ufl team mailing list archive

[Harish Narayanan <harish.mlists@xxxxxxxxx>]

Garth N. Wells wrote:
> On Nov 9 2009, Harish Narayanan wrote:
> 
>> Anders Logg wrote:
>>> On Mon, Nov 09, 2009 at 02:05:29PM +0100, Anders Logg wrote:
>>>> On Mon, Nov 09, 2009 at 12:58:32PM +0000, Garth N. Wells wrote:
>>>>>
>>>>> Marie Rognes wrote:
>>>>>> Kristian Oelgaard wrote:
>>>>>>> Quoting Anders Logg <logg@xxxxxxxxx>:
>>>>>>>
>>>>>>>
>>>>>>>> This is something we've discussed before, but I want to discuss it
>>>>>>>> again.
>>>>>>>>
>>>>>>>> Could we change the definition of the gradient of a vector to be
>>>>>>>> the
>>>>>>>> "normal" definition:
>>>>>>>>
>>>>>>>>   grad(u) = du_i / dx_j
>>>>>>>>
>>>>>>>> ?
>>>>>>>>
>>>>>>> I think this sounds like an excellent idea.
>>>>>>>
>>>>>>>
>>>>>> Me too.
>>>>>>
>>>>> Cowards! ;) You wait for Martin the leave and then jump on the
>>>>> bandwagon.
>>>>>
>>>>> I suggested this March/April, and Martin's argument against was for
>>>>> consistency with other operations. I think that it's fine to change so
>>>>> long that other operations do not break or lead to
>>>>> unexpected/inconsistent results.
>>>> :-)
>>>>
>>>> I've had this discussion with Martin at least twice and both times he
>>>> managed to convince me of his definition. But I keep forgetting the
>>>> arguments in favor of the current definition (and more so when it
>>>> gets me into trouble).
>>>
>>> I've created a blueprint:
>>>
>>> https://blueprints.launchpad.net/ufl/+spec/redine-grad
>>>
>>> Add your comments there. (Harish, add comments on div.)
>>
>> I will. And there might be something similar with curl too, but who uses
>> curl anyway! :)
>>
>> I'd just like to point out that in future discussion, I think it will be
>> clearer if you used the following notation grad(u)_ij = du_i / dx_j.
>>
> 
> Yes, otherwise we could have
>  grad(u)_ji = du_i / dx_j
> which would keep Martin happy :).

Exactly!

Harish