Bug: ABACUS LCAO memory cost increases during SCF iteration and causes OOM

[kirk0830]

Describe the bug

I am testing the very large basis set for small molecule. Each time my SCF can only run for 8 steps, then the calculation will be killed like OOM:

                                                                                     
                              ABACUS v3.8.2
               Atomic-orbital Based Ab-initio Computation at UStc                    
                     Website: http://abacus.ustc.edu.cn/                             
               Documentation: https://abacus.deepmodeling.com/                       
                  Repository: https://github.com/abacusmodeling/abacus-develop       
                              https://github.com/deepmodeling/abacus-develop         
                      Commit: 0d455cb (Thu Nov 7 10:08:20 2024 +0800)
 Tue Nov 12 17:38:24 2024
 MAKE THE DIR         : OUT.ABACUS/
 RUNNING WITH DEVICE  : CPU / Intel(R) Xeon(R) Platinum
 UNIFORM GRID DIM        : 864 * 864 * 864
 UNIFORM GRID DIM(BIG)   : 216 * 216 * 216
 DONE(71.6958    SEC) : SETUP UNITCELL
 DONE(72.3571    SEC) : INIT K-POINTS
 ---------------------------------------------------------
 Self-consistent calculations for electrons
 ---------------------------------------------------------
 SPIN    KPOINTS         PROCESSORS  THREADS     NBASE       
 1       Gamma           4           4           308         
 ---------------------------------------------------------
 Use Systematically Improvable Atomic bases
 ---------------------------------------------------------
 ELEMENT ORBITALS        NBASE       NATOM       XC          
 Cl      18s17p17d-6au   154         2           
 ---------------------------------------------------------
 Initial plane wave basis and FFT box
 ---------------------------------------------------------
 DONE(77.2381    SEC) : INIT PLANEWAVE
 -------------------------------------------
 SELF-CONSISTENT : 
 -------------------------------------------
 START CHARGE      : atomic
 DONE(252.931    SEC) : INIT SCF
 * * * * * *
 << Start SCF iteration.
 ITER       ETOT/eV          EDIFF/eV         DRHO     TIME/s
 GE1     -8.18827885e+02   0.00000000e+00   1.0112e-01 139.17
 GE2     -8.18862826e+02  -3.49406469e-02   2.3170e-02 138.44
 GE3     -8.18867751e+02  -4.92507973e-03   9.5219e-03 138.75
 GE4     -8.18869501e+02  -1.75062999e-03   4.3400e-03 139.12
 GE5     -8.18870165e+02  -6.63091835e-04   3.0995e-03 139.60
 GE6     -8.18870635e+02  -4.70805126e-04   2.3152e-03 140.63
 GE7     -8.18870793e+02  -1.57412850e-04   1.7121e-03 141.15
 GE8     -8.18870831e+02  -3.85875513e-05   1.2741e-03 142.04

When I check the trajectory of memory cost, I find there is a significant increasing during the SCF iteration:

I guess there is some memory leak. Please find the input files below:
Cl2-dimer.zip

Expected behavior

No response

To Reproduce

No response

Environment

No response

Additional Context

No response

Task list for Issue attackers (only for developers)

• Verify the issue is not a duplicate.
• Describe the bug.
• Steps to reproduce.
• Expected behavior.
• Error message.
• Environment details.
• Additional context.
• Assign a priority level (low, medium, high, urgent).
• Assign the issue to a team member.
• Label the issue with relevant tags.
• Identify possible related issues.
• Create a unit test or automated test to reproduce the bug (if applicable).
• Fix the bug.
• Test the fix.
• Update documentation (if necessary).
• Close the issue and inform the reporter (if applicable).

[WHUweiqingzhou]

Could you make some tests? @kirk0830