# Convect( ) in parallel
**URL:** https://community.freefem.org/t/convect-in-parallel/1270
**Category:** General Discussion
**Created:** [October 21, 2021, 10:32am UTC](https://community.freefem.org/t/convect-in-parallel/1270 "2021-10-21T10:32:10Z")
**Posts on this page:** 6
**Page:** 1
### Author:  [@yongxing](https://community.freefem.org/u/yongxing)
#### Post date: [October 21, 2021, 10:32am UTC](https://community.freefem.org/t/convect-in-parallel/1270/1 "2021-10-21T10:32:10Z")
Just to confirm that `convect( )` has already been implemented in parallel?
For example, if `[ux,uy]` is convected **from one partition of mesh to another,** has this been considered already in the parallel implementation so that one can directly use this function?
`convect([uox,uoy],-dt,ux), convect([uox,uoy],-dt,uy)`
I did a simple test, and it seems that this has **NOT** been implemented. Can you confirm this please @prj .
If this has not been implemented, is there a problem of your parallel fluid code @mojtaba.barzegari ? [GitHub - mbarzegary/navier-stokes-solver: Parallel Navier-Stokes solver for coupled diffusion-convection PDEs](https://github.com/mbarzegary/navier-stokes-solver)
---
### Author:  [@prj](https://community.freefem.org/u/prj)
#### Post date: [October 21, 2021, 11:40am UTC](https://community.freefem.org/t/convect-in-parallel/1270/2 "2021-10-21T11:40:23Z")
Just look at the examples… [FreeFem-sources/convect.edp at develop · FreeFem/FreeFem-sources · GitHub](https://github.com/FreeFem/FreeFem-sources/blob/develop/examples/hpddm/convect.edp)
Using `convect` inside a `varf` is wrong, and you’ll get incorrect results for high velocities/timesteps.
---
### Author:  [@yongxing](https://community.freefem.org/u/yongxing)
#### Post date: [October 21, 2021, 11:43am UTC](https://community.freefem.org/t/convect-in-parallel/1270/3 "2021-10-21T11:43:14Z")
Great! Thank you, I just realised this problem…
---
### Author:  [@yongxing](https://community.freefem.org/u/yongxing)
#### Post date: [October 21, 2021, 1:41pm UTC](https://community.freefem.org/t/convect-in-parallel/1270/4 "2021-10-21T13:41:22Z")
one more question:
What is about computing gradient inside `varf`, such as `dx(u_n)`, with `u_n` being the previous step velocity?
Should I compute `dx(u_n)` first, then `exchange( ..)`, and then put it into `varf` ?
---
### Author:  [@prj](https://community.freefem.org/u/prj)
#### Post date: [October 21, 2021, 1:42pm UTC](https://community.freefem.org/t/convect-in-parallel/1270/5 "2021-10-21T13:42:22Z")
No, that is fine, you don’t need to exchange.
---
### Author:  [@mojtaba.barzegari](https://community.freefem.org/u/mojtaba.barzegari)
#### Post date: [October 28, 2021, 12:46pm UTC](https://community.freefem.org/t/convect-in-parallel/1270/6 "2021-10-28T12:46:59Z")
sorry for my late reply.
@yongxing thank you for pointing this out. I had already noticed that the code doesn’t work well for larger time steps, but since I use small time steps, I didn’t investigate it. maybe, it’s a better idea to switch to newton method and a SNES solver for NS equations, and I should update the code to work on that basis.
I used such technique (putting `convect` inside `varf`) because I has seen a wide range of FF examples for solving NS writing it in that way (like [this](https://github.com/FreeFem/FreeFem-sources/blob/master/examples/3d/NSI3d-carac.edp), [this](https://github.com/FreeFem/FreeFem-sources/blob/master/examples/mpi/NSCaraCyl.edp), and [this](https://github.com/FreeFem/FreeFem-sources/blob/master/examples/mpi/NSI3d-carac-mumps.edp) to name a few), but apparently it has the problem @prj mentioned (at least in parallel).
for clarifying this to people seeing this later, I prepared these minimal examples, one with `convect` in `varf` and one using `convectParallel`. they produce the same results in small time steps, but the results will be different in larger time steps.
first example:
```auto
// using convect in the varf statement
load "PETSc"
macro dimension()2// EOM
include "macro_ddm.idp"
real dt = 0.02;
int n = 200;
mesh Mesh = square(n, n);
fespace SpaceP1(Mesh, P1);
SpaceP1 c, cold;
c = exp(-5*((x-0.3)^2 + (y-0.3)^2));
Mat A;
createMat(Mesh, A, P1)
c = c;
// SpaceP1 u1 = 0.3, u2 = 0.3;
SpaceP1 u1 = y, u2 = x;
varf convection(c, v) = intN(Mesh)(c/dt*v)
+ intN(Mesh)(convect([u1, u2], -dt, cold)*v/dt);
matrix Loc;
real[int] b(SpaceP1.ndof);
Loc = convection(SpaceP1, SpaceP1);
A = Loc;
set(A, sparams="");
for (int i=0; i < 50; i++)
{
if (mpirank == 0)
cout << "step " + i << endl;
cold = c;
b = convection(0, SpaceP1);
c[] = A^-1 * b;
savevtk("vel-1.vtu", Mesh, c, dataname="c", append = i?true:false);
}
```
second example:
```auto
// using convectParallel
load "PETSc"
macro dimension()2// EOM
include "macro_ddm.idp"
real dt = 0.02;
int n = 200;
mesh Mesh = square(n, n);
fespace SpaceP1(Mesh, P1);
SpaceP1 c;
c = exp(-5*((x-0.3)^2 + (y-0.3)^2));
buildDmesh(Mesh);
c = c;
// SpaceP1 u1 = 0.3, u2 = 0.3;
SpaceP1 u1 = y, u2 = x;
for (int i=0; i < 50; i++)
{
if (mpirank == 0)
cout << "step " + i << endl;
convectParallel(Mesh, [u1, u2], -dt, c, 20)
savevtk("vel-2.vtu", Mesh, c, dataname="c", append = i?true:false);
}
```