Update tutorial to use ADMaterialProperty in consumers

[lindsayad]

This was reported by a user here

We happen to know here that the producer of the _permeability is not using any non-linear variables in its calculation, but we want to teach our users that in their AD consumer objects they should (almost) always use ADMaterialProperty. It is consistent with our modular philosophy because who knows if sometime in the future we want to change the _permeability to be calculated based on non-linear variables.

[moosebuild]

Job Documentation on 3354045 wanted to post the following:

View the site here

This comment will be updated on new commits.

[tophmatthews]

When is the case where you don't want to use ADMaterialProperty? Isn't there memory overhead? If there isn't, why not just make all materials AD?

[aeslaughter]

So, I made this part of the tutorial as such exactly for the reason of showing and explaining the difference between the two. I don't really want to get rid of it unless we are actually going to deprecate the non-AD version.

[lindsayad]

When is the case where you don't want to use ADMaterialProperty? Isn't there memory overhead? If there isn't, why not just make all materials AD?

After #12599 I believe the memory expense incurred by AD is not all that high. The greatest memory expense for materials occurs when we have stateful material properties. After #12599, there is still an additional memory cost to having AD, but it's only 17 bytes to 8 bytes. Moreover, this expense is actually incurred no matter what. All material properties are created as AD material properties behind the scenes. It has to be this way to preserve inter-operability between non-AD and AD objects. Which brings us to...

So, I made this part of the tutorial as such exactly for the reason of showing and explaining the difference between the two. I don't really want to get rid of it unless we are actually going to deprecate the non-AD version.

This is perfectly mine with me. So perhaps instead I'll just put some doxygen comments inside the header to explain why we have the difference here.

[lindsayad]

When is the case where you don't want to use ADMaterialProperty? Isn't there memory overhead? If there isn't, why not just make all materials AD?

There is CPU expense. The expense of computing a DualNumber*DualNumber vs. a double*DualNumber is twice as large. However, depending on what you care about, this may mean less to you than increased modularity. However, obviously there's a time and place to care about speed! That's perhaps the number one reason why hand-coded materials may stay around for a long time. @friedmud achieved faster Jacobian computation using AD vs hand-coded in his Julia code, so maybe we can get there some-day. Certainly roystgnr/MetaPhysicL#53 will give us greatly increased AD flexibility, but it does not speed up existing AD simulations

[tophmatthews]

I'm all for AD. It has really changed the game for my work. It's good to know to just make EVERYthing an AD material.

[tophmatthews]

And automatic scaling...that should be automatically turned on it kicks so much ass!!!!!

[lindsayad]

You’ve experienced benefits from AS? (That’s how I refer to automatic scaling in my brain as the AD analog :-)

…

On Aug 21, 2019, at 5:13 PM, Topher Matthews ***@***.***> wrote: And automatic scaling...that should be automatically turned on it kicks so much ass!!!!! — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[hugary1995]

Wow! Do you know the reason for AD being faster than hand-coded Jacobian in Julia?

[tophmatthews]

Yeah, AD and AS made an unsolvable problem solvable with thermo-mechanics. At first I couldn't understand why it wasn't running any nonlinear iterations, and I thought it was doing worse. Turns out my tolerances were so loose to get anything to work before that it wasn't even taking a single iteration. Once I cranked them down, it flew by the difficult spots.

[lindsayad]

Glad to hear!

…

On Thu, Aug 22, 2019 at 6:40 AM Topher Matthews ***@***.***> wrote: Yeah, AD and AS made an unsolvable problem solvable with thermo-mechanics. At first I couldn't understand why it wasn't running any nonlinear iterations, and I thought it was doing worse. Turns out my tolerances were so loose to get anything to work before that it wasn't even taking a single iteration. Once I cranked them down, it flew by the difficult spots. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#13948?email_source=notifications&email_token=ACOGA4BKIHQ6XDJY7YIMSQDQF2JMDA5CNFSM4IOOLMZ2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD45D6RI#issuecomment-523910981>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACOGA4DVAPKD3AGPDJ2KP7DQF2JMDANCNFSM4IOOLMZQ> .