Re: Simplify and Noncommutativity
- To: mathgroup at smc.vnet.net
- Subject: [mg60372] Re: Simplify and Noncommutativity
- From: "Carl K. Woll" <carlw at u.washington.edu>
- Date: Wed, 14 Sep 2005 03:27:44 -0400 (EDT)
- Organization: University of Washington
- References: <dfrhi4$g4l$1@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
"Robert Schoefbeck" wrote:
>I have a rather lengthy expression of abstract products of matrices
> of the form
>
> myDot[M1,M2,...]
>
> If Inv[M] denotes the inverse matrix
>
> i have told mathematica that
>
> myDot[P1___,P2_,Inv[P2_],P3___]:=myDot[P1,P3];
> myDot[P1___,Inv[P2_],P2_,P3___]:=myDot[P1,P3];
> myDot[]=Unity;
>
> and that
>
> myDot[P1___,P2_+P3_,P4___]:=myDot[P1,P2,P4]+myDot[P1,P3,P4];
>
> and a lot more things.
>
> My Problem is:
>
> In big expressions i have huge cancellations of the form
>
> myDot[M1,Inv[M1+M2+M3+....]] + myDot[M2,Inv[M1+M2+M3+....]]
> + myDot[M3 ,Inv[M1+M2+M3+....]]+...
>
> such that the summands M1,M2.... should be summed and then cancel
> against the Inv[...] part.
>
> I have a very slow workaround,
>
> myDotSimp[HoldPattern[Plus[P6___, myDot[P5___, P1_, P3___],
> myDot[P5___,P2_, P3___]]]] := P6 + myDot[P5, P1 + P2, P3];
>
> SetOptions[Simplify, TransformationFunctions ->
> {Automatic,myDotSimp}];
>
> this thing, however, is immensly time consuming.
>
> On the other hand, cancellations of the type
> Simplify[b/(b+c)+c/(b+c)]
> are extremly fast.
> Is there a way to combine the power of Simplify on Rational functions
> with a noncommutative multiplication?
>
> kind regards
> robert schoefbeck
>
One approach is to use regular multiplication, but to change your variables
by adding ordering information. Here are the definitions:
Clear[ov]
Format[ov[ord_, var_]] := var
ov /: ov[i_, x_] + ov[i_, y_] := ov[i, x + y]
ov /: ov[i_, x_] ov[j_, Inv[x_]] := ov[Max[i, j], 1] /; Abs[i - j] == 1
ov /: a_. ov[i_, 1] := a /. ov[j_ /; j > i, x_] -> ov[j - 2, x]
Since it is cumbersome to type in ov for each variable, I create a rule that
converts noncommutative multiplication using CenterDot into regular
multiplication with ov variables. However, since CenterDot can't be
displayed in plain text, for the purposes of this post I use
NonCommutativeMultiply instead:
Unprotect[NonCommutativeMultiply];
NonCommutativeMultiply[a__] := Times @@ MapIndexed[ov[First[#2], #1] &, {a}]
Protect[NonCommutativeMultiply];
If you try it out, use CenterDot (or something else that looks nice) and you
won't need the Unprotect/Protect statements.
Now, let's try out your problem:
In[50]:=
p2**p1**Inv[p1+p3+p4]**p5+p2**p3**Inv[p1+p3+p4]**p5+p2**p4**Inv[p1+p3+p4]**p5
Out[50]=
p2 p1 Inv[p1+p3+p4] p5+p2 p3 Inv[p1+p3+p4] p5+p2 p4 Inv[p1+p3+p4] p5
Notice that the proper order of the factors is maintained. The displayed
form of ov[1,p2] is p2, but the ordering is based on the first argument of
ov.
Now, let's factor:
In[51]:=
Factor[%]
Out[51]=
p2 p5
Just what you wanted. Now, for a few comments on the ov definitions. It is
possible that the ov definitions can by very slow due to combinatorial
explosion associated with the pattern matcher. If this happens, I have more
complicated definitions which will avoid the combinatorial explosion and
will hence be much quicker. Second, the rule with a_. ov[i_,1] is there to
reindex all of the ov variables, so that stuff like
x y Inv[y] Inv[x]
will simplify. If you like this approach, and decided to experiment, I would
be happy to have a dialog about your results.
Carl Woll
Wolfram Research