Fix: Bug in EXX PW, nspin=2 due to wrong k-point index (#6721)

Flying-dragon-boxing · mohanchen · WHUweiqingzhou · web-flow · commit 812c485d9669 · 2025-11-21T08:07:25.000+08:00
* modify for changed gint interface * correct nspin related behaviors * add efermi passthrough * Revert "add efermi passthrough" This reverts commit d7b402d. * commits to resolve conversations related to codes * DM and EDM pointers in pexsi now handled by diagopexsi, and copying h s matrices no longer needed * add pexsi examples * fix pexsi unit test (original version shouldn't run) * add building docs for pexsi * set cxx standard to c++14, which is required in make_unique * Fix: Fix typo related to pexsi * update to PPEXSIDFTDriver2 * default npoints to 1, so single core pexsi will work * Feature: exx operator for pw basis, single kpt * apply pexsi changes(?) * q-e style exx_div * Correct exxdiv * Fix Compile errors * refactor to abandon `pdiagh` * Fix mu_buffer and nspin * HSE examples * Feature: Multi-K exx * Feature: Multi-K exx * Updates with latest * Remove redundant global vars * Update to v3.9.0 * Update to v3.9.0, now code works * Remove Redundant cal_exx_energy in esolver_ks_pw.cpp * Some mess * Minor Fixes * Fix separate loop and screening * Add EXX stress * EXX Energy??? * Multi-K is broken??? * Fix: Multi-K and stress * Feature: ACE for single-K * Feature: ACE should work for multi-K, but not for sure * Feature: ACE works. Next step is ACE energy. * Fix: adapt to the latest instruction for variable `conv_esolver` * Reconstruct: move exx_helper to hamilt_pwdft * Refactor: in ESolver_KS_PW, calculate deband in iter_finish, not in hamilt2density * Fix: make files in consistent with upstream * Fix: Now EXX PW doesn't depend on LibRI * Fix: Add input constraints for EXX PW * Fix: Remove redundant mpi barrier * Fix: Clean irrelevant files * Fix: Clean irrelevant files * Feature: add ace flag, exit on using gpu * Refactor: Phase 1 for refactoring exx energy * Feature: now ace calculates energy * Feature: enable exx energy * Fix: fix makefile compilation error * Fix: One minor fix for a segmentation fault * Tests: one integrate test for exx pw, only for verifying whether exx pw works * Revert "Tests: one integrate test for exx pw, only for verifying whether exx pw works" This reverts commit e7b606f. * Fix: EXX PW ACE open only when separate_loop is on * add timer * Feature: Double Grid method of EXX PW * Feature: Double Grid method of EXX PW Stress * Fix: Double Grid method of EXX PW Stress * Feature: add double grid variable * Feature: add double grid variable * Fis: HSE stress * Fix: HSE Stress * Fix: Timer * Fix: Timer * For non mp sampling, disable extrapolation * Modify test * Modify mp * Format * Format * Feature: nspin == 2 scf * Fix: nspin == 2 scf * Docs: EXX PW Docs * Feature: EXX PW for nspin=2 * Docs: EXX PW Docs * Docs: EXX PW Docs * Docs: EXX PW Docs, minor fixes * Refactor * Refactor * Refactor * Refactor * Refactor * Refactor: fix unit test * Refactor: fix unit test * Refactor: fix unit test * Refactor: fix unit test * Bump version v3.9.0.7 * Refactor: Remove set kvec funcs in `K_Vectors` * Refactor: Remove final_scf * Refactor: Fix kvecc2d/d2c * Fix: Tests * Fix: Tests * Fix: Tests * Fix: Tests * Refactor: Final? * Fix * Fix * Fix * Fix * GPU EXX PW Support * Fix: Compile Error on CUDA > 12.9 * Fix: Compile Error on CUDA > 12.9 * NVTX3 * F***ing new version * Feature: Support linear combination of coulomb_param for EXX PW * Fix: Fix compile issue * F***ing new version * F***ing new version * F***ing new version * Uploading hybrid gauge tddft (#6369) * hybrid gague * update tests * update * update * update * update * update unit test * fix tests * update tests * fix read_wfc * fix catch_properties.sh * fix restart * update gpu test * update tests * fix * fix input_conv * Improve md calculation stress output in running log (#6366) * Improve md calculation stress output in running log * Module_IO Unittest modify * ModuleMD Unittests modify * modify code comment in fire_test.cpp * maintain setprecision(8) for md stress output * Refactor: Remove redundant Input_para from ESolver Class (#6370) * Refactor: Replace PARAM.inp with inp in ESolver classes for consistency * Refactor: Replace local input parameters with PARAM.inp in ESolver classes for consistency * Refactor: Use PARAM.inp.scf_ene_thr in ESolver_KS_LCAO iter_finish method * Revert "Refactor: Use PARAM.inp.scf_ene_thr in ESolver_KS_LCAO iter_finish method" This reverts commit b1bd0fd. * Revert "Refactor: Replace local input parameters with PARAM.inp in ESolver classes for consistency" This reverts commit f4f81e3. * Fix: Fix memory leak introduced by new gint module (#6375) * fix memory leak * delete copy assignment * refactor Exx_Opt_Orb (#6378) Co-authored-by: linpz <linpz@mail.ustc.edu.cn> * Add use sw and fix Floating point exception (#6372) * remove float error in sunway * fix ig=0 * add the sw * change the make_dir * unify the gg use * fix compile bug * add init * temporarily remove the sunway define * add the pesduo * fix compile bug * fix bug in the betar * modify the test * Update the output formats of rt-TDDFT (#6381) * update the output formats of rt-TDDFT * update the output formats of rt-TDDFT * fix a bug * update initialized velocities * found some output information is still lacking in MD module * [Refactor] Rename grid to module_grid and genelpa to module_genelpa (#6386) * Rename grid to module_grid * Rename genelpa to module_genelpa * Fix cmake * Update the outputs of geometry relaxation (#6387) * update the output formats of rt-TDDFT * update the output formats of rt-TDDFT * fix a bug * update initialized velocities * found some output information is still lacking in MD module * update output information * remove some global variables in relax_driver * update outputs * update relaxation outputs * update relaxation output messages * update tests of print info * fix a test * fix cg outputs * udpate cg test * update relax tests * update LCAO output stress format * change update_cell.cpp algorithm, when the ion move is larger than the cell length, it is fine to proceed the relaxation calculations * fix tests for unitcells * update cell * Feature: support the output of matrix representation of symm_ops (#6390) * Feature: support output the matrix representation of symmetry operation * Feature: support the output of matrix representation of symm_ops * update the document * Feature: Output real space wavefunction and partial charge density when `device=gpu` (#6391) * Fix GPU output of out_pchg and out_wfc_norm, out_wfc_re_im * GPU integrate test is functional again * Optimize RT-TDDFT dipole output (#6393) * Perf: support GPU version of cal_force_cc with LCAO basis (#6392) * support GPU version of cal_force_cc with LCAO basis * fix a bug * [Refactor] Move module_lr to source_lcao and add a new folder module_external in source_base (#6388) * Move module_lr to source_lcao * Fix test build * Move blas_connector to module_external * Fix header use * Fix internal header use * A fierce battle with Makefile😡 * Move blacs_connector.h to module_external * Move lapack_connector.h and lapack_wrapper.h to module_external * Fix header usage * Move scalapack_connector.h to module_external * Fix a bug for the output information after relaxation (#6395) * update the output formats of rt-TDDFT * update the output formats of rt-TDDFT * fix a bug * update initialized velocities * found some output information is still lacking in MD module * update output information * remove some global variables in relax_driver * update outputs * update relaxation outputs * update relaxation output messages * update tests of print info * fix a test * fix cg outputs * udpate cg test * update relax tests * update LCAO output stress format * change update_cell.cpp algorithm, when the ion move is larger than the cell length, it is fine to proceed the relaxation calculations * fix tests for unitcells * update cell * update some function names, update output A to Angstrom * change eV/A to eV/Angstrom * bump version to 3.9.0.10 (#6397) Co-authored-by: Liang Sun <50293369+sunliang98@users.noreply.github.com> * Fix: fix exx_gamma_extrapolation error in MPI * Fix: fix exx_gamma_extrapolation error in MPI * Update lapack.cu * Refactor: Use LAPACK interfaces from ATen * Fix: Integrate test * Fix: implement devinfo for potrf * Fix: MPI and Makefile * Fix: get_potential * Fix: ace * Refactor * Refactor * Refactor * Refactor * Refactor * Fix: conv * Revert "Fix: conv" This reverts commit d2da506. * Fix: conv * Fix: conv hard code thr for now * Fix: conv hard code thr for now * Fix: conv hard code thr for now * Fix: conv hard code thr for now * Refactor * Refactor * Refactor * Refactor * Refactor * Mod * Begin EXX KPAR * Begin EXX KPAR * Begin EXX KPAR * Begin EXX KPAR * Begin EXX KPAR * EXX KPAR WORKS * EXX KPAR WORKS Alternative * Fix GPU, but so ugly... * Undo cuda aware mpi * Undo cuda aware mpi * Revert "Undo cuda aware mpi" This reverts commit a8d71b2. * EXX KPAR WORKS on NSPIN=2 * Fix without MPI * Fix header * Separate ecut for exx pw * Documents related * Documents related * EXX KPAR BUG FIX * EXX KPAR BUG FIX --------- Co-authored-by: Mohan Chen <mohan.chen.chen.mohan@gmail.com> Co-authored-by: wqzhou <33364058+WHUweiqingzhou@users.noreply.github.com> Co-authored-by: HTZhao <104255052+ESROAMER@users.noreply.github.com> Co-authored-by: lanshuyue <140165754+lanshuyue@users.noreply.github.com> Co-authored-by: Liang Sun <50293369+sunliang98@users.noreply.github.com> Co-authored-by: dzzz2001 <153698752+dzzz2001@users.noreply.github.com> Co-authored-by: linpeize <linpeize2024@163.com> Co-authored-by: linpz <linpz@mail.ustc.edu.cn> Co-authored-by: liiutao <74701833+A-006@users.noreply.github.com> Co-authored-by: Mohan Chen <mohanchen@pku.edu.cn> Co-authored-by: Critsium <tsfxwbbzxy@163.com> Co-authored-by: kirk0830 <67682086+kirk0830@users.noreply.github.com> Co-authored-by: Taoni Bao <baotaoni@pku.edu.cn> Co-authored-by: Chen Nuo <49788094+Cstandardlib@users.noreply.github.com>
diff --git a/source/source_pw/module_pwdft/operator_pw/exx_pw_ace.cpp b/source/source_pw/module_pwdft/operator_pw/exx_pw_ace.cpp
@@ -119,7 +119,8 @@ void OperatorEXXPW<T, Device>::construct_ace() const
     {
         for (int ik0 = 0; ik0 < nk_max; ik0++)
         {
-            int ik = ik0 + ispin * nk_max;
+            int ik = ik0 + ispin * wfcpw->nks / nspin_fac;
+            // printf("ik: %d\n", ik);
             int npwk = wfcpw->npwk[ik];
 
             T* Xi_ace = Xi_ace_k[ik];
@@ -146,7 +147,12 @@ void OperatorEXXPW<T, Device>::construct_ace() const
                 // ik fixed here, select band n
                 for (int iq0 = 0; iq0 < nqs; iq0++)
                 {
-                    int iq = iq0 + ik;
+                    // For nspin=2, iq should be in the same spin channel as ik
+                    int iq;
+
+                    int nk = wfcpw->nks / 2;
+                    iq = iq0 + ispin * nk; // iq in the same spin channel
+
                     // for \psi_nk, get the pw of iq and band m
                     get_exx_potential<Real,  Device>(kv, wfcpw, rhopw_dev, pot, tpiba, gamma_extrapolation, ucell->omega, ik, iq);
 
@@ -157,7 +163,6 @@ void OperatorEXXPW<T, Device>::construct_ace() const
                     for (int m_iband = 0; m_iband < psi.get_nbands(); m_iband++)
                     {
                         double wg_mqb = 0;
-                        bool skip = false;
                         if (iq_pool == GlobalV::MY_POOL)
                         {
                             wg_mqb = (*wg)(iq_loc, m_iband);
diff --git a/source/source_pw/module_pwdft/operator_pw/op_exx_pw.cpp b/source/source_pw/module_pwdft/operator_pw/op_exx_pw.cpp
@@ -306,6 +306,8 @@ void OperatorEXXPW<T, Device>::act_op_kpar(const int nbands,
     setmem_complex_op()(psi_nk_real, 0, wfcpw->nrxx);
     setmem_complex_op()(psi_mq_real, 0, wfcpw->nrxx);
     int nqs = kv->get_nkstot_full();
+    int nspin_fac = PARAM.inp.nspin == 2 ? 2 : 1;
+    int ispin = this->ik < (wfcpw->nks / nspin_fac) ? 0 : 1;
 
     // ik fixed here, select band n
     for (int iq = 0; iq < nqs; iq++)
@@ -332,7 +334,7 @@ void OperatorEXXPW<T, Device>::act_op_kpar(const int nbands,
 
             if (iq_pool == GlobalV::MY_POOL)
             {
-                const T* psi_mq = get_pw(m_iband, iq_loc);
+                const T* psi_mq = get_pw(m_iband, iq_loc + ispin * wfcpw->nks / nspin_fac);
                 wfcpw->recip_to_real(ctx, psi_mq, psi_mq_real, iq_loc);
                 // send
             }

Original file line number	Diff line number	Diff line change
`@@ -119,7 +119,8 @@ void OperatorEXXPW<T, Device>::construct_ace() const`
`119`	`119`	`{`
`120`	`120`	`for (int ik0 = 0; ik0 < nk_max; ik0++)`
`121`	`121`	`{`
`122`		`- int ik = ik0 + ispin * nk_max;`
	`122`	`+ int ik = ik0 + ispin * wfcpw->nks / nspin_fac;`
	`123`	`+ // printf("ik: %d\n", ik);`
`123`	`124`	`int npwk = wfcpw->npwk[ik];`
`124`	`125`
`125`	`126`	`T* Xi_ace = Xi_ace_k[ik];`
`@@ -146,7 +147,12 @@ void OperatorEXXPW<T, Device>::construct_ace() const`
`146`	`147`	`// ik fixed here, select band n`
`147`	`148`	`for (int iq0 = 0; iq0 < nqs; iq0++)`
`148`	`149`	`{`
`149`		`- int iq = iq0 + ik;`
	`150`	`+ // For nspin=2, iq should be in the same spin channel as ik`
	`151`	`+ int iq;`
	`152`	`+`
	`153`	`+ int nk = wfcpw->nks / 2;`
	`154`	`+ iq = iq0 + ispin * nk; // iq in the same spin channel`
	`155`	`+`
`150`	`156`	`// for \psi_nk, get the pw of iq and band m`
`151`	`157`	`get_exx_potential<Real, Device>(kv, wfcpw, rhopw_dev, pot, tpiba, gamma_extrapolation, ucell->omega, ik, iq);`
`152`	`158`
`@@ -157,7 +163,6 @@ void OperatorEXXPW<T, Device>::construct_ace() const`
`157`	`163`	`for (int m_iband = 0; m_iband < psi.get_nbands(); m_iband++)`
`158`	`164`	`{`
`159`	`165`	`double wg_mqb = 0;`
`160`		`- bool skip = false;`
`161`	`166`	`if (iq_pool == GlobalV::MY_POOL)`
`162`	`167`	`{`
`163`	`168`	`wg_mqb = (*wg)(iq_loc, m_iband);`