Hi!
I tried to implement the SIR model taking into account the fact that I will only use 3 MPI ranks at this moment.
I built vectors and matrices following the examples already available. In particular, I defined the functions required similarly (RHSFunction, IFunction, IJacobian), as follows:

static PetscErrorCode RHSFunction(TS ts,PetscReal t,Vec X,Vec F,void *ctx)
{ 
  PetscErrorCode    ierr;
  AppCtx            *appctx = (AppCtx*) ctx;
  PetscScalar       f;//, *x_localptr; 
  const PetscScalar *x;
  PetscInt          mybase;
  
  PetscFunctionBeginUser;
  ierr = VecGetOwnershipRange(X,&mybase,NULL);CHKERRQ(ierr);
  ierr = VecGetArrayRead(X,&x);CHKERRQ(ierr);
  if (mybase == 0) {
    f    = (PetscScalar) (-appctx->p1*x[0]*x[1]/appctx->N);
    ierr = VecSetValues(F,1,&mybase,&f,INSERT_VALUES);
  }
  if (mybase == 1) {
    f    = (PetscScalar) (appctx->p1*x[0]*x[1]/appctx->N-appctx->p2*x[1]);
    ierr = VecSetValues(F,1,&mybase,&f,INSERT_VALUES);
  }
  if (mybase == 2) {
    f    = (PetscScalar) (appctx->p2*x[1]);
    ierr = VecSetValues(F,1,&mybase,&f,INSERT_VALUES);
  }
  ierr = VecRestoreArrayRead(X,&x);CHKERRQ(ierr);
  ierr = VecAssemblyBegin(F);CHKERRQ(ierr);
  ierr = VecAssemblyEnd(F);CHKERRQ(ierr);
  PetscFunctionReturn(0);
}


Whilst for the Jacobian I did:


static PetscErrorCode IJacobian(TS ts,PetscReal t,Vec X,Vec Xdot,PetscReal a,Mat A,Mat B,void *ctx)
{ 
  PetscErrorCode    ierr;
  AppCtx            *appctx = (AppCtx*) ctx;
  PetscInt          mybase, rowcol[] = {0,1,2};
  const PetscScalar *x;
  
  PetscFunctionBeginUser;
  ierr = MatGetOwnershipRange(B,&mybase,NULL);CHKERRQ(ierr);
  ierr = VecGetArrayRead(X,&x);CHKERRQ(ierr);
  if (mybase == 0) {
    const PetscScalar J[] = {a + appctx->p1*x[1]/appctx->N, appctx->p1*x[0]/appctx->N, 0};
    ierr = MatSetValues(B,1,&mybase,3,rowcol,J,INSERT_VALUES);CHKERRQ(ierr);
  }
  if (mybase == 1) {
    const PetscScalar J[] = {- appctx->p1*x[1]/appctx->N, a - appctx->p1*x[0]/appctx->N + appctx->p2, 0};
    ierr = MatSetValues(B,1,&mybase,3,rowcol,J,INSERT_VALUES);CHKERRQ(ierr);
  }
  if (mybase == 2) {
    const PetscScalar J[] = {0, - appctx->p2, a};
    ierr = MatSetValues(B,1,&mybase,3,rowcol,J,INSERT_VALUES);CHKERRQ(ierr);
  }
  ierr    = VecRestoreArrayRead(X,&x);CHKERRQ(ierr);
  
  ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
  ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
  if (A != B) {
    ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
    ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
  }
  PetscFunctionReturn(0);
}

This code does not provide the correct result, that is, the solution is the initial condition, either using implicit or explicit methods. Is the way I defined these objects wrong? How can I fix it? 
I also tried to print the Jacobian with the following commands but it does not work (blank rows and error message). How should I print the Jacobian?

ierr = TSGetIJacobian(ts,NULL,&K, NULL, NULL); CHKERRQ(ierr);
ierr = MatAssemblyBegin(K,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(K,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatView(K,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);                            

I would very much appreciate any kind of help or advice.
Best,
Francesco

Il giorno 2 apr 2021, alle ore 04:45, Barry Smith <bsmith@petsc.dev> ha scritto:



On Apr 1, 2021, at 9:17 PM, Zhang, Hong via petsc-users <petsc-users@mcs.anl.gov> wrote:



On Mar 31, 2021, at 2:53 AM, Francesco Brarda <brardafrancesco@gmail.com> wrote:

Hi everyone!

I am trying to solve a system of 3 ODEs (a basic SIR model) with TS. Sequentially works pretty well, but I need to switch it into a parallel version. 
I started working with TS not very long time ago, there are few questions I’d like to share with you and if you have any advices I’d be happy to hear.
First of all, do I need to use a DM object even if the model is only time dependent? All the examples I found were using that object for the other variable when solving PDEs.

Are you considering SIR on a spatial domain? If so, you can parallelize your model in the spatial domain using DM. Splitting the three variables in the ODE among processors would not scale.

 Even though it will not scale and will deliver slower performance it is completely possible for you to solve the 3 variable problem using 3 MPI ranks. Or 10 mpi ranks. You would just create vectors/matrices with 1 degree of freedom for the first three ranks and no degrees of freedom for the later ranks. During your function evaluation (and Jacobian evaluation) for TS you will need to set up the appropriate communication to get the values you need on each rank to evaluate the parts of the function evaluation needed by that rank. This is true for parallelizing any computation.

 Barry





Hong (Mr.)  

When I preallocate the space for the Jacobian matrix, is it better to decide the local or global space?

Best,
Francesco