Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: A few examples



Dan Zuras wrote:
But ignoring that, if I throw in a few examples of my own
together with your case analysis with xx = {[0,4],D3} we
get (please check my work):

D4 ==> empty & result = empty:
sqrt(xx - 5) + sqrt(xx - 5) = {empty,D4}
{empty,D0}  +  {empty,D0}  = {empty,D4}

By Definition 3, the tracking decoration of the addition should be:
   (Empty,D0) + (Empty,D0)
       = (Empty+Empty,inf(S(+,Empty,Empty),D0,D0))
       = (Empty,inf(D4,D0,D0))
       = (Empty,D0) // this does not agree with your answer


D4 ==> empty & result = nonempty:
f(xx) == xx \union {[5,6],doesItMatter?}
==> f(xx) = {[0,6],D4}
By Def. 3, the tracking decoration should be
   ([0,4],D3) \union ([5,6],??)
       = ([0,4] \union [5,6],inf(S(\union,[0,4],[5,6]),D3,??)
       = ([0,6],inf(D3,D3,??))  // the answer depends on ??


D3 ==> inDomain&Continuous & result = empty:
xx \intersect (xx + 10) = {empty,D3}
By Def. 3:
   ([0,4],D3) \intersect ([10,14],D3)
       = ([0,4] \intersect [10,14], inf(S(\intersect,[0,4],[10,14]),D3,D3)
       = (Empty,inf(D3,D3,D3))
       = (Empty,D3)  // agrees with your answer

D3 ==> inDomain&Continuous & result = nonempty:
xx \intersect (xx + 1) = {[1,4],D3}
   yes

xx + xx = {[0,8],D3}
   yes

sqrt(xx - 3) +   (xx + 10)  = {[10,15],D3}
{[0,1],D1}   + {[10,14],D3} = {[10,15],D3}
   sqrt(([-3,1],D3)) + ([10,14],D3)
       = ([0,1],D1)+([10,14],D3)
       = ([0,1]+[10,14],inf(S(+,[0,1],[10,14]),D1,D3))
       = ([10,15],inf(D3,D1,D3))
       = ([10,15],D1) // does not agree with your result

floor(xx) +  (xx + 10)   = {[10,18],D3}
{[0,4],D2} + {[10,14],D3} = {[10,18],D3}
   floor(xx) + (xx+10)
       = ([0,4],D2)+([10,14],D3)
       = ([0,4]+[10,14],inf(S(+,[0,4],[10,14]),D2,D3))
       = ([10,18],inf(D3,D2,D3))
       = ([10,18],D2) // does not agree with your result

...snip...


Do I have all these correct?

No.


. . .

> Well, as I recall only a partial ordering was necessary to
> prove FTDIA.  It was:
>
> empty < inDomain&Continuous < inDomain < in&Out
> empty <       outDomain         < in&Out
>
> Where, as you pointed out, 'ill' need only be placed at
> some level below level 1.
>
> So anything that fits this description would do the job.

Agreed.

IMHO, Dan, it seems you understand pretty well.

Nate

Well, it *appeared* you were understanding, but now I see I was wrong, and that you've completely missed the point... :-(



Then, in keeping with your assertion that this is a
level 1 only motion, let me suggest an amendment.

...

So the amendment is: Eliminate discussion of implementation
details such as bits & numbering of decorations; rename the
decorations as John once did; & make an explicit statement
about the partial ordering required.

That way this motion gives us everything we need for an
FTDIA at level 1 with as little controversy as possible.

Will you accept this amendment as friendly or should I ask
for a second?

Dan, what you are proposing is to eliminate the whole purpose of the motion: property tracking. Clearly when you don't follow the Definition 3, you get lots of wrong answers that will quickly cause most interval algorithms to fail.

So I will certainly not accept this as a friendly amendment!!

Nate