Re: 1D PDE solver for Octave

[Bill Greene]

I just checked in a Makefile that works for me on linux.

But my strong recommendation, at this point, is just to use the pre-built

shared library I provided the link to.

As you recall, the main point of my original post is how to deal with these build

issues in a reasonable way. My belief is that very few Octave users will try this

pde solver if they have to build it themselves.

On Mon, Jul 4, 2016 at 4:30 AM, Juan Pablo Carbajal <address@hidden> wrote:

On Thu, Jun 30, 2016 at 7:56 PM, Bill Greene <address@hidden> wrote:
>>sure, were is the code?
> OK, great.
>
> The source code is here:
> https://github.com/wgreene310/pde1d
>
> Unfortunately, my approach to building this is somewhat ad-hoc
> at this point so you might find that challenging. I can send you a
> Linux Makefile if you are sufficiently motivated.
>
You should distribute your code with a makefile if you have the
intention that other people test it.

> Alternatively, I have a pre-built Linux binary here:
>
> https://drive.google.com/folderview?id=0B0HVh_MUHujMU0MzY2FxSlVUWWs&usp=sharing
>
> that seems to work fine with Octave 4.02 on my system.
>
> Let me know if you run into problems with either approach.
>
> Thanks.
>
> Bill
>
> On Thu, Jun 30, 2016 at 9:54 AM, Carlo De Falco <address@hidden>
> wrote:
>>
>>
>> On 30 Jun 2016, at 15:08, Bill Greene <address@hidden> wrote:
>>
>> > Thanks for taking the time to reply although I am disappointed
>> > that you didn't address any of my issues.
>>
>> As I said in my previus email I needed more info in order to address
>> some of them.
>>
>> > I am using the C API to ida; my code is written in C++.
>>
>> This is the same approach Franceso is using in ode15i/ode15s.
>>
>> There may be lots of code duplication, it would be nice to compare
>> your implementations ...
>>
>> > My code is currently using a simple linear interpolation in the
>> > spatial dimension and then using ida to solve the resulting system
>> > of ODE.
>>
>> So to summarize, given a parabolic PDE you discretize in space
>> using linear finite elements / first order finite differences
>> the solve the resulting ODE with BDF time-stepping via IDA,
>> is this correct?
>>
>> > I intend to improve the spatial discretization in the future
>> > but the current implementation could certainly be considered "inferior"
>> > to pdepe. I'm not trying to duplicate pdepe-- just support the same
>> > problem class and the command line interface as closely as possible.
>>
>> If my interpretation of your algorithm above is correct, it sounds
>> like, in order to reduce code duplication, this could be implemented
>> like a thin wrapper around BIM and ode15i or daspk, but we need to see
>> the code to be more specific.
>>
>> > I should say one more thing about Sundials ida. The latest version of
>> > ida is only just "adequate" to support my current implementation,
>> > mainly because its sparse matrix support is rather rudimentary. I am
>> > hoping they will improve ida in this area and expect that for the
>> > foreseeable future I'll want to be using the very latest ida release.
>> >
>> > I am aware of the ongoing Octave ode15s work; I alluded to that in my
>> > original
>> > message.
>>
>> I missed that, sorry. Replying below.
>>
>> >> I know that some work is ongoing using Sundials ida to improve the
>> >> Octave ODE solvers. But I'm unclear on when or if this might result
>> >> in the Sundials libraries being available in Octave.
>>
>> The changes for linking Octave to Sundials in Francesco's code are
>> essentially
>> done, but I'm not sure whether they'll be merged into core for the 4.2 or
>> 4.4
>> release ...
>>
>> If you want to start using those you can just clone Francesco's repo
>> from bitbucket and build Octave from there.
>>
>> >
>> > I would be happy for anyone to try the current code. At this point, I
>> > consider it to be solid beta-quality code.
>>
>> sure, were is the code?
>>
>> >
>> > Thanks.
>> >
>> > Bill
>>
>>
>> c.
>
>