[lanshuyue]

Summary

• Initialize per-ionic-step DFT+U occupation-matrix output in shared ESolver_KS::iter_init() for both PW and LCAO.
• Remove duplicate preparation calls and file-output parameters from iter_init_dftu_pw() and finish_dftu_lcao().
• Rename dftu_base_io_read.cpp and dftu_base_io_write.cpp to dftu_base_read.cpp and dftu_base_write.cpp, updating CMake and legacy Makefile source lists. The renamed source contents are unchanged.
• In PW, when the occupation matrix was loaded from file, write occ_mat.txt at the first electronic step as well as the numbered snapshot. Both output calls now use the same readiness condition; fresh PW calculations still wait until the matrix is available.

Verification

• Earlier KS initialization commit: CMake configured with IntelLLVM 2024, MPI, LCAO and tests enabled; syntax compilation passed for its five changed C++ translation units. Existing neighboring-checkout PW/LCAO I/O test binaries passed 2/2. The earlier python3 tools/03_code_analysis/agent_governance_check.py --staged check reported no errors.
• Current commit, using this branch: cmake -S . -B /tmp/abacus-dftu-rename-build -DBUILD_TESTING=ON -DENABLE_LCAO=ON -DENABLE_MPI=ON -DENABLE_LIBXC=OFF -DENABLE_EXX=OFF -DENABLE_DEEPKS=OFF -DENABLE_CUDA=OFF -DENABLE_ROCM=OFF completed. CMake reported that the ENABLE_EXX/DEEPKS/CUDA/ROCM variables were unused.
• cmake --build /tmp/abacus-dftu-rename-build --target MODULE_PW_dftu_base_test MODULE_DFTU_base_io -j 8: both targets built.
• OMP_NUM_THREADS=1 /tmp/abacus-dftu-rename-build/source/source_pw/module_pwdft/unittests/MODULE_PW_dftu_base_test --gtest_brief=1: 27/27 passed.
• OMP_NUM_THREADS=1 /tmp/abacus-dftu-rename-build/source/source_lcao/module_dftu/unittests/MODULE_DFTU_base_io --gtest_brief=1: 6/6 passed.
• git diff --check and git diff --cached --check: passed.
• A full PW restart calculation was not run locally. The first-step loaded-matrix esolver gate has no dedicated integration test in this focused change.

Review notes

• No linked issue: this PR is a focused DFT+U output cleanup and PW/LCAO alignment.
• The PARAM.globalv.global_out_dir reference moved from the PW solver call site to the shared KS solver in the earlier commit; the PR-wide global dependency count is unchanged.
• No INPUT parameter or occupation-matrix file-format changes.
• Core module impact: ESolver_KS prepares ion-step output for both basis paths; PW iteration finish uses the existing matrix-readiness state for both outputs.
• Existing I/O tests cover file initialization and reader/writer behavior. No new test fixture was added for the source renames or one-line esolver readiness condition because constructing an ESolver_KS fixture would substantially expand this PR. Full PW/LCAO runtime validation remains for CI/review.
• No governance exception requested.

[mohanchen]

reopen when it is not a draft

[ABACUS-PR-BOT]

Static review at f34f8ee found no implementation defect in the shared output preparation, readiness gate, or source renames. A regression coverage suggestion follows. No builds, runtime tests or CI diagnostics were performed; author-reported checks were not independently rerun.

[ABACUS-PR-BOT]

Non-blocking: The existing tests exercise loaded readiness and output helpers separately, so they would still pass if this gate reverted to iter > 1 || istep > 0. Add a one-step file-loaded PW restart case that checks occ_mat.txt contains the first-step matrix, with a fresh-start control and a check that numbered output does not retain stale sections.