How to make commands in Mathematica 8 use all cores?
From the Parallelize doumentation, under Examples > Possible Issues:
Expressions that cannot be parallelized are evaluated normally:
Parallelize[Integrate[1/(x - 1), x]]
As mentioned in the other questions and comments, things like Integrate
and Simplify
would be really difficult to parallelize, so Mathematica returns the message Parallelize::nopar1
and proceeds "with sequential evaluation."
(Although on reflection, maybe FullSimplify
could be parallelized, since it basically works by trying lots of different rules and doing leafcounts on them...)
If you have many integrals or simplifications to do, then you could use ParallelTable
or ParallelMap
etc...
As a trivial example, if you have the integrands
In[1]:= ints = Table[x^n, {n, 1, 10}]
Out[1]= {x, x^2, x^3, x^4, x^5, x^6, x^7, x^8, x^9, x^10}
You can use ParallelTable
In[2]:= ParallelTable[Integrate[int, x], {int, ints}]
Out[2]= {x^2/2, x^3/3, x^4/4, x^5/5, x^6/6, x^7/7, x^8/8,\
x^9/9, x^10/10, x^11/11}
or ParallelMap
In[3]:= ParallelMap[Integrate[#, x] &, ints]
Out[3]= {x^2/2, x^3/3, x^4/4, x^5/5, x^6/6, x^7/7, x^8/8,\
x^9/9, x^10/10, x^11/11}
Obviously for small lists of integrals like the above, the parallelization overhead is probably larger than benefit. But if you have really large lists and complex integrals, then it's probably worth it.
Edit in response to comments
Given the really messy integrand that the OP is interested in (note: you should really simplify your results as you go!), here's some code that breaks the integral into a sum of monomials and performs the integrals using ParallelDo
.
First we import the integral from pastebin
In[1]:= import = Import["http://pastebin.com/raw.php?i=JZ0CXewJ", "Text"];
extract the integration domain
In[2]:= intLimits = Rest@(2 Pi^5 ToExpression[StringReplace[import, "Integrate" -> "List"]])
vars = intLimits[[All, 1]];
Out[2]= {{\[Theta]3, 0, 2*Pi}, {\[Theta]2, 0, 2*Pi},
{\[Theta]1, 0, 2*Pi}, {\[CurlyPhi]2, 0, Pi/2}, {\[CurlyPhi]1, 0, Pi/2}}
and the integrand, which comes as the sum of 21 monsterous terms
In[4]:= integrand = First@(2 Pi^5 ToExpression[StringReplace[import, "Integrate" -> "Hold"]]);
Length[integrand]
LeafCount[integrand]
Out[5]= 21
Out[6]= 48111
We need to break the horrible mess down into bite sized chunks. First we extract all of the different functions from the integral
In[7]:= (fns=Union[vars, Cases[integrand, (Cos|Sin|Tan|Sec|Csc|Cot)[x_]/;!FreeQ[x,Alternatives@@vars],Infinity]])//Timing
Out[7]= {0.1,{\[Theta]1, <snip> ,Tan[\[CurlyPhi]2]}}
We find the (13849 nonvanishing) coefficients of monomials constructed from fns
In[8]:= coef = CoefficientRules[integrand, fns]; // Timing
Length@coef
Out[8]= {35.63, Null}
Out[9]= 13849
Check that all of the coefficients are free of any integration variables
In[10]:= FreeQ[coef[[All, 2]], Alternatives@@vars]
Out[10]= True
Note that we can actually clean up the coefficients using Factor
or Simplify
and decrease the ByteSize
by about 5 times...
But since the integrals of most of the monomials are zero, we might as well leave simplifications until the very end.
This is how you reconstruct a monomial, integrate it and recombine with its coefficient, for example, the 40th monomial gives a nonvanishing integral:
In[11]:= monomialNum=40;
Times@@(fns^coef[[monomialNum,1]])
Integrate[%, Sequence@@intLimits]
coef[[monomialNum,2]] %//Factor
Out[12]= \[Theta]1 Cos[\[Theta]1]^2 Cos[\[CurlyPhi]1]^4 Cos[4 \[CurlyPhi]1] Cos[\[CurlyPhi]2]^4 Cos[2 \[CurlyPhi]2] Sin[\[Theta]1]^2
Out[13]= \[Pi]^6/256
Out[14]= -((k1^2 (k1-k2) (k1+k2) (-2+p) p^3 \[Pi]^6 \[Sigma]^4)/(131072 \[Omega]1))
For now I'll reduce the number of terms, since it would take forever to do all of the integrals on my dual-core laptop. Delete or comment out the following line when you want to evaluate the whole set of integrals
In[15]:= coef = RandomChoice[coef, 100]; (* Delete me!! *)
OK, initialize an empty list for the monomial integration results
In[16]:= SetSharedVariable[ints]
ints = ConstantArray[Null, Length@coef];
As we perform the integrals, we Print
out
num: {timing, result}
for each monomial integrated.
The CellLabel
of each printed cell tells you which core did the integral.
The printing can get annoying - if it does annoy you,
then replace Print
with PrintTempory
or ##&
.
You could also monitor the calculation using a Dynamic variable of some sort: e.g. a progress bar.
ParallelDo[Print[c, ": ", Timing[
ints[[c]] = Integrate[Times@@(fns^coef[[c,1]]), Sequence@@intLimits]]],
{c, Length@coef}]
Combine with their coefficients
1/(2 Pi^5) Simplify[ints.coef[[All, 2]]]
And (hopefully) that's that!