# [Trilinos-Users] Transforming Laplacians of Intrepid FEM basis functions

Roberts, Nathan V. nvroberts at alcf.anl.gov
Mon Mar 21 10:15:04 EDT 2016

Hello all,

I'm implementing an SUPG formulation for convection-diffusion, and the stabilization term for this involves a Laplacian.  So, I need to take second-order derivatives of the FEM basis functions in Intrepid.  I see how to do this in reference space; I can apply OPERATOR_D2 and use the Intrepid::getDkEnumeration() to navigate the returned values.

My question is this: how do I appropriately transform the values to physical space?  I.e., the second-order analogue to the FunctionSpaceTools::HGRADtransformGRAD()?

Below, I offer what I've worked out thus far—I'm hoping that someone can (a) confirm that what I have is reasonable, and (b) comment on the simplest/best way to approach this in terms of what's available in Intrepid.

Mathematically, what HGRADtransformGRAD() is doing goes something like this.  If the physical function is u and the reference-space function is \hat{u}, and there is a vector transformation function x = x(\xi) from reference to physical space, then the gradient of u can be computed as:

This can be computed by multiplying the inverse of the Jacobian of the reference-to-physical map by the reference space gradient values; this is what HGRADtransformGRAD does.

Proceeding to the second derivatives needed for the Laplacian, we have

I believe the first term on the right hand side is computable in terms of the reference-space OPERATOR_D2 values and the inverse of the Jacobian applied twice (I would be glad to see example code if anyone has that to offer).  The second term requires second derivatives of the reference-to-physical map (i.e. the Hessian), as well as first derivatives of the reference-space function.

Is that correct?  Can anyone comment on the best way to implement this?

Thanks!
Nate

P.S. In case my pasted PDF graphics do not come through, the LaTeX for the first equation is:
\frac{\partial u}{\partial x_i} = \frac{\partial \hat{u}}{\partial \xi_k} \frac{\partial \xi_k}{\partial x_i}

and the second is:
\frac{\partial^2 u}{\partial x_i^2} = \frac{\partial^2 \hat{u}}{\partial \xi_k \partial \xi_l} \frac{\partial \xi_k}{\partial x_i} \frac{\partial \xi_l}{\partial x_i} + \frac{\partial \hat{u}}{\partial \xi_k} \frac{\partial^2 \xi_k}{\partial x_i^2}

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://trilinos.org/pipermail/trilinos-users/attachments/20160321/f7f9e3e0/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PastedGraphic-1.pdf
Type: application/pdf
Size: 37088 bytes
Desc: PastedGraphic-1.pdf
URL: <https://trilinos.org/pipermail/trilinos-users/attachments/20160321/f7f9e3e0/attachment-0002.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PastedGraphic-2.pdf
Type: application/pdf
Size: 46037 bytes
Desc: PastedGraphic-2.pdf
URL: <https://trilinos.org/pipermail/trilinos-users/attachments/20160321/f7f9e3e0/attachment-0003.pdf>