MathGroup Archive 2008

[Date Index] [Thread Index] [Author Index]

Search the Archive

RE: Re: Workaround for an unexpected behavior of Sum


Well, it is so great to have these responses to my post. Thank you too,
Albert

Another issue, somehow related with your answer, Albert, is that Mathematica
has two different ways to specify double sums, the two of them format the
same in standard form (with the sigma notation), but they are not consider
to be equal by Mathematica (I already posted this in another thread of this
forum, however I considered relevant to mention it again):

Evaluate in a Mathematica notebook this code:

Sum[Sum[f[j], {j, 1, k}], {k, 1, n}]==Sum[f[j], {k, 1, n}, {j, 1, k}]

As you can see, both sides of the equation display exactly equal in a
notebook, using two sigmas, etc. However Mathematica does not simplify this
to True, not even using Simplify or Reduce:

Simplify[
 Sum[Sum[f[j], {j, 1, k}], {k, 1, n}]==Sum[f[j], {k, 1, n}, {j, 1, k}]
]

Mathematica is a great software, I feel happy of making a small contribution
to possible improvements/fixes. I already reported these two issues (dummy
index versus global variables, and nested sums not equal to two-indices
form) to the support team of Wolfram Research.

Please if anyone has further ideas besides those of Andrzej and Albert, I
will certainly be happy to read about them

Best regards
Jose
Mexico


-----Mensaje original-----
De: Albert Retey [mailto:awnl at gmx-topmail.de] 
Enviado el: Viernes, 08 de Agosto de 2008 06:15
Para: mathgroup at smc.vnet.net
Asunto: [mg91172] Re: Workaround for an unexpected behavior of Sum

Hi,

> Yes Andrzej, I agree, you are right, I can see know very important
> advantages of your approach over mine:
> 
> ONE: Using the flag, you avoid an infinite loop (actually, infinite
> recursion). The equivalent action in my approach was to verify if
> Unevaluated[dummyindex] is the same as the evaluated dummyindex. In my
case,
> the evaluation of dummyindex could take a long time. We actually cannot
know
> in advance how much time would that evaluation take, it depends whatever
the
> end user has stored in dummyindex, which could be a long program that
takes
> a lot of time to evaluate. Even worst, the evaluation of dummyindex could
> have unwanted results or side effects, again it depends on the content of
> dummyindex. Therefore my approach is potentially inefficient and
potentially
> dangerous. On the other hand, in your approach you only verify if the flag
> is True or False: always fast, always safe, always reliable, great!

I'm not in a position to criticize Adrzejs approach, but the following
solution seems to achieve the same in a different way and has some
advantages in my opinion, so it might also be of interest.

By using ValueQ (or accessing OwnValues directly) you can easily decide
whether or not a Symbol has a value without evaluating. So something
like this would also do the job:

Unprotect[Sum];
Sum[
   sumand_,
   before___,
   {dummyindex_Symbol /; ValueQ[dummyindex], rest___},
   after___
   ] := Module[{dummyindex},
   Sum[sumand, before, {dummyindex, rest}, after]
   ];
Protect[Sum];

I'm sure this approach has it's limits too (there could be additional
trouble with UpValues or DownValues of the index), but I like it for the
following reasons:

1) no dependence on the state of a global variable, which for whatever
reason could be left in an inappropriate state (see e.g. nested sums below).
2) it is minimal invasive: it only intervenes when obviously necessary
3) it works also with nested sums:

j=1;k=2;
Sum[Sum[f[j], {j, 1, k}], {k, 1, n}]

j = 1; k = 2;
Sum[f[j, k], {k, 1, n}, {j, 1, k}]

>> This confirms that the behavior of Sum is a bug.
>>
>> Andrzej Kozlowski

After all a reliable solution can only be achieved by Wolfram in this
case, everything else is just a workaround...

hth,

albert




  • Prev by Date: Re: "stepwise" integrating a 2D array
  • Next by Date: Re: More Inquiries
  • Previous by thread: Re: Workaround for an unexpected behavior of Sum
  • Next by thread: Re: Workaround for an unexpected behavior of Sum