MathGroup Archive 2002

[Date Index] [Thread Index] [Author Index]

Search the Archive

RE: On Defining Functions Symmetric wrt Some Indices

  • To: mathgroup at smc.vnet.net
  • Subject: [mg34328] RE: [mg34316] On Defining Functions Symmetric wrt Some Indices
  • From: "Wolf, Hartmut" <Hartmut.Wolf at t-systems.com>
  • Date: Wed, 15 May 2002 03:35:18 -0400 (EDT)
  • Sender: owner-wri-mathgroup at wolfram.com

> -----Original Message-----
> From: Alexei Akolzin [mailto:akolzine at uiuc.edu]
To: mathgroup at smc.vnet.net
> Sent: Tuesday, May 14, 2002 10:13 AM
> Subject: [mg34328] [mg34316] On Defining Functions Symmetric wrt Some Indices
> 
> 
> Hello,
> 
> For the purposes of formula simplification I need to specify that some
> function "f" is symmetric upon SOME of its indices. For example,
> f[a,b,c] == f[a,c,b] but NOT equal to f[b,a,c]. 
> 
> The proposed command SetAttribute[f,Orderless] makes the function
> symmetric wrt ALL of its indices, which I want to avoid.
> 
> Is there is a way to neatly solve this problem?
> 
> Thanks.
> 
> Alexei.
> 

Alexei,

from your question I suppose that you intend to use f merely as a container
to transform the ordering of the arguments. Otherwise, if you have a
definition for f, you were free to bring the arguments to any order you like
at rhs, e.g.

In[1]:= f[x_,y__]:={x}~Join~Sort[{y}]

In[2]:= f[a,b,c]===f[a,c,b]
Out[2]= True
In[3]:= f[a,b,c]===f[b,a,c]
Out[3]= False
In[4]:= Quit[]

But the problem with this presumably is just that head f is lost (and cannot
be transformed further). This will keep it
 
In[1]:= f[x_, y__] /; ! OrderedQ[{y}] := f[x, ##] & @@ Sort[{y}]

In[2]:= f[a, b, c] === f[a, c, b]
Out[2]= True
In[3]:= f[a, b, c] === f[b, a, c]
Out[3]= False
In[4]:= f[1, 2, 3] /. f[a_, c_, b_] :> {a, b, c}
Out[4]= {1, 2, 3}
In[5]:= f[1, 2, 3] /. HoldPattern[f[a_, c_, b_]] :> {a, b, c}
Out[5]= {1, 3, 2}

Deplorably Out[5] is not consistent with pattern matching of Orderless
attribute:
 
In[6]:= Attributes[g] = {Orderless};

In[7]:= g[1, 2, 3] /. g[a_, c_, b_] :> {a, b, c}
Out[7]= {1, 2, 3}
In[8]:= g[1, 2, 3] /. HoldPattern[g[a_, c_, b_]] :> {a, b, c}
Out[8]= {1, 2, 3}

In[9]:= Quit[]


Perhaps a good way to reach your ends would be to transform your expression
explicitely using a rule:
 
In[1]:=
normalizingRule = f[x_, y__] :> RuleCondition[f[x, Sequence@@Sort[{y}]]]

In[2]:= Unevaluated[f[a, b, c] === f[a, c, b]] /. normalizingRule
Out[2]= True
In[3]:= Unevaluated[f[a, b, c] === f[b, a, c]] /. normalizingRule
Out[3]= False
In[4]:= 
Unevaluated[f[1, 2, 3] /. f[a_, c_, b_] :> {a, b, c}] /. normalizingRule
Out[4]= {1, 2, 3}
In[5]:=
Unevaluated[
    f[1, 2, 3] /. HoldPattern[f[a_, c_, b_]] :> {a, b, c}] /.
normalizingRule
Out[5]= {1, 2, 3}
In[6]:= Quit[]

What is ugly with this is the need to deliberately hold your expressions
unless the rule is tried. But that can be done in a rather mechanical
fashion.

If you know in advance which arguments are not to be ordered (stretching
your example) perhaps you might try:

In[1]:= Attributes[f] = Orderless;
In[3]:= f[b, c][a] === f[c, b][a]
Out[3]= True
In[4]:= f[b, c][a] === f[a, c][b]
Out[4]= False
In[5]:= f[2, 3][1] /. f[c_, b_][a_] :> {a, b, c}
Out[5]= {1, 2, 3}
In[6]:= f[2, 3][1] /. HoldPattern[f[c_, b_][a_]] :> {a, b, c}
Out[6]= {1, 2, 3}
In[7]:= Quit[]

Another idea would be this
 
In[1]:= Attributes[f0] = Orderless;
In[2]:= f[x_, y__] := f[x][f0[y]]

In[3]:= f[a, b, c] === f[a, c, b]
Out[3]= True
In[4]:= f[a, b, c] === f[b, a, c]
Out[4]= False
In[5]:= f[1, 2, 3] /. f[a_, c_, b_] :> {a, b, c}
Out[5]= {1, 2, 3}
In[6]:= f[1, 2, 3] /. HoldPattern[f[a_, c_, b_]] :> {a, b, c}
Out[6]= f[1][f0[2, 3]]
In[3]:= Quit[]

yet having more disadvantages.

It is difficult to tell the "right way" unless you tell more about what you
finally intend. I would be surprised, if there were a simple and direct way
to reach that "partially orderless" property for f you quested.

--
Hartmut



  • Prev by Date: Re: Re: Limit with restrictions
  • Next by Date: Re: Assigning to superscripted variables
  • Previous by thread: RE: On Defining Functions Symmetric wrt Some Indices
  • Next by thread: RE: RE: On Defining Functions Symmetric wrt Some Indices