Re: [Jprogramming] Multidimensional Root Finding with Newton Solver

```There's also this - http://code.jsoftware.com/wiki/Scripts/nlls - an
implementation of the Levenberg-Marquardt algo that I cribbed from someone
who copied it from APL.```
```
On Thu, Aug 10, 2017 at 6:47 PM, Don Kelly <d...@shaw.ca> wrote:

> I also suggest that you look at references to the use of NR for power
> system  load flow problems which are non-linear and generally expressed in
> terms of complex numbers in the  polar format. These do converge well. I
> haven't written one in J but one written in APL has 11 readable lines-most
> of which are setting up the matrices to solve (with no complex number
> capability- J should be more compact).
>
> Don
>
>
>
> On 2017-08-10 1:10 PM, Ian Clark wrote:
>
>> Is it also possible to solve a system of equation like the following
>>>
>> one… ?
>>
>> Basically, yes.
>>
>> Because not only can x = (x1,x2) be a vector, but so can y = (y1,y2) in
>>
>>   y1 = a*(1-x1)
>>   y2 = b*(x2-x1^2)
>>
>> TABULA is an app (distributed as a JAL "addon") which employs
>> Newton-Raphson (occasionally) to "solve" systems of non-linear equations
>> supplied by the user. "Solve" includes replacing y2, say, with (y2+∆y2)
>> and
>> relying on TABULA to adjust x2, x1 and y1 accordingly.
>>
>> …And to do so without forcing a debate with the clueless user as to what
>> "accordingly" means here.
>>
>> As you'll be aware, N-R algorithms don't always converge, especially when
>> you don't have control over what the user throws at you. Pure
>> mathematicians do; engineers don't. Hyperbolic functions behave badly,
>> e.g.
>> y-->(y+∆y) in: y = k/x unless (∆y) is "small enough"… again there is no
>> point asking the poor user what "small enough" is supposed to mean in this
>> instance.
>>
>> For a problem that has no general solution, TABULA performs remarkably
>> well
>> with the systems of (sometimes non-linear) equations that physicists and
>> engineers typically need solving. Occasionally TABULA throws up its hands
>> in despair – as even the perfect app would have to, when faced with
>> y-->(y+∆y)
>> in: y = sin(x) whenever ∆y takes (y+∆y) outside the domain: [-1,1].
>>
>> I suggest you read these articles:
>>
>>     http://code.jsoftware.com/wiki/TABULA/LaunchElephant
>>     http://code.jsoftware.com/wiki/TABULA/ChurchClock-NEW
>>
>> as examples of the sort of empirical problem TABULA is designed to
>> Then examine the code of the addon: CAL, the engine used by TABULA to do
>> the real work. Particularly the verb: (inversion) and its ancillary (and
>> alternative) verbs.
>>
>> … in a similar elegant manner?
>>>
>> Since TABULA is an empirical tool for non-mathematicians, such as K-12
>> pupils and college students training to be physicists, engineers and
>> ecologists, it sacrifices a great deal of elegance in the interests of
>> generality and practicality, not to mention giving the user a
>> comprehensible answer when things go wrong. Thus in practice the "_" in "
>> N^:_" needs replacing by a finite value simply to timeout the infinite
>> iterations which can and do occur.
>>
>> Accordingly you'll see that the different variants of the verb (inversion)
>> are nowhere near as pretty as: N=: 1 : '- u % u d. 1'
>>
>> You don't say whether you are a mathematician (in particular a functional
>> analyst) or an engineer. Whichever it is, you'll either be baffled by, or
>> scornful of, the code in CAL.ijs. But it should give you a bit of a start
>> in whatever you want to do.
>>
>>
>> On Thu, Aug 10, 2017 at 7:07 PM, Martin <d...@famic.de> wrote:
>>
>> Hi there,
>>>
>>> J looks very interesting. I have no previous experience with array
>>> languages and, being curious, started to experiment.  Now, I would
>>> like to solve a system of non-linear equations. I could only examples
>>> solving single equations like this one:
>>>
>>>    N=: 1 : '- u % u d. 1'   NB. Adverb implementing Newton-Raphson
>>> iteration.
>>>    (_2 + *:) N^:_ ]1        NB. Find root of “0 = _2+x^2”, starting guess
>>> of “1”.
>>>
>>> Is it also possible to solve a system of equation like the following
>>> one in a similar elegant manner?
>>>
>>>    f1(x) = a*(1-x1)
>>>    f2(x) = b*(x2-x1^2)
>>>
>>> Example from https://www.gnu.org/software/gsl/doc/html/multiroots.html#
>>> examples
>>>
>>> Thanks for any ideas!
>>> -Martin
>>>
>>> ----------------------------------------------------------------------
>>> For information about J forums see http://www.jsoftware.com/forums.htm
>>>
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm
>>
>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
>

--

Devon McCormick, CFA

Quantitative Consultant
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm```