Re: [AMBER] random number seed - range

From: Paul S. Nerenberg <psn.berkeley.edu>
Date: Sat, 24 Jul 2010 18:11:00 -0700

Hi GT,

I'd like to add to what Ross, Bill, and Tom have stated. First, as
Bill has already said, you may be overlooking the thermostat
information that is printed out further down in the output. For
example, here is a pmemd (AMBER 10) run that I recently did using ig =
-1:

           -------------------------------------------------------
           Amber 10 SANDER 2008
           -------------------------------------------------------

| PMEMD implementation of SANDER, Release 10

| Run on 06/16/2010 at 18:17:23

---snip---

  Here is the input file:

Aax3 + water: 50 ns NPT MD at 300 K, 1 atm
  &cntrl
   imin = 0, irest = 1, ntx = 5,
   iwrap = 1, ntc = 2, ntf = 2,
   ntb = 2, cut = 10.0, ntr = 0,
   ntp = 1, pres0 = 1.0, taup = 5.0,
   ntt = 2, vrand = 1000, ig = -1, temp0 = 300.0,
   nstlim = 25000000, dt = 0.002,
   ntpr = 500, ntwx = 500, ntwr = 5000
  /

---snip---

Anderson (strong collision) temperature regulation:
      ig = 28163, vrand = 1000
      temp0 = 300.00000, tempi = 0.00000

---snip---

As you can see from this example, the random seed used for the
thermostat is printed down in the temperature regulation section of
the output and *not* up where the input file is displayed. [Side
note: the above run was conducted using pmemd in AMBER 10, so I can
verify that pmemd in AMBER 10 *does* support ig = -1, although since
you (GT) want to do TI, you're assumably using sander instead.]

Finally, as Tom has said, the output using random seed n should be
sufficiently different from random seed n+1, provided that the random
number generator is "good". But if you are uncomfortable with that,
you can always use your shell's built-in random number generator
(i.e., the $RANDOM variable) to generate a new random number each time
you create a new mdin file. When I do TI calculations, I first
execute a shell script that generates all of the relevant mdin files
at once (basically a customized version of the one given in the TI
tutorial on the AMBER website) and I use ig=$RANDOM within that
script. That way I can ensure that the same random seed is
automatically placed in the mdin files for both the V0 and V1 systems.

Best,

Paul


On Jul 24, 2010, at 4:12 PM, Ross Walker wrote:

> Hi GT,
>
> Are you using PMEMD or Sander? If I remember correctly pmemd in
> AMBER 10
> does not support the use of ig=-1. Only Sander does. PMEMD from
> AMBER 11
> supports the use of ig=-1.
>
> All the best
> Ross
>
>> -----Original Message-----
>> From: g t [mailto:sketchfoot.gmail.com]
>> Sent: Saturday, July 24, 2010 12:46 PM
>> To: AMBER Mailing List
>> Subject: Re: [AMBER] random number seed - range
>>
>> Hi!
>>
>> Thanks for your quick reply. I checked the outfiles when ig was set
>> to
>> -1
>> and they stated that ig had been set to -1 every instance, instead
>> of a
>> random number based upon the current time & date. I'm afraid I no
>> longer
>> have those files to post though.
>>
>> My version of AMBER is 10 and is running on linux Red Hat Enterprise
>> Linux
>> Server release 5.4. I'm afraid I do not know the compiler for sure,
>> but
>> suspect they may have used the comercial intel compiler & fortran
>> options.
>>
>> I think the issue is probably due to the TI limitation you speak
>> off in
>> the
>> second paragraph. If I do have to specify the ig value manually, what
>> do you
>> think would be an acceptable range?
>>
>> Thanks again,
>> GT
>>
>> On 24 July 2010 14:22, case <case.biomaps.rutgers.edu> wrote:
>>
>>> On Sat, Jul 24, 2010, g t wrote:
>>>>
>>>> I am ruinning a TI calculation. When I set the random number seed
>> value
>>> to:
>>>>
>>>> ig = -1
>>>>
>>>> The number is not set randomly based on the current date & time.
>>>
>>> First, what is you evidence that "ig=-1" is not using a different
>> seed for
>>> each run? (The seed chosen is printed in the output file). This
>> sounds
>>> like
>>> a bug, but we need to know the version of Amber, the compiler
>> (+version)
>>> and
>>> OS you are using.
>>>
>>> Second, note that you generally cannot use "ig=-1" for TI, since the
>>> integration scheme (which depends on ig for ntt=3) needs to be the
>> same
>>> for each of the two groups. So, for TI, you have to manually choose
>> a
>>> different seed for each run, and make it the same in the two mdin
>> files.
>>>
>>> ....dac
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jul 24 2010 - 18:30:04 PDT
Custom Search