Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

Many commands in Mathematica 8 (Integrate, Simplify, etc.) seem to only be using a single core on my system. Is there any way I can change the affinity so that it utilizes all cores for computations?

share|improve this question
4  
My guess is that those commands can not be easily parallelized and therefore won't deliver better performance when run on different cores at the same time. That's why Wolfram didn't bother to implement it for parallel computing. I'm no expert in theoretical computer science and mathematics though, so I wouldn't know how to implement an integration algorithm. –  slhck Jul 26 '11 at 16:05

2 Answers 2

up vote 2 down vote accepted

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!

share|improve this answer
    
I have integral by five variables, ok for beginning I can integrate one by one, but the problem is how to use all cores for this one integral. Of course FullSimplify command also can not be parallelized. –  derdack Jul 28 '11 at 13:20
    
An iterated/multidimensional integral? If it's numeric, then you could parallelize a Monte Carlo evaluation. If it's an analytic result you want, you might be in trouble. What I meant by the FullSimplify comment is that the algorithm probably has room for some parallelization - not that you could get Mma to parallelize it in its current form. –  Simon Jul 28 '11 at 14:01
    
Problem is that I want analytic result. For a beginning I can decouple my multiple integral. But Analytic calculation for just one can be one month or I don't know how long, because I have computer with 16 cores and I can use just one for calculating. "Maple 15" wrote that they have parallel computation in this version, but I tested and for integrating also couldn't work on many cores in same time. maplesoft.com/products/maple/new_features/index.aspx#parallel –  derdack Jul 28 '11 at 15:39
    
Sorry, I mean my limits of integrals are numerical values. I have for example: Integrate [0..5] Integrate [0..8] Integrate [0..2](large function) dx dy dz. Large function is in the form: a*x+b*y+c*z, where a,b,c are analytical constants in form of a, b, c. Can use parallelize in this case? –  derdack Jul 28 '11 at 17:28
    
So you have an integral of the form int (a x + b y + c z) dx dy dz where a, b and c are independent of x, y, and z? Isn't that trivial - or am I missing something? –  Simon Jul 29 '11 at 1:37

From the Parallelize doumentation, under Examples > Possible Issues:

Expressions that cannot be parallelized are evaluated normally:

Parallelize[Integrate[1/(x - 1), x]]

enter image description here

share|improve this answer
    
Do you think that it is the same problem under the Linux? –  derdack Jul 26 '11 at 16:36
    
Because I got this message on Win7: ParallelCombine::nopar: No parallel kernels available; proceeding with sequential evaluation. >> –  derdack Jul 26 '11 at 16:40
1  
A "kernel" in computation is, simply speaking, only the part of an algorithm that is applied on the data over and over again. If there is no parallel kernel for a task, it can't be run in parallel. This is the same for every operating system and solely dependent on the software, i.e. Mathematica. –  slhck Jul 26 '11 at 16:48

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.