus 4 you
     get more concerned.
         That is kind of what these two regions are.  You have
     defined two regions where you are beginning to worry about it, and what
     I was saying is they used an order of magnitude difference in 1.174 and
     you are using a little factor when you use the 95 percentile.
         DR. KHATIB-RAHBAR:  Yes.  It is probably a factor of 2 or 3.
         DR. KRESS:  Yes, and that bothers me.  I have trouble
     discriminating a two or three factor with PRAs and that's what is
     bothering me about it.
         DR. KHATIB-RAHBAR:  I understand what you're saying.  You
     are talking about uncertainty at least in order of magnitudes
     typically --
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  -- rather the factors of two or three.
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  But on the other hand, these are the
     type of things we compute -- I mean knowing the state of knowledge
     today, this is what we can come up with in terms of uncertainties.  I
     mean I agree with you.  These are driven by many other factors.
         Anyways, if you come essentially outside this 94th
     percentile, no action is necessary.  You are below that so it's not a
     problem.  However, if your CCDF at 95 percent level exceeds the
     criteria, then you see whether your mean exceeds this criteria or not. 
     If the mean does not exceed the criteria, then you will evaluate some
     procedural modifications -- you know, accident management actions, et
     cetera.
         DR. KRESS:  A little clarification on your 95th.  Is that
     the 95-95 or is that the 95th percentile on the mean?
         DR. KHATIB-RAHBAR:  95th percentile on the means.
         DR. KRESS:  It's not the 95-95, okay.
         DR. APOSTOLAKIS:  No, no, it's the curve -- you know, the
     log-normal curve you mentioned -- the uncertainty on the frequency of
     this or greater release.
         DR. KHATIB-RAHBAR:  On the greater releases, right.
         DR. APOSTOLAKIS:  Typically log-normal shaped.
         DR. WALLIS:  This doesn't help until you say how beneficial
     they have to be before they are worthwhile implementing.  You can't just
     say beneficial.  You have to say --
         DR. KHATIB-RAHBAR:  -- by how much.
         DR. WALLIS:  What they need to be before you do them because
     otherwise you just say it's infinitesimally beneficial, therefore I must
     do it.
         DR. KRESS:  Two thousand dollars, two thousand dollars per
     manrem.
         DR. KHATIB-RAHBAR:  This is just a qualitative picture.
         DR. WALLIS:  A 50.59 thing again.
         DR. KHATIB-RAHBAR:  This was just to think ahead if you
     wanted to implement such a system, what are the type of processes you
     have to go through, and the idea here was to distinguish amongst two
     things, one is things which are cost beneficial, what those cost
     beneficial factors are --
         DR. WALLIS:  Finding what's beneficial may turn out to be as
     much of a problem as deciding what is safe.
         DR. KHATIB-RAHBAR:  I agree.
         DR. APOSTOLAKIS:  Also, this is not PSA-based.  It is
     PSA-informed really, isn't it?
         DR. KHATIB-RAHBAR:  Define what is PSA-based.
         DR. APOSTOLAKIS:  Based means that -- what based means is
     that you will do all these things looking only at the risk numbers, and
     I bet you you are not going to do that.
         DR. KHATIB-RAHBAR:  No.
         DR. APOSTOLAKIS:  Okay, so it is informed.  You are going to
     look at what's driving them -- yes?
         DR. KHATIB-RAHBAR:  I don't know what is meant by risk --
         DR. APOSTOLAKIS:  Based means only the results of the PSA.
         DR. KHATIB-RAHBAR:  100 percent on the risk -- no, they're
     not.
         DR. APOSTOLAKIS:  And also I don't believe that the intent
     here is that, you know, you compare with the 95th -- yes/no?
         DR. KHATIB-RAHBAR:  No, it's not.
         DR. APOSTOLAKIS:  It's -- well, these are fuzzy rules
     really.
         DR. KHATIB-RAHBAR:  Precisely.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  This is, as I will again show through
     the example, the idea here is not to treat these lengths as absolute
     lines.
         DR. APOSTOLAKIS:  Sure.
         DR. KHATIB-RAHBAR:  This is just to provide some guidelines. 
     If a power plant exceeds these criteria --
         DR. SEALE:  -- leaky pipes, in other words.
         DR. KHATIB-RAHBAR:  Right -- to encourage the utilities to
     come up with ways to circumvent some of these issues rather than -- in
     fact, by and large a lot of the decisions -- the venting decision was
     made before even this thing was put into place.
         If you want to justify any cost beneficial basis, $1000 per
     manrem averted, a venting system probably would not be cost beneficial. 
     You know, these venting systems have cost millions of dollars to
     install, and the benefit you get from them for example depends on the
     plant.  For large dry containment it may not be very significant for
     some cases.
         But that is not the issue.  The idea is that some of the
     changes because societally they were considered to be things that people
     wanted to have done and there were some technical bases for them it
     would give an additional degree of comfort to the society if they were
     implemented regardless of cost.  That is an example.  Venting is one
     example.
         DR. FONTANA:  Dollars per manrem can be kind of misleading
     because it should be dollars per the what does it cost to recognize the
     surroundings -- it's a totally different number.
         DR. KHATIB-RAHBAR:  Even that.  Even that, for the venting
     even that will become very difficult.
         DR. KRESS:  There was some argument that that number
     included that at one time.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  Okay.
         DR. WALLIS:  I think you should have sort of a stock
     exchange where the price per manrem goes up and down depending on what
     we know and what people think.
         DR. KHATIB-RAHBAR:  Of course for future reactors it is to
     be at the 94 percent level, as indicated.
         DR. APOSTOLAKIS:  Let's go to your risk management exercise.
         DR. KHATIB-RAHBAR:  Let's look at some applications for some
     plants.
         Again, there are four plants, five units in Switzerland. 
     The two smaller, older units -- one GE boiling water reactor, one
     Westinghouse two-loop PWR, a BWR MARK-III, BWR-6 MARK-III and the German
     KWU plant.
         DR. APOSTOLAKIS:  And they all have detailed PRAs, I
     understand?
         DR. KHATIB-RAHBAR:  They all have done PRAs during shutdown,
     yes.
         DR. APOSTOLAKIS:  Including shutdown?
         DR. KHATIB-RAHBAR:  Shutdown, right.
         These are typical risk curves that one comes up with for
     these plants.  These are three of them only shown here.  We have the
     last one just completed but I don't have the curve here.
         Again, these are a combination of apples and oranges.  Let
     me explain a few things for you.
         The black line here represents the Westinghouse plant, the
     Betsnau plant.  This curve includes the risk associated with internal
     events and external events, everything, seismic, floods, fires,
     everything is included.
         For this particular plant, I believe this is Millerburg,
     which is a boiling water reactor.  This is a small reactor.  This
     excludes fires, but it does exclude -- include earthquakes, and aircraft
     crashes.
         DR. APOSTOLAKIS:  Why do they exclude fire?
         DR. KHATIB-RAHBAR:  It was not -- the utility study had the
     fires in there.  So this part, this is the evaluation that went by the
     Inspectorate.
         DR. APOSTOLAKIS:  Oh.
         DR. KHATIB-RAHBAR:  They found problems with the fire, so
     the fire was excluded from the analysis that they performed.  And the
     yellow line is the -- one, which is a boiling water similar to Grand
     Gulf, and this curve is internal events only.  Okay.  This does not have
     any external events at all.
         DR. WALLIS:  It seems incredible that with something as
     vague as PRA you get absolute coincidence of CDF with three curves.
         DR. KHATIB-RAHBAR:  It is interesting.  But, again, as I
     said, they are different size reactors.  This, for example, this reactor
     is about three times --
         DR. WALLIS:  But it looks so unusual as to wonder why it
     happened that way.
         DR. KHATIB-RAHBAR:  Yes.  Well, I guess it is not unusual,
     in this end of it, you just study the containments.
         DR. APOSTOLAKIS:  But why is it 10 to the minus 5 and not,
     you know, some spread?
         DR. KHATIB-RAHBAR:  Because all the reactors have been
     backfitted substantially, that basically the backfits that they have
     done, they have brought them to them level of current designs, which are
     typically 10 to the minus 5.
         DR. WALLIS:  So they aim for 10 to the minus 5.  It is not
     just fortuitous.
         DR. KHATIB-RAHBAR:  Because they have been backfitted by and
     large to satisfy those types of requirements.
         DR. POWERS:  They are cumulative distribution functions that
     get pinned, for that reason, on that side and get pinned by fission
     product release physics on the other point.  And, as we all know, all
     high entropy distributions are about the same.  And then we plot them on
     a log scale, so they really look coincident at this point.
         DR. KHATIB-RAHBAR:  No, no, no, that's not true, because,
     actually, we have done a comparison with about five or ten different
     plants, all over the place.
         DR. POWERS:  What's not true, that they are plotted on a log
     scale?
         DR. KHATIB-RAHBAR:  No, no, no.  They are all over the
     place.
         DR. POWERS:  They are plotted on a log scale.  There is no
     controversy that they are plotted on a log scale.  It is
     uncontrovertible that they are on a log scale.
         DR. KRESS:  I would definitely agree with that.
         DR. KHATIB-RAHBAR:  What is -- what is the problem?
         DR. APOSTOLAKIS:  Mohsen said that -- I mean if they did
     these PRAs, say, 20 years ago, then the flat parts most likely would not
     be around 10 to the minus 5.
         DR. KHATIB-RAHBAR:  Not at all.  Not at all.
         DR. APOSTOLAKIS:  What you said was that in the last several
     years the Swiss have passed requirements or regulations --
         DR. KHATIB-RAHBAR:  This plant, the core damage frequency,
     the first time it was done, it was up here.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  As I said, they have gone through about
     $200 million worth of backfits to the plants to bring them to the
     standards of modern PWRs or BWRs, and from modern BWRs and PWRs, that is
     where you are at.  Believe it or not, that is where you are at.
         DR. POWERS:  And that is what fits that side, the other side
     gets fixed by physics.
         DR. KHATIB-RAHBAR:  The other side is really fit by this
     line here, which, as I explained.
         DR. POWERS:  It is physical --
         DR. KHATIB-RAHBAR:  It is a consequence, absolutely.
         DR. KRESS:  And the black line then would have to have
     something done?
         DR. KHATIB-RAHBAR:  Yeah, we will talk about the black line
     in the next figures.
         DR. KRESS:  Okay.
         DR. KHATIB-RAHBAR:  Now, even these older plants are very
     similar to, in terms of risk, to some of the most modern plants.
         DR. APOSTOLAKIS:  So the earlier regulatory decisions were
     not based on these curves?
         DR. KHATIB-RAHBAR:  No, they were not.  They are not. 
     Absolutely.  But I mean just, it is coincidental after the make the
     modifications in areas.  For example, some of these plants, one of them
     did not have an ECCS system, an accumulator system.
         DR. APOSTOLAKIS:  But did they have the PRAs when they
     proposed these changes?
         DR. KHATIB-RAHBAR:  They actually did the modifications
     recently.
         DR. APOSTOLAKIS:  Okay.  So then the coincidence is --
         DR. KHATIB-RAHBAR:  But these are not coincidental from that
     point of view, yes.  Now, let's look at that black line, this is the
     Betsnau, the PWR one.  This is the 95th percentile curve.  This is the
     5th percentile and this is the mean.  I guess you had asked, George,
     that you were amazed that our uncertainties are relatively narrow
     compared to what you have seen before.
         DR. APOSTOLAKIS:  Yes.
         DR. KHATIB-RAHBAR:  I think if you look at these, comparing
     it probably to NUREG-1150, the uncertainties are wider because the 1150,
     my understanding was all the data was generic data, which they used in
     the 1150 PSAs.  These plants, by and large, are using plant-specific
     data, they have extensive databank.
         DR. APOSTOLAKIS:  And this is --
         DR. KHATIB-RAHBAR:  This is based on plant-specific data.
         DR. APOSTOLAKIS:  This is a complete PRA external events?
         DR. KHATIB-RAHBAR:  This is a complete PRA, including
     everything -- excluding for power, not other modes.
         DR. APOSTOLAKIS:  Not shutdown?
         DR. KHATIB-RAHBAR:  It does have a shutdown study, but these
     curves do not include the shutdown.  And the other factor is that there
     is some -- of course, in general, the seismicity for Swiss plants has
     been found to be problematic.  The uncertainties are not properly
     reflected in the seismicity curves for Switzerland, so that is being
     reassessed.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  With the exception of that, you do
     expect slightly narrower uncertainties here as compared to 1150, for the
     reasons I just stated.  But, by and large, if you put the 1150 numbers,
     they are not that far apart, they are probably about here.  They are not
     that much narrower than NUREG-1150.  In fact, there is a curve in the
     paper I sent you, it has a comparison.  For the boiling water reactor it
     will be even wider than 1150 in terms of uncertainties.
         Here you find that the 95th percentile level definitely
     exceeds the criteria, and the mean level down here, of course, does
     exceed it substantially.  There are a number of procedural
     modifications, not modifications -- accident management actions which
     have been developed for Westinghouse.
         DR. APOSTOLAKIS:  We are following in your previous block
     diagram.
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  In the region where the 95th is above, but
     the mean is almost okay.
         DR. KHATIB-RAHBAR:  Almost okay.  So now we are going to
     look at procedural modifications, which is really accident management
     actions.  These are like Westinghouse Owners Group procedures, if you
     look at it for a typical PWR, okay.  Additional water to degrade a core,
     manual depressurization of reactor cooling system, steam generator
     isolation, early isolation of steam generator during a tube rupture
     accident, additional fire water to a damaged part of steam generator,
     recovery of containment isolation early, cavity flooding, et cetera, et
     cetera, you go down the line.  These are some of them, if not all or
     most of the Westinghouse Owners Group, what they call SAMGs, severe
     accident management guidelines.
         DR. WALLIS:  So what you are telling us is that there is
     considerable flexibility.  If you change the regulations, we can change
     how we run the plant to meet them.
         DR. KHATIB-RAHBAR:  What?
         DR. WALLIS:  There is considerable flexibility, that is very
     nice to know.  It is not as if you are locked into having to have a high
     CDF, there are things you can do to bring it down.
         DR. KHATIB-RAHBAR:  You can do, exactly.  The idea here is
     to, again, to see how -- which part of the curve these will benefit the
     most and which one of them are going to be beneficial.  So if you took
     these actions and put -- use a risk model, the PRA model, and basically
     calculate the effect, this is what you find.
         DR. KRESS:  Cumulative.
         DR. KHATIB-RAHBAR:  This is the same thing, cumulative,
     right.  This is all of them, this is all of the actions considered
     together.  In fact, it turns out what brings this part of the curve down
     is the things which can eliminate or reduce the releases from bypass
     events, steam generator tube ruptures, additional water, for example. 
     This bought you essentially DF of a 100 right here.
         DR. WALLIS:  Is it true to say that all these procedures are
     thermal-hydraulic in nature?
         DR. MILLER:  I thought everything was.
         DR. KHATIB-RAHBAR:  All fission product, really, aerosol in
     nature.  This sends out really more aerosol physics than
     thermal-hydraulics, really.
         DR. POWERS:  Aerosol physics is thermal-hydraulics.
         DR. KHATIB-RAHBAR:  Not quite.
         DR. POWERS:  It is simply two-phase flow.  Yes, it is just
     two-phase flow.
         DR. KHATIB-RAHBAR:  Everything is thermo-dynamics from that
     standpoint.
         DR. APOSTOLAKIS:  Is the recovery of containment isolation a
     thermal-hydraulic issue?
         DR. KHATIB-RAHBAR:  No, that is basically a way for the
     operators to recover the isolation early, primarily --
         DR. APOSTOLAKIS:  So not everything is thermal-hydraulic.
         DR. WALLIS:  This is like losing a valve?
         DR. KHATIB-RAHBAR:  This is AC power recovery.
         DR. WALLIS:  Like losing a valve or something?
         DR. APOSTOLAKIS:  I don't know, that was his V-5.
         DR. KHATIB-RAHBAR:  Yeah, this is recovering AC power,
     essentially.
         DR. APOSTOLAKIS:  Let me change the question, give it a
     different flavor.  Does this mean, if all of them are thermal-hydraulic
     in nature, that this is the weak link in the plant, thermal-hydraulics?
         DR. WALLIS:  No, these are cures, George.  These are cures.
         DR. SEALE:  Removable heat capacity, George,
         DR. APOSTOLAKIS:  Let's not pursue this.
         DR. KHATIB-RAHBAR:  So all this shows is that even though,
     again, you are exceeding these lines --
         DR. WALLIS:  No, I think it is very interesting, the
     solution to a PRA problem is thermal-hydraulic.
         DR. KHATIB-RAHBAR:  If you look at the uncertainty --
         DR. APOSTOLAKIS:  You have to remove the heat, right?  There
     must be something there.
         DR. POWERS:  Well, I mean I think it is a truism, if we
     could do -- if we had good, keen, physical insight, we would not have to
     do PRA, except possibly for a certain amount of aleatory uncertainty,
     right, George?  If our epistemic uncertainties were zero, --
         DR. APOSTOLAKIS:  Yes.
         DR. POWERS:  -- then we would only be handling the aleatory
     part, right?
         DR. APOSTOLAKIS:  That's right.
         DR. POWERS:  If our insight was perfect, there would be no
     aleatory part.
         DR. APOSTOLAKIS:  There would be no aleatory part, that's
     correct.
         DR. KHATIB-RAHBAR:  I don't have the 5th percentile shown in
     figure because this was not very relevant.  But if I had looked at
     considering these, all the actions, there was an overall narrowing of
     the uncertainty band around the mean value, and this is really to
     demonstrate that there are procedural things we can do even you are not
     narrowing the uncertainty, but you are circumventing areas where you can
     get into high uncertainty.
         DR. APOSTOLAKIS:  You didn't really do much to the core
     damage frequency.
         DR. KHATIB-RAHBAR:  No, core damage frequency was not
     changed.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  The core damage frequency, because we
     did -- with the exception of recovering of AC power in a couple of
     cases, by and large, --
         DR. WALLIS:  You are just mitigating it once it has
     happened.
         DR. KHATIB-RAHBAR:  Yeah.  That is precisely it.  So this is
     how, again, a plant which falls outside the 95th percentile can fulfill
     this requirement.
         But to summarize, if there is such a summary, in my mind, I
     think FC curves can represent the common risk metric for all sources and
     all modes of operations.  It is just a question of semantics and also
     having to do that.  And the curves should not be treated as speed limits
     but as indicators for safety optimization more so than anything.
         Uncertainties cannot be expected to be fully eliminated, at
     least not in the foreseeable future.
         DR. WALLIS:  I am going to ask you how you optimize, unless
     you have some value function like dollars.
         DR. KHATIB-RAHBAR:  Yes, that will have to be done. 
     Absolutely.  And, as I said, an example I have at the very last figure
     which shows how I propose that would do such an optimization.
         DR. WALLIS:  It tells you where you need to do optimization.
         DR. KHATIB-RAHBAR:  Where you -- right.
         DR. WALLIS:  How do you do it?
         DR. KHATIB-RAHBAR:  What you define as your cost benefits
     criteria, et cetera, of course, are still issues to be addressed, no
     question about it.  You know, there are -- some people have drawn the
     parallel within this and the radiation protection requirements.  And I
     personally do not, but some have done so.  But you definitely will need
     to have a cost indicator before you can -- some cost standard before you
     can do this.
         But the idea here is that if the risk is way too high, it
     falls way outside of these curves, this is an area where they call --
     risk is not acceptable, you have to do something.
         DR. APOSTOLAKIS:  You are changing all the acceptable --
     acceptance criteria because you are going an order of magnitude higher.
         DR. KHATIB-RAHBAR:  This is just an example.
         DR. APOSTOLAKIS:  So this would be a different formulation
     of what you showed earlier.
         DR. KHATIB-RAHBAR:  Exactly.  Exactly.
         DR. APOSTOLAKIS:  Okay.
         DR. KHATIB-RAHBAR:  If you take that back to somewhere here.
         DR. APOSTOLAKIS:  Yeah.
         DR. KHATIB-RAHBAR:  But in my own mind, if I want to look at
     an optimization process where I am not going to treat this as a solid
     line, --
         DR. APOSTOLAKIS:  Right.
         DR. KHATIB-RAHBAR:  -- I would look at the region over which
     I am going to make some changes.
         DR. APOSTOLAKIS:  Is this part of the Swiss proposal?
         DR. KHATIB-RAHBAR:  No, it is not.
         DR. APOSTOLAKIS:  This is Mohsen's proposal?
         DR. KHATIB-RAHBAR:  This is my own, right.
         DR. POWERS:  If we think about how institutions behave, will
     they, in fact, attempt an optimization or will they, in fact, say I have
     this amount of money that I am willing to devote to this activity, how
     much can I do for that?  I mean I think that is how most agencies --
     most institutions behave, that they really don't do an optimization,
     just because it is so difficult to do.
         DR. KHATIB-RAHBAR:  Yeah, but the British, for example, have
     adopted such an approach.  If you look at the British way, the way they
     are approaching this whole --
         DR. APOSTOLAKIS:  NEI is proposing something similar here. 
     I guess the words are not the right ones, risk -- instead of
     optimization, maybe you can say increased regulatory concern.  So now
     they have a motivation for doing something.
         DR. KHATIB-RAHBAR:  But the process which is --
         DR. APOSTOLAKIS:  Yes.  I understand the spirit of it.
         DR. KHATIB-RAHBAR:  It is really the same thing.
         DR. APOSTOLAKIS:  Now why isn't this beautiful stuff that
     you proposed, you described here, accepted by the Swiss Nuclear
     Regulatory Authority?
         DR. KHATIB-RAHBAR:  That is a question you are going to have
     to ask --
         DR. APOSTOLAKIS:  Why are they still discussing it?  I mean
     are there any obstacles on the way?
         DR. KHATIB-RAHBAR:  No, I think there are a number of issues
     like the ones, I guess, Dr. Wallis raised.  They have to do with
     defining a lot of these other terminologies which are significant.  You
     know, how do you do cost benefit?  What are some of these other aspects
     you still have to define?
         DR. APOSTOLAKIS:  The practical implementation.
         DR. KHATIB-RAHBAR:  The practical implementation of it which
     will drive this thing.  And the other factor is that, by and large, even
     though this has naturally been applied on a plant-specific,
     plant-to-plant basis, they think they are fairly confident, by and
     large, they can satisfy these requirements.  As I have shown you, by and
     large, they do fall within the criteria on limits.
         So in areas where they have already exceeded those they have
     already made decisions a priori to modify the plants and make the
     changes necessary.  As I said, some of these plants had fairly high core
     damage frequencies before the modifications were instituted.  So --
         DR. APOSTOLAKIS:  Yes.  Actually one of your goals had to do
     with core damage frequency.  You say it had to be less than 10 to the
     minus 5 per year.
         DR. KHATIB-RAHBAR:  Right.
         DR. APOSTOLAKIS:  Now when you showed the curves, you
     focused on the consequences really, because even after the procedures,
     the 95th percentile was higher than 10 to the minus 5.  So they intend
     to look at that separately, or is that acceptable?
         DR. KHATIB-RAHBAR:  Yes, they're actually still -- they're
     looking at some of those core damage frequency figures.  As I said,
     there are a number of things which still have to be done to complete
     those studies.
         DR. APOSTOLAKIS:  Right.
         DR. KHATIB-RAHBAR:  And they will await -- at first we
     wanted to see what could be -- what benefits could be realized if we
     just looked at the consequence end of the picture to see whether indeed
     these numbers are totally out of whack and they're too conservative and
     no plant can ever meet them, or they are something that with reasonable
     actions and reasonable things you can bring these plants into a realm
     that you think they ought to be at.
         DR. KRESS:  Can I see your last example curve?  Now that
     you've already put it up.
         DR. KHATIB-RAHBAR:  This one here, the example?
         DR. KRESS:  Yes.
         DR. KHATIB-RAHBAR:  Yes.
         DR. KRESS:  There is some level down there where that line
     -- in a straight line across there all the way and still be
     risk-negligible, right?
         DR. KHATIB-RAHBAR:  Sure.
         DR. KRESS:  Yes.  That being the case, I envision a series
     of curves starting with a straight line, and as you go up --
         DR. KHATIB-RAHBAR:  You can --
         DR. KRESS:  It's getting steeper and steeper.
         DR. KHATIB-RAHBAR:  Yes.
         DR. KRESS:  And I was surprised as you got --
         DR. KHATIB-RAHBAR:  Yes, this --
         DR. KRESS:  It suggests --
         DR. KHATIB-RAHBAR:  This is just, yes --
         DR. KRESS:  Symbolic.
         DR. KHATIB-RAHBAR:  This is symbolic.  Not look at the
     numbers per se.  I'm just trying to show the process of what you should
     go through.  I should have actually not put any numbers on any of these. 
     It's just to show the process that one needs to go through and how these
     types of criteria should be treated more so than --
         DR. KRESS:  That's what I thought you meant.
         DR. BONACA:  There was another on the summary slide.
         DR. APOSTOLAKIS:  Mario, can you speak to --
         DR. BONACA:  Yes, the third bullet.  Could you -- no, no,
     no, the summary slide.
         DR. KHATIB-RAHBAR:  The summary slide.  I'm sorry.
         DR. BONACA:  Yes.
         DR. KHATIB-RAHBAR:  I'm going to have to find it.
         DR. BONACA:  Just expand a little bit on the third bullet,
     safety criteria.
         DR. KHATIB-RAHBAR:  On this one here?
         DR. BONACA:  Yes.
         DR. KHATIB-RAHBAR:  Uncertainties cannot be expected to be
     fully eliminated.
         DR. BONACA:  Yes.
         DR. KHATIB-RAHBAR:  I was talking about this tail of the
     distribution really, the tail of the exceedence frequency curve.  I'm
     saying in this region it's going to be very difficult that we can
     demonstrate with a high confidence deterministically that we can tell
     them with some confidence that release is 5.6 percent cesium, okay?  And
     eliminate all that certainty altogether.  All I'm saying is we have to
     divide it away by design or by procedures to overcome that issue. 
     That's the whole idea.
         DR. BONACA:  Okay.
         DR. APOSTOLAKIS:  So in other words instead of analyzing and
     talking about it forever --
         DR. KHATIB-RAHBAR:  Precisely.
         DR. BONACA:  Do something.  That's basically it, do
     something.  Right.
         DR. KHATIB-RAHBAR:  Yes.
         DR. APOSTOLAKIS:  Well, now the floor is open to general
     discussion.
         Joe, do you have anything to say?
         MR. MURPHY:  Always.
         I think there's one factor where I -- I think I agree with
     just about every technical fact that Mohsen said, but there's one area
     where I'm not sure I agree with him, and that has to do with the need to
     set a criterion for interdicted land.  The reason for this is that there
     is a very clear playoff between societal risk in terms of person-rem and
     interdicted land.  And what we really should do is try to optimize both. 
     And if I could use a blackboard or something --
         DR. KRESS:  There is a slate up there.
         MR. MURPHY:  I want you to know one of the highlights, just
     as an aside when I put this on, the pastor of my church left his mike on
     when he left the church and proceeded to comment on some other people
     while he was still wearing the mike, and it got broadcast all over.
         If I -- when I do a PRA for each accident sequence, I get --
     and all the way through level 3 -- I get a certain amount of
     contamination that is released.  It comes down, it is a certain amount
     of material is released at a place out on the ground, whatever.  From
     that point on the question is what dose do I get, how contaminated is
     the land, and what do I do with it.
         What it amounts to is that if I take the societal risk in
     person-rem, and the interdicted land area --
         DR. KRESS:  A person-rem meaning if there was somebody
     standing there?
         MR. MURPHY:  Actually, yes.  The actual dose that I would
     expect from the PRA making the various assumptions I have about
     evacuation --
         DR. KRESS:  Oh, okay.
         MR. MURPHY:  And nonevacuation.
         DR. KRESS:  You are putting in evacuation.
         MR. MURPHY:  Yes.  What it comes out as is that for a given
     amount I have a tradeoff.  What I'm saying is I have a certain amount of
     material on the ground.  I can either not clean it up and let people
     return to it, in which case I get very high person-rem, or I can clean
     it all up, and which I get no person-rem, but I get a tremendous amount
     of integrated area -- of interdicted area.  And the nature of this for
     the limit of stuff we did at the draft of NUREG-1150, we didn't do it
     for the final, and so the models are not quiet exact, but I think the
     general pattern is the same, is that you get a curve linearly that looks
     like that.  It's continuing dropping.
         And what we really want to do is find the null point, the
     minimum.  And then we can either set a criterion on societal person-rem
     or we can set a criterion on interdicted area, but we don't need both.
         DR. WALLIS:  You are assuming that all areas are equal? 
     There are some areas that people love more than others.
         MR. MURPHY:  In terms of my ability to do a calculation
     today, I'm assuming that, and I grant you that that's a bad assumption. 
     I have a way that we've accepted already of monetizing person-rem, and
     that's $2,000 person-rem discounted.  In terms of -- I can also convert
     the interdicted area to dollars, and in fact the codes that we used in
     NUREG-1150 supposedly can do that.  It looks at the land around the
     plant and in terms of what the products are, whether it's city or farm,
     if it's farm, whether it grows grain or wheat or peanuts or whatever. 
     Those are built into the code, and they try to monetize.  The problem is
     those codes, the economic model in those codes, has not been updated for
     I guess at least ten years.  Some of the ex-Brookhaveners in the room
     may remember when that was done.
         The models themselves as they are now have deficiencies, and
     the updating of that economic model will take a bit of work.  But, yes,
     you really ought to take a city is worth more than farmland, and you
     should take that into consideration.  But there's a basic randomness to
     this that Mohsen identified earlier.
         DR. WALLIS:  I think it's more than that, too.  Are you
     going -- the Swiss don't want Switzerland to become interdicted
     completely.  I mean, this is their whole nation.
         MR. MURPHY:  Well, I understand.
         DR. WALLIS:  And there are small States that might feel the
     same way.
         MR. MURPHY:  Well, clearly.  But you don't want everything
     interdicted, but by the same token, you don't want a high person-rem
     dose.  So it seems to me that because you have this general pattern,
     what you really want to do is find a limit, and then set a societal risk
     on one or the other.
         I am more confident in my ability to calculate person-rem
     than I am interdicted land area because of the fact that I have an
     outdated economic model.  And for that reason I would tend to pick the
     person-rem.  But you could do it either way.  And there's a balance,
     because certainly I don't calculate person-rem all that -- as well as I
     would like to yet either.  Person-rem calculations -- Jeff knows more
     about this than I do -- but my recollection is that about 50 percent of
     the person-rem dose in a PRA comes from distances greater than 50 miles. 
     Of that within 50 miles, better than half of it comes beyond 20 miles.
         Now there's a problem in that at least in terms of the max
     code that we use, in that it's a unidirectional model.  And if you look
     at statistical data on wind persistence in the United States, just
     looking at the whole map, you find that the median for wind blowing in
     one direction, looking at -- oh, I guess there must be 40 or 50 points
     on the map that's in the old Meteorology and Atomic Energy, is on the
     order of three to four hours.  And with a reasonable wind speed of five
     to six miles an hour, about average, taking a calculation out more than
     20, 25, 30 miles in one direction probably doesn't make sense.  But the
     model we have right now does.
         Now why does it?  Well, there have been attempts to build
     other models along that line, but the problem is the lack of downfield
     data in the wind field.  Even if you have good data at the site and you
     know that the wind changes direction at the site, and there are some
     models that then take the plume further on down and bend it the same
     way.
         DR. KRESS:  Joe?
         MR. MURPHY:  That data doesn't necessarily agree.
         DR. KRESS:  Being able to plot one parameter versus another
     like this implies to me that there is a third parameter that they're
     both correlated by that is varying along this curve.  Is it what,
     fission product release?  If I go from the left -- if I follow that
     curve from the left side to the right, what is varying?
         MR. MURPHY:  This is essentially for the same beast.  I take
     one accident and I release -- I let nature do its thing, if you will.  I
     have a release at 10 to the 8 curies of cesium, and it goes out and it
     gets dispersed by the wind, and some of it comes down and some of it
     doesn't.
         DR. KRESS:  But you've got one -- one point --
         MR. MURPHY:  Now once it is down --
         DR. KRESS:  That gives you one point, though.
         DR. SEALE:  No.
         DR. FONTANA:  No, he's saying is that he can -- you can
     interdict more area --
         MR. MURPHY:  Once it is down, I have one point now, if you
     will, one physical specimen.  Now I say what is the dose to the public? 
     Well, the dose to the public depends on whether there's any public there
     to get a dose.  If I interdict all the land, there's no person-rem.  The
     people are gone, because the dose to the public in terms of the
     long-term effects --
         DR. KRESS:  Okay.  Now --
         MR. MURPHY:  Comes from ground shine.  Hardly any of it
     comes from the cloud.  And so if I get them out before they have down
     shine and then don't let them back, which is what interdicted means,
     I'll go all the way down to zero on person-rem.  By the same token, if I
     say I don't want to interdict an acre, I may have some very high
     person-rem.
         DR. KRESS:  So this basically -- yes, I understand.
         MR. MURPHY:  And what I'm saying is what we really should do
     is -- and the curve does look something like this.  It's an L-shaped
     curve.
         DR. WALLIS:  But you're talking as if you can decide.  I
     think of so many historical situations where groups of people have been
     willing to fight to the death, all of them, in order not to be
     interdicted from their area.
         MR. MURPHY:  That's true.
         DR. WALLIS:  So there's some human values here which you
     can't just --
         MR. MURPHY:  From a technical standpoint it makes sense to
     do this, but then there's the political standpoint, and this is a policy
     decision.  And it should have broad stakeholder input.
         DR. KRESS:  You see this though to me says given I have an
     accident, this is a decision on what do I do.  It's not an acceptance
     criteria related to these F-C curves.  It has very little to do with it,
     it seems to me.
         MR. MURPHY:  It's both.
         DR. KRESS:  It is?
         MR. MURPHY:  For a given accident, as it's in progress,
     somebody's going to have to decide when to interdict.  We have
     guidelines that are out there now.  EPA has guidelines on this.  Those
     guidelines were set by doing this kind of a calculation using the
     WASH-1400.
         DR. KRESS:  Restating my question, it has nothing to do with
     the design of the reactor.
         MR. MURPHY:  It doesn't until I get to setting a criterion
     here that's not in terms of person-rem or person-rem per year, and then
     the basic core damage frequency plays in.
         DR. SHACK:  Well, you get families of curves depending on --
     so that there is a curve for a reactor.  And it would be affected by the
     design in a sense.
         MR. MURPHY:  Yes.
         MR. HOLAHAN:  Dr. Kress, I think what you would like to do
     is have some socially acceptable point, and then find out what reactor
     design and emergency plan and all that stuff would draw a line through
     that point.
         DR. APOSTOLAKIS:  Before we go on, I have to leave at this
     point, but the discussion will go on.  Dr. Kress will chair the rest of
     the meeting.
         DR. KRESS:  Before you leave, George, what would you like to
     accomplish today, or did you want to put it off till tomorrow in terms
     of what to carry over to the main meeting or --
         DR. APOSTOLAKIS:  Well, one thing is what Dana raised there,
     we want to brief the Committee --
         DR. KRESS:  Are we deciding that today or tomorrow?
         DR. APOSTOLAKIS:  Tomorrow is a different meeting, actually. 
     It's on a different subject.  So today.
         DR. KRESS:  We need to decide --
         DR. APOSTOLAKIS:  Today.  Yes.
         DR. KRESS:  That's what I need to do before we close.
         DR. APOSTOLAKIS:  Yes, and maybe get comments from around
     the table.  Actually I had a question myself, but there is no time.  I
     mean, developing subsidiary goals from the curves.  Because if you want
     to use these things in 50.59, you definitely have to go below this.  But
     that's for something else.
         So thank you very much, Mohsen.  This was an excellent
     presentation.  I appreciate your coming.  In fact, thanks both to Dr.
     Kaiser and Dr. Khatib-Rahbar for agreeing to come here as invited
     experts, which means they were not paid.  So we do appreciate this.
         And now you can go on with the discussion.
         DR. MILLER:  And they're not off the hook yet.  We will ask
     them some more.
         DR. APOSTOLAKIS:  And we'll ask them some more.  Yes.  Thank
     you.
         MR. MARKLEY:  I'd just like to make one point.  There is no
     briefing session to the full Committee.  Since this is an ACRS
     initiative, it's really sorting between what you want to do, so staff's
     not bringing something forward to talk about, nor are invited experts
     bringing something for you to consider.  This has been offered for your
     information.
         MR. HOLAHAN:  My suggestion would be that this and many
     other papers and meetings would be background for sometime later in the
     spring, whether it's, you know, somewhere between probably March and
     June, when Mr. Murphy brings the where do we want to go with the safety
     goal paper around.  I'm sure --
         DR. KRESS:  This does sound like a safety-goal issue.
         MR. HOLAHAN:  Right.  Yes, I think it is.  It's what should
     the safety goal look like and how should you use it, and I think -- in
     my mind, there's no -- I don't see any direction action from this
     absent, you know, some thoughts on where we ought to go with safety
     goal.
         DR. KRESS:  Well, I was in my mind relating this somewhat to
     the considerations one ought to have in mind when one risk-informs Part
     50 --
         MR. HOLAHAN:  Um-hum.
         DR. KRESS:  Because we're talking about how to risk-inform
     regulation.  So it may have some relevance to that also.  I don't know
     how to work it into that.
         MR. HOLAHAN:  It could, although I have to confess that I've
     been thinking of risk-informing Part 50 like 1.174 with, you know, LERF
     and CDF.
         DR. KRESS:  See, the problem I have with that is 1.174 has
     things in it like you will meet the strictures of the current
     regulation, which are mostly dose strictures --
         MR. HOLAHAN:  Yes.
         DR. KRESS:  And then -- but you're going to change the
     regulation.
         MR. HOLAHAN:  Yes.
         DR. KRESS:  And it's sort of a circular argument to say meet
     the regulations when we're going to change the regulations, and it's
     these dose criteria that I worry about.  And this was my attempt to
     attack that part of the problem, although I like the idea of the
     thinking of 1.174.  But that was the problem I had with it.  That's why
     I thought this might be relevant in that part of 1.174.
         MR. HOLAHAN:  i think it can be relevant in that context.
         Since I happen to be talking, why don't I -- my observations
     from a meeting like this are not that I'm inclined to take the
     regulatory process to, you know, frequency consequence curves, because I
     think there are a lot of, you know, practicality and stability and
     understandability issues, that we're already changing things maybe, you
     know, about as fast as they can be changed without being sort of
     chaotic.  But in my mind these sort of studies and this sort of
     discussion helps identify areas in which the requirements we have and
     the criteria and approaches that we're thinking of may be either well
     calibrated or poorly calibrated with respect to, you know,
     overregulation or underregulation.
         So at the moment I don't -- I'm not enthusiastic about
     establishing a whole new regulatory paradigm, but it's a good testing
     mechanism to see where the things that we're currently doing are in some
     way, you know, out of calibration.  And I think -- so the insights from
     some generic studies like this in my mind are more valuable than an idea
     that everybody ought to go off, do such analyses, and make regulatory
     decisions based on them.
         DR. WALLIS:  How about the long-term view, Gary, though?  I
     mean, you're saying that today you have to do what's practical.  That's
     true, but I think you also ought to have some idea of where the Agency
     might want to be 10, 20 years from now.
         MR. HOLAHAN:  I think I'd like to make that decision after I
     learned how good or how poorly, you know, the deterministic set of
     regulations and something as simple as core damage frequency and large
     early release, you know, how good a job they do.
         And then whatever residual dissatisfaction you had would be
     the driving force for changing the next 10 or 20 years.  But I wouldn't
     want to change it just because it's intellectually, you know, more --
         DR. BONACA:  Yes, I have a question in fact regarding --
     probably on the same issue, Mohsen, which has to do with what is a
     contest of this application in Switzerland.  What I mean is that if I
     look at 50.59, it doesn't accomplish a risk evaluation in itself. 
     However, it accomplishes other things that are important too, which is
     closure and commitments and things of that kind.
         And the question I'm having is I don't understand to what
     this process is applied in Switzerland.  Is it applied to any change
     that is made, and in that case, and if we transfer to a risk-based
     decision, how do they accomplish other issues such as closure and
     commitments and maintaining documentation and all of that kind of stuff?
         DR. KHATIB-RAHBAR:  Yes.  In terms of specifics, I don't
     believe that they have really gotten that far in developing how far they
     are going to apply such a thing.  But one of the requirements that they
     have by law is that their powerplants have to stay current with the
     currency of the art and technology.  That means that the -- this is a
     moving target in a sense, that as you improve the technology, you have
     to improve the plants.
         So the plants are -- they go through a periodic safety
     review there every ten years, every plant is reviewed from bottom up,
     completely.  And there is no such a thing as a 30-year or a 40-year
     license life.  So the license life could be indefinite theoretically. 
     So such a process is to be applied to see how you can -- and essentially
     how good is safe enough and how safe is safe enough, and then as this
     technology evolves, how you improve your plants over the ten-year
     period, how do you measure relatively speaking.  So insofar as specific
     changes, I don't believe they have gotten that far to address that
     issue.
         DR. SEALE:  But presumably the assumption is that if they
     maintain consistency with the state of the art and the technology, that
     plant is a reasonable candidate for an extension of its lifetime
     regardless of how old the concrete happens to be.
         DR. KHATIB-RAHBAR:  That's correct.  That's an implicit
     assumption.
         DR. MILLER:  So in a way they're measuring cumulative
     changes over that ten-year period and seeing if that's moved them toward
     a safer reactor, so to speak.
         DR. SEALE:  Well, and they're also adjusting the
     configuration of the plant.
         DR. KHATIB-RAHBAR:  It's mostly, yes, the plant
     configuration which is changing as a function of time.  For example,
     after Three Mile Island there was a number of major changes made to the
     plants.  After Chernobyl, there were additional changes made.  And now,
     as these older plants are coming for their evaluation, ten-year-period
     evaluation, they're making additional changes.  Some of them, in fact,
     if you look at them, they probably are more suited than even some more
     modern plants which are currently under construction in some of the
     other countries, in the neighboring countries.
         But the idea is that you have to be able to change your
     level of safety to improve your safety with time, because technology
     allows you to do that.  And that is the basic concept behind this
     ten-year evolutionary process.
         DR. MILLER:  Dr. Seale --
         DR. KHATIB-RAHBAR:  I'm sorry?
         DR. MILLER:  Dr. Seale said that was configuration.  I
     assume that was a combination of configuration and equipment.
         DR. SEALE:  Both.
         DR. MILLER:  So I'm in the area of INC -- you mean,
     configuration, you mean both.
         DR. KHATIB-RAHBAR:  Yes, both.  Yes, both, both.  For
     example, this is one area they're going to for example to some digital
     INC, and they're looking at that as one way where technology has changed
     and made things perhaps better, perhaps worse.  And you have to look at
     it.
         DR. MILLER:  The challenge is to prove it's better.
         DR. KHATIB-RAHBAR:  Yes.
         DR. MILLER:  I'm sorry, Mario?
         DR. FONTANA:  My thought was at some point the capability of
     the plant to meet current technology would not be possible.
         DR. KHATIB-RAHBAR:  As it's currently configured and
     designed.  But they are changing them.
         DR. FONTANA:  Look into the future --
         DR. KHATIB-RAHBAR:  It's possible.  It's possible.  I don't
     know.
         DR. FONTANA:  Haven't hit it yet.
         DR. KHATIB-RAHBAR:  But in general, I mean, if you look at
     some of these older plants, they have made major changes.  For example,
     all the plants are required to have a totally independent redundant
     shutdown heat-removal system.
         DR. MILLER:  Filter vents.
         DR. KHATIB-RAHBAR:  Filter vents.  In other words, these
     plants are de facto designed against, you know, other factors, external
     factors, other than just accidents.
         DR. FONTANA:  Well, I was thinking, it was kind of an absurd
     case, like a whole new material for pressure vessels that makes all
     these problems go away.  Well, you can't --
         DR. KHATIB-RAHBAR:  That's true.  There are certain boundary
     conditions you cannot change.  Absolutely.
         DR. SEALE:  Gary, I know you had your hand up.
         MR. HOLAHAN:  Going back to Dr. Kress' question, about if
     the only thing you really believe is PRA, how can you have -- in effect,
     how can you have a risk-informed process, because everything that the
     PRA tells you isn't important, some of those you're going to keep.  And
     how are you going to decide which ones to keep.
         I think there is some information you can draw from these
     sort of studies to help you with that.  The first thing you can do is
     you can certainly identify existing requirements or systems or whatever
     that are irrelevant to risk.  And, you know, to land contamination or
     whatever you decide are the metrics that you're interested in.
         And I think even though you're going to keep, you know, some
     defenses, almost by definition contrary to what the PRA would tell you
     to do, you can identify where those additional margins are much greater
     than the uncertainties involved.  And I don't mean -- I mean the real,
     total uncertainties, where you have to really think about what you're
     modeling and, you know, as well as the parameters.
         And I think that if the purpose of defense-in-depth and
     safety margins is to cover for what you really don't know and what is
     not modeled in the PRAs.  Then I think it is legitimate to go back and
     test your defense-in-depth engineering margins against what the PRA
     tells you you know really well on what you don't know so well.  And you
     keep those things that you are not so comfortable with, and you let go
     those things that seem to be irrelevant or have, you know, orders of
     magnitude margin more than is necessary.
         DR. KRESS:  Good thought.  That complies with my thinking
     about defense-in-depth ought to be utilized in a manner that tends to
     minimize your uncertainty.  And what you are saying is you look at
     places where uncertainty is very hard, and there you keep
     defense-in-depth, and I think that is a good concept.
         I like the proposal that Gary made that -- we have got most
     of the committee here now.  It just seems like we don't really gain much
     by having a presentation or more discussion at the next meeting at all.
         DR. SEALE:  Yeah.
         DR. KRESS:  And that we can utilize this as background
     material for thinking about -- I think we ought to use in thinking about
     risk-informing Part 50, too, but mainly for changing the safety policy
     statement.  And so I can't really, at this time, see the need for, say,
     asking Mohsen to come back to the meeting and make a presentation.
         MR. MARKLEY:  The other benefit you have is that you are
     going to be meeting with the Commission and they may ask what your
     current thinking is since you put this in a letter on 50.59.
         DR. KRESS:  Well, the problem with it is most of this is not
     really relevant, in my mind, to 50.59.  But you are right, that probably
     will get asked.
         DR. MILLER:  Are we going to tell -- that should be our
     response to the Commission, we have decided it is not relevant?
         DR. KRESS:  Well, I don't want to speak for the Committee
     and George is making that particular part of the talk.  But I would
     like, at this point, to ask if there are other members that want to make
     some comments or ask question.
         DR. SEALE:  I have a question.  We haven't heard very much
     from Jeff, and I would like to ask him one in particular.
         DR. KRESS:  Oh, wonderful.  Wonderful.  We have heard quite
     a bit from him, though.
         DR. SEALE:  Well, but not nearly as much as our more vocal
     members and Mohsen.  When George introduced you, he made the comment,
     Jeff, that you were in charge of PRA application for process industries,
     is that -- did I hear that correctly?
         MR. KAISER:  Yes.
         DR. SEALE:  One of the difficulties we have occasionally
     when we talk about the flexibility of PRA or the application of PRA to
     other topics, or to things that are as yet unsatisfactorily covered in
     the existing PRAs, is the idea that somehow we have a limitation because
     our -- the PRA methodology that we generally find used in the nuclear
     industry is, to be redundant, the nuclear industry's version of the PRA
     methodology.  And, undoubtedly, in the other industries that have
     applied PRA kinds of -- or PSA kinds of methodology to analysis of their
     problems, they have had to have invented at least one or two things that
     aren't on the radar screen of the nuclear industry.
         DR. MILLER:  And probably eliminated a few that are.
         DR. SEALE:  Yes.  And I realize, and you are a better judge
     of this than I am, that is probably a topic that could last at least as
     long as we have been here today.  But recognizing that, could you
     perhaps, and you don't have to mention it right now, but if you could
     tell Mike what some references or places we might go to get ourselves
     more up to speed on what some of the other applications of PRA have
     been.  I think you get the idea of what I am driving at.
         MR. KAISER:  Yes.  I think I would say I don't think you
     will find, at least in the chemical and petrochemical and petroleum
     industries, that they have done sophisticated things that the nuclear
     industry doesn't do.  I think it is more a case of them doing much more
     quick and dirty PRAs or QRAs, as they call them, quantitative risk
     assessments.  For example, very often they will simply do point estimate
     risk assessments.  Spend some time thinking about the values of the
     individual parameters that go into it, but, in the end, they will just
     select one of those parameters and then generate a single CCDF or a
     societal risk or something like that, and then use that to make
     decisions about where they are going to put their dollars to reduce
     risk.
         DR. SEALE:  Okay.  So this is an input to a cost benefit
     assessment then that they might do?
         MR. KAISER:  Yes.
         DR. SEALE:  I guess some thoughts on how they discipline
     that cost benefit analysis might also be interesting, because, you know,
     we talk about $2,000 a person rem, but that is -- well, undoubtedly,
     there are other considerations.
         DR. POWERS:  There is another point that you can't let slip
     by, and that is the frequency with which societal, as opposed to
     individual risk is used in the process industry as a criterion.
         DR. SEALE:  Yes.
         MR. KAISER:  Yes.  Well, they do use both, there's no doubt
     about that.  I can dig up some applications.
         DR. SEALE:  We would certainly appreciate that.
         DR. POWERS:  Where they have, in my experience, you find
     that societal risk frequently gets used when you are looking at
     protecting workers in a facility.  And when you think about societal
     risk in that concept, you find that it has a complementary nature to the
     individual risk in exploring the space of safety.  I haven't thought
     about it in the grander public arena, but, certainly, among workers
     within a facility, you create two dimensions in looking at safety when
     you look at both societal and individual risk.
         MR. KAISER:  However, I would say most industries that I
     know would not use quantitative risk assessment for the workers.  They
     use process safety management to ensure that their workers are safe, and
     that is entirely a qualitative endeavor.  It's very rare in my
     experience for a Mobil or an Exxon to use a --
         DR. POWERS:  Go to Dupont.  You get to use it in spades.
         MR. KAISER:  Yeah.  At a Dupont commercial site or a Dupont
     Savannah River place.
         DR. POWERS:  My only familiarity with it is -- and
     Wilmington.
         MR. KAISER:  Wilmington, Delaware.
         DR. POWERS:  Yes.
         MR. KAISER:  Okay.
         DR. MILLER:  Well, they use it down at Savannah River,
     though.
         DR. POWERS:  Well, that is a relatively modern, I mean
     relatively new innovation at Savannah River.
         DR. MILLER:  Dupont has been out of Savannah River for quite
     some time..
         DR. POWERS:  Yeah.
         DR. MILLER:  But they were using it before they left.
         MR. KAISER:  Right.
         DR. KRESS:  Are there other comments from the members, that
     they would like to make before we -- I guess you have all commented out.
         DR. MILLER:  Well, I think the only comment I would make is
     I agree with apparently what you said early on, and others, I don't see
     a relation of this with -- applied to 50.59, and it was very valuable
     for me to better understand what is going on.  I agree with Gary Holahan
     that it gives you a good context to look at the LERF and the core damage
     frequency, but I don't -- I think it is a way before it is a practical
     situation to even -- when you look at risk-informed Part 50.  But I
     don't see how it is sensitive enough for a 50.59 application.
         DR. KRESS:  That's my problem.
         DR. MILLER:  Of course, I had the same opinion of core
     damage frequency, too.  The two are related, obviously.
         DR. FONTANA:  If you look at the FC curve, and you look at
     the lefthand side of it, and you start thinking of 50.59, it doesn't
     help you much because the allowed probability doesn't go up.  If you use
     50.59, I don't think you have -- the way lefthand side of the curve is
     just.  I think, Tom, you didn't discuss it much, but your relationship
     on relating the FC curves to steam generators, I thought that was a very
     good job.  I think it helps to correlate these and rationalize them.
         I also feel that land interdiction really ought to be
     calculated independently and added up.  I think it is potentially so
     important, and I know there is a lot of emotion involved there.
         DR. MILLER:  So you wouldn't agree with Joe Murphy said
     then?
         DR. KRESS:  And what I said, yeah, you calculate it
     independently and add up the costs along with the other costs.  You
     would put it on a common metric.
         DR. FONTANA:  But I think, basically, that the FC curves are
     a very good concept There is a tendency to get balled up with
     uncertainties, but I think the way you are proposing to try to get
     around that looked pretty good.
         DR. MILLER:  It does handle -- it does give uncertainties a
     better context than just the core damage frequency.
         DR. KRESS:  Any other comments?
         DR. WALLIS:  Well, I have a comment.  When ways of doing
     engineering become mature, they get taught.  Now, what you presented
     here is sort of the level of research and development and advising
     governments and that sort of thing, but, eventually, this gets taught to
     people as this is the way it is done, or this is the way that you as
     professionals will be doing things in the future.  And I wonder how far
     along this is evolving to the kind of thing that appears in classrooms
     and students learn to do it this way, and, eventually, they will be
     hired by people like the NRc.
         DR. KRESS:  I didn't know there were any more students.
         DR. WALLIS:  Because that eventually has happened.  If
     something becomes a worthwhile way of doing things in technology, it
     becomes taught and becomes practice.  Has this evolved to that point?
         DR. MILLER:  That sounds like a challenge for engineers.
         DR. WALLIS:  Are universities teaching this stuff as being
     the way to do things?
         DR. MILLER:  No, the challenge would be to --
         DR. WALLIS:  Does George teach it as the way to do things at
     MIT?
         DR. MILLER:  George, maybe.
         DR. KRESS:  Only national laboratories can do that.
         DR. WALLIS:  So, as far as you know, it is still at the sort
     of research level?
         DR. KHATIB-RAHBAR:  I have no idea what they are teaching.
         DR. SEALE:  I have another concern about all of this.  In
     the present environment, if a plant gets shut down for six months by the
     NRC, it is dead, almost certainly.  Now, maybe if it is a station that
     has two or three other plants with it, and they can keep a cash flow up
     that will help take care of people to do the things that are necessary
     and so on, maybe it isn't quite that bad, but it is close to that.  And
     those kinds of events are over here on that part of the curve that we
     haven't talk about at all today, that flat part there, that mesa, if you
     will, that vast wasteland, that whatever you want to characterize it as. 
     We haven't talked about that at all.
         And if I look at the methods we are using for PRA, they
     don't just tell us a hell of a lot about what is happening there either. 
     And I am just concerned that as we talk about changing regulations and
     going to risk-informed methodologies and so forth, and we don't have the
     methodology to evaluate risk where it is most likely to have severe
     impact, are we really serving the industry well, I mean in the sense of
     -- and the public?
         DR. KRESS:  I think that is a real good point, Bob.  I don't
     agree with you that it does not incorporate it in the curve.  I think
     the PRAs can address those things.  My feeling is, you know, I had these
     other considerations on one of my lists, and one of them is time risk. 
     I didn't lift it, I had time left on license, but actually I mean time
     at risk.  And I think that can be used as one of the gauges.
         DR. SEALE:  Well, I will share --
         DR. POWERS:  I don't think that the comment is accurate,
     though.  That we have been looking today exclusively at the integral
     measures that come from a risk assessment and those attract our
     attention.  But the truth is that those regulations that we have been
     risk-informing call upon looking at the differentials, that is, looking
     at things like the risk achievement worth and the risk reduction worth. 
     We have not spoken of those using this FC curve as the object function
     where you are getting the differentials.  And it is in those regions
     that are used in the regulations.
         For instance, you can use it in the maintenance rule to
     decide what are the important systems components and structures.  There
     are a variety of applications for those used, and it is not in a
     wasteland, it is on the differential of those curves where these things
     get applied, and it is built into our regulatory philosophy, that when
     we degrade elements of the defense-in-depth, that is significant to us
     and the quantification of it comes in the differential of these curves. 
     We just haven't talked about it.
         DR. KRESS:  I think you are basically right and that is one
     reason I like to retain the concepts of things like LERF and
     complementary -- or conditional containment failure problem and CDF,
     because those things tend to be amenable to a differential evaluation,
     whereas, these FC curves, you are going to have a little problem when
     you talk about --
         DR. SEALE:  I think that is why Gary is not all that
     enthusiastic about FC curves, or at least one reason.
         DR. KRESS:  Yes.  That is why I like to retain these things,
     too, as part of our system.
         DR. BONACA:  Yes.  Just one observation.  It seems to me
     that this approach that you presented seems to me appropriate for macro
     decisions.  By macro, I mean where you are talking about measure changes
     that can be really looked at favorably or unfavorably at the high level,
     which means, you know, talking about putting in a filter vented
     containment, that definitely would be a decision, even smaller changes
     probably.  But to me, that doesn't preclude, however, the value of PRA
     in the just day-to-day engineering used, you know, in a much more
     limited framework of the decision you make as part of an engineering
     analysis.
         Another thing that, you know, supporting this approach here,
     should exclude our encouragement for the other application, which,
     really, it is almost complementary to --
         DR. KRESS:  Yes, this is almost an additional complementary.
         DR. BONACA:  That's right.  This is really --
         DR. KRESS:  It doesn't necessarily -- it may bring some
     consistency to these other things if you do it properly.
         DR. BONACA:  That's right.  So, I mean, you know, I don't
     think that -- if we say that this is not applicable today to 50.59, we
     don't want to use it, in that sense, it doesn't discount, in my mind,
     still, the value of encouraging use of PRA in more engineering type of
     applications, as we have seen, extensively, and they are being done
     today.  So, anyway, that is my perspective there.
         DR. KRESS:  With that, what I think I will do is add my
     thanks to the valued experts who did a really good discussion and we are
     really pleased that you were able to come, plus the staff.  I was very
     glad you could come and valuable comments and valuable discussions, and
     we really appreciate it.  So, with that, I think I will adjourn this
     particular meeting.
         Tomorrow is a different meeting, right, so I can adjourn.
         DR. SEALE:  Start at 8:30?
         DR. KRESS:  8:30, yes.  So this meeting is now adjourned.
         [Whereupon, at 5:07 p.m., the meeting was concluded.]
 

Page Last Reviewed/Updated Tuesday, July 12, 2016