Re: locating overlow/underflow
- To: mathgroup at smc.vnet.net
- Subject: [mg112247] Re: locating overlow/underflow
- From: Leslaw Bieniasz <nbbienia at cyf-kr.edu.pl>
- Date: Mon, 6 Sep 2010 06:36:24 -0400 (EDT)
- References: <i5sucr$pej$1@smc.vnet.net>
Hi Bill,
Thanks for your comments. I have reduced the range of y, and I am now
obtaining results without any error messages. However, this exercise
raises my doubts whether the desired "arbitrary precision" is indeed
reached, especially when y approaches 100. The program output yields
infinity for the number of valid significant digits, but how many
digits can I really trust if I declare 70 digits? Shouldn't the output
contain warnings if the number of valid digits is smaller than 70?
I don't perhaps need as many as 70, something like 30 would be enough,
but I need to be sure that I really have that accuracy.
I don't know how all that works, but I can imagine MATHEMATICA should
automatically allocate appropriately more bytes to the floating point
representations used, if the current number of bytes is insufficient.
There are algorithms that can do that, I believe, like interval
arithmetics? How can I check whether the accuracy is indeed as I want it
to be?
Leslaw
On Sat, 4 Sep 2010, Bill Rowe wrote:
> On 9/3/10 at 6:10 AM, nbbienia at cyf-kr.edu.pl (Leslaw Bieniasz) wrote:
>
>> I am not very much familiar with MATHEMATICA, but I have to tabulate
>> a certain mathematical expression in a possibly large range of
>> independent variable, and with the highest possible number of
>> accurate digits. I have a notebook file that does the job (included
>> below in text form, together with the results) but during the cell
>> evaluation I obtain error messages stating that underflow or
>> overflow occurred. My question is: is there any way to rewrite the
>> code in such a way so that I can obtain an information where exactly
>> (that is for which values of independent variable) the errors
>> actually occur? Or, is there any way to rewrite the code or change
>> the program settings in such a way so that the errors do not occur
>> at all? I am using MATHEMATICA 6.0.
>
>> In[1]:= Table[{N[y],
>> SetPrecision[ 2 ((y - 1) Exp[y] Erfc[Sqrt[y]] + (y/3 - 1) 2
>> Sqrt[y/Pi] + 1)/(y y), 70], Precision[ 2 ((y - 1) Exp[y]
>> Erfc[Sqrt[y]] + (y/3 - 1) 2 Sqrt[y/Pi] + 1)/(y y)]
>> },
>> {y, { 1/10^19, ..., 5 10^15
>> }}]
>
> Where to start?
>
> SetPrecision simply doesn't do what you seem to think it does.
> Using the first value for y (1/10^19) I can get 50 digit
> precision by doing:
>
> In[18]:= f =
> 2 ((y - 1) Exp[y] Erfc[Sqrt[y]] + (y/3 - 1) 2 Sqrt[y/Pi] +
> 1)/(y y);
>
> In[19]:= N[f /. y -> 10^-19 // Simplify, 50]
>
> Out[19]= 0.99999999971454014148222328832683872045949424371145
>
> There is no need to compute this number twice to get the
> precision since doing N[expr, 50] will always return a number
> with a precision of 50 if all of the values in expr are exact values.
>
> Strictly speaking, I did not need to simplify your expression
> before converting it to a number with 50 digit precision. But
> experience indicates this is a good idea for a complex
> expression with extreme exact values.
>
> While this approach will work for several values of y, it will
> not work for all values.
>
> Notice for y = 100
>
> In[44]:= Log[10, Exp[{100, 200, 500}]] // N
>
> Out[44]= {43.4294,86.8589,217.147}
>
> and
>
> In[45]:= Log[10, Erfc[Sqrt[{100, 200, 500}]]] // N
>
> Out[45]= {-44.6802,-88.2591,-218.746}
>
> That is for y values of about 100, the first term
>
> (y - 1) Exp[y] Erfc[Sqrt[y]]
>
> will be about 1/10 the size of the second term
>
> (y/3 - 1) 2 Sqrt[y/Pi] + 1)
>
> But the next decade gives
>
> In[46]:= Log[10, Exp[{1000, 2000, 5000}]] // N
>
> Out[46]= {434.294,868.589,2171.47}
>
> and
>
> In[47]:= Log[10, Erfc[Sqrt[{1000, 2000, 5000}]]] // N
>
> Out[47]= {-436.043,-870.488,-2173.57}
>
> That is the first term decreases about one order of magnitude
> while the second one increase by one order of magnitude.
> Clearly, the first term can be dropped well before you get to
> the maximum value you have for y without any significant error.
>
> The bottom line is if you truly need 70 digits of precision for
> this expression with a value of y like 5x10^15 you will need to
> have a far more sophisticated approach than what you are using.
>
>
>