Bug: Segmentation fault in `Exx_LRI::cal_exx_ions` during hybrid functional (LRC_WPBE) SCF calculation for 2D h-BN

[AsTonyshment]

Description

I encountered a segmentation fault while performing an SCF calculation for a 2D hexagonal boron nitride (h-BN) system using the lrc_wpbe hybrid functional.

Under the default exx_erfc_omega = 0.3 and a 24x24x1 K-point grid, the program runs normally (though it struggles to converge, which is expected for hybrid functionals). However, the program crashes with a segmentation fault (KILLED BY SIGNAL: 6 (Aborted)) under two specific parameter modifications:

1. Case A: Changing exx_erfc_omega to 0.126 (keeping the 24x24x1 K-point grid).
2. Case B: Keeping the default exx_erfc_omega = 0.3, but reducing the K-point grid to 8x8x1.

To Reproduce

Run the SCF calculation using the provided input files.

1. INPUT

INPUT_PARAMETERS
pseudo_dir            ./
orbital_dir           ./
basis_type            lcao
ecutwfc               100
calculation           scf
scf_thr               1e-7
device                cpu

dft_functional        lrc_wpbe
exx_erfc_omega        0.126

2. KPT

K_POINTS
0
Gamma
24 24 1 0 0 0

3. STRU

ATOMIC_SPECIES
B  10.8100 B_ONCV_PBE-1.0.upf upf201
N  14.0070 N_ONCV_PBE-1.0.upf upf201

NUMERICAL_ORBITAL
B_gga_10au_100Ry_2s2p1d.orb
N_gga_10au_100Ry_2s2p1d.orb

LATTICE_CONSTANT
1.8897260000

LATTICE_VECTORS
        2.5127465422        0.0000000000        0.0000000000
       -1.2563732711        2.1761025724        0.0000000000
        0.0000000000        0.0000000000       30.0000000000

ATOMIC_POSITIONS
Direct

B #label
0.0000   #magnetism
1 #number of atoms
        0.0000000000        0.0000000000        0.5000000000 m 0 0 0

N #label
0.0000   #magnetism
1 #number of atoms
        0.3333333333        0.6666666667        0.5000000000 m 0 0 0

Error Logs

In both failure cases, the program aborts right before the SCF iterations start, immediately after initializing potentials.

Standard Output

...
 DONE(0.279238    SEC) : INIT PLANEWAVE
 START CHARGE      : atomic
 DONE(0.416281    SEC) : CHARGE
 DONE(12.0791     SEC) : POTENTIALS

 ================================================================
 SELF-CONSISTENT: 
 ================================================================
terminate called recursively
terminate called recursively
terminate called recursively

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 0 PID 3579772 RUNNING AT cn81042
=   KILLED BY SIGNAL: 6 (Aborted)
===================================================================================

Source Code Investigation

To narrow down the issue, I inserted std::cerr print statements inside the Exact Exchange evaluation interface for LibRI (source/source_lcao/module_ri/Exx_LRI.hpp).

I found that the segmentation fault is triggered exactly at the this->exx_objs[settings_list.first].cv.cal_Cs_dCs(...) function call within the cal_exx_ions method. Specifically, the crash happens here:

// source/source_lcao/module_ri/Exx_LRI.hpp (around line 140)
std::pair<std::map<TA, std::map<TAC, RI::Tensor<Tdata>>>,
    std::map<TA, std::map<TAC, std::array<RI::Tensor<Tdata>, 3>>>>
        Cs_dCs = this->exx_objs[settings_list.first].cv.cal_Cs_dCs(
            ucell,
            list_As_Cs.first, list_As_Cs.second[0],
            {{"cal_dC",PARAM.inp.cal_force||PARAM.inp.cal_stress},
            {"writable_Cws",true}, {"writable_dCws",true}, {"writable_Vws",false}, {"writable_dVws",false}});

A print statement immediately before this block executes successfully, but a print statement immediately after it is never reached. While the root cause might stem from memory allocation or data structures passed into cal_Cs_dCs earlier in the workflow, this is the exact point of failure.

Environment

• ABACUS Version: v3.9.0.25 (Branch: develop)
• Commit: d999af798 (Wed Mar 4 16:09:43 2026 +0800)
• Hardware: CPU / Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz (x2)
• Compiler: IntelLLVM (icpx) 2024.0.2 (-fiopenmp enabled)
• MPI: Intel MPI (API Version 3.1)
• MKL: Intel MKL 2024.0 (libmkl_intel_lp64.so)
• Libxc: v6.2.2
• ELPA: Enabled (libelpa_openmp.so)