Error when adding particles in Fortran

[HoulinLee]

Dear all,
I’m a Fortran user of AMReX, and I use the functionalities of Particles in AMReX. In my codes, I consider the ionization collision between particles, which means that new particles will be added during the simulation. I add particles by the following procedure (PC_ele_particles is the ParticleContainer of electrons and PC_ion_particles is the ParticleContainer of ions):

type(amrex_particle), pointer :: particles(:)
particles=>PC_ele_particles%get_particles(lev,mfi) 
elec_num=size(particles)
do ll = 1, elec_num
! Determine whether the ionization occurs
if (an ionization collision occurs) then
……
! Add a new electron
call amrex_set_particle_cpu(amrex_get_cpu(),p_new1)
call amrex_set_particle_id(amrex_get_next_particle_id(),p_new1)
call PC_ele_particles%add_particle(lev,mfi,p_new1)
! Add a new ion
call amrex_set_particle_cpu(amrex_get_cpu(),p_new2)
call amrex_set_particle_id(amrex_get_next_particle_id(),p_new2)
call PC_ion_particles%add_particle(lev,mfi,p_new2)
……
enddo

However, I found that an error occurs when the frequency of ionization collision is high. There is no specific error message but the codes suddenly stop and produce a file of Backtrace.0 to tell me the location where the error occurs. After some tests, I found that any call to the pointer particles(ll), such as particles(ll)%pos, will cause a segment fault after the call of PC_ele_particles%add_particle. Also, I found the error disappear if I call PC_ele_particles%add_particle and PC_ion_particles%add_particle outside the cycle of do ll = 1, elec_num enddo. So, I wonder if there is a restriction for the use of add_particle, such as one could not add particles in the cycle where particles, which is a pointer of amrex_particle, is used. Thanks.

[atmyers]

I think the issue is that add_particle invalidates the pointer because it can potentially trigger a resize in the underlying data vector. After calling it, you need to do particles=>PC_ele_particles%get_particles(lev,mfi) again before indexing in to particles or otherwise using it.

[HoulinLee]

Thanks for your reply. I'll try as you suggested.