Quartz piezoelectric tensor: LEPSILON vs LCALCPOL differ by 2x rigid.aug. term

Problems running VASP: crashes, internal errors, "wrong" results.


Moderators: Moderator, Global Moderator

Post Reply
Message
Author
julian_juan
Newbie
Newbie
Posts: 1
Joined: Thu Oct 08, 2026 7:40 pm

Quartz piezoelectric tensor: LEPSILON vs LCALCPOL differ by 2x rigid.aug. term

#1 Post by julian_juan » Thu Nov 19, 2026 6:40 pm

VASP version: 6.5.0 (16Dec24)

Dear VASP developers and users,

We are computing the clamped-ion (electronic) piezoelectric tensor of α-quartz with three VASP routes (LEPSILON, LCALCEPS and finite-strain LCALCPOL). The two field-based routes agree with each other, but both differ from the finite-strain Berry-phase result. The difference is, to within 1×10⁻⁵ C/m², twice the parenthesised contribution printed in the "PIEZOELECTRIC TENSOR FIELD DIRECTION" blocks. Details follow.

System and settings

  • α-quartz, 9-atom cell (P3₂21 / P3₁21), LDA.

  • PAW Si 02Apr1999 and PAW O 31May2000.

  • ENCUT = 850 eV, PREC = Accurate, EDIFF = 1E-8, LASPH = .TRUE., LREAL = .FALSE., ISMEAR = 0.

  • Γ-centred 5×5×4 k-mesh (8×8×6 for the convergence checks); V = 111.069 ų.

  • All values below are in the Cartesian frame of the POSCAR, in C/m², and are clamped-ion (electronic) values.

Results (clamped-ion, C/m²)

Code: Select all

Method                                                                 e11 (x,XX)  e14 (x,YZ)  e11/e14
---------------------------------------------------------------------  ----------  ----------  -------
LEPSILON (IBRION = -1), 5x5x4                                            -0.16299    -0.07112    2.292
LEPSILON, 8x8x6                                                          -0.16301    -0.07113    2.292
LCALCEPS (EFIELD_PEAD = 0.01 eV/A), 5x5x4                                -0.16198    -0.07079    2.288
LCALCPOL, finite strain +/-0.005, 5x5x4 (*)                              -0.14504    -0.07011    2.069
LCALCPOL, finite strain +/-0.005, 8x8x6 (*)                              -0.14533    -0.07068    2.056
QE 7.5, norm-conserving (PseudoDojo LDA), Berry phase + finite strain    -0.14860    -0.07221    2.058
QE 7.5, norm-conserving, finite field (lelfield) + stress                -0.14827    -0.07201    2.059

The finite-strain values use proper differences: the dipoles are converted to reduced coordinates before taking the ±δ differences. For e11 and e14 the improper terms vanish identically, so this choice does not affect them.

(*) Note on the LCALCPOL runs. The finite-strain runs at both k-meshes were set up as ionic relaxations (IBRION = 2, ISIF = 2, NSW = 100), and LCALCPOL is evaluated only at the first ionic step. The clamped-ion values in the table are taken from that first step, i.e. the strained cell with the reduced coordinates of the unstrained structure; the later relaxation steps are not used. As a check, genuinely static clamped-ion runs (IBRION = −1, NSW = 0, same settings) were done for the rigidly translated crystal of the origin test and give e11 = −0.14512 and e14 = −0.07011 C/m² at 5×5×4, in agreement with the first-step values.

The per-field-direction block printed by LEPSILON reads:

Code: Select all

 PIEZOELECTRIC TENSOR FIELD DIRECTION 1  (e Angst)
    -1.12977 -0.00000  0.00000           ( -0.06126 -0.00000  0.00000)
    -0.00000  1.12967 -0.49303           ( -0.00000  0.06126 -0.00078)
     0.00000 -0.49303  0.00008           (  0.00000 -0.00078  0.00000)

The parenthesised XX value, −0.06126 e·Å, corresponds to −0.008837 C/m² with Ω = 111.069 ų. We assume it is the rigid-augmentation part, by analogy with the "(rigid.aug. ionic)" columns of the Born-charge block.

Observation. Subtracting twice the parenthesised term from the LEPSILON value reproduces the finite-strain result (C/m²):

Code: Select all

Component (8x8x6)  LEPSILON  -2 x (paren.)  Corrected  LCALCPOL  Difference
-----------------  --------  -------------  ---------  --------  ----------
e11                -0.16301       +0.01767   -0.14534  -0.14533        1e-5
e14                -0.07113       +0.00023   -0.07090  -0.07068        2e-4

For e14 the parenthesised term is tiny (−0.00078 e·Å), and the remaining difference is smaller than the k-point variation of the finite-strain result itself (0.00057 C/m² between 5×5×4 and 8×8×6).

What we have checked

  • k-points: LEPSILON changes by 0.01 % between 5×5×4 and 8×8×6, and LCALCPOL by 0.2 %.

  • Run settings: the LEPSILON result is the same with ISPIN = 1 or 2, with or without LCALCPOL in the same run, and with LPEAD = .TRUE. or .FALSE.

  • Origin: a rigid translation of the whole crystal (with DIPOL shifted accordingly) changes neither result.

  • Linearity: the ±δ finite strain is linear (p(+δ) ≈ −p(−δ)), and no polarization quantum is involved (Δn ~ 10⁻³).

  • Ionic relaxation: e(relaxed) − e(clamped) from LCALCPOL agrees with the ionic piezoelectric contribution from IBRION = 6 to 0.04 %. Born-charge and internal-strain responses therefore agree between the two routes.

  • Independent code: with norm-conserving pseudopotentials (no augmentation charges) in Quantum ESPRESSO, the strain-derivative route and the field–stress route agree with each other to 0.2 %. They also agree with VASP LCALCPOL to about 2 % (ε∞ and Z* from the two codes differ by about 0.5 %). They do not agree with LEPSILON or LCALCEPS.

We may be misinterpreting this block or missing a term in the comparison, so any clarification would be very helpful.
Questions

  1. What exactly is the parenthesised contribution in the "PIEZOELECTRIC TENSOR FIELD DIRECTION" blocks?

  2. Is its sign correct in the ion-clamped piezoelectric tensor computed from the stress response to the field? The discrepancy is exactly twice that term.

  3. Does LCALCEPS share the same term in its finite-field stress? Its result is within 0.6 % of LEPSILON.

Attached is vasp_quartz_e11.zip with the INCAR, POSCAR, KPOINTS and OUTCAR of the minimal set needed to reproduce the e11 discrepancy at 5×5×4: LEPSILON, LCALCEPS and the two LCALCPOL runs at S1 = ±0.005. A README.txt maps each folder to the values above. The remaining runs (8×8×6, S4 strain for e14, origin test) are available on request.

Thank you in advance for your help.

Best regards,

You do not have the required permissions to view the files attached to this post.

manuel_engel1
Global Moderator
Global Moderator
Posts: 276
Joined: Mon May 08, 2023 4:08 pm

Re: Quartz piezoelectric tensor: LEPSILON vs LCALCPOL differ by 2x rigid.aug. term

#2 Post by manuel_engel1 » Fri Nov 20, 2026 1:35 pm

Hello,

Thank you for posting on the VASP forum, and for your detailed bug report. This is indeed a bug. In many cases, the induced errors will be rather small, so it went unnoticed for a long time.

The term in parentheses is a PAW contribution: it comes from how the projections onto the PAW projectors change under strain. It is added with the wrong sign in both LEPSILON and LCALCEPS. That is why the two agree with each other but not with your finite-strain LCALCPOL results. The dielectric tensor and the Born effective charges are not affected.

Until this is fixed, your correction is the right one:

e(correct) = e(LEPSILON) - 2 x (term in parentheses)

We checked this for cubic BN and alpha-quartz. With the correction, LEPSILON agrees with finite-strain LCALCPOL to within 0.25%.

The fix will be in the next VASP release.

Manuel Engel
VASP developer


Post Reply