Quantum Espresso

Quantum ESPRESSO DFT Calculations

You are executing Quantum ESPRESSO density functional theory calculations.

CRITICAL: Pseudopotentials Are NOT Pre-Installed

You must find and download pseudopotentials yourself.

Pseudopotentials are NOT stored locally. You need to:

1. Determine what elements and functional you need
2. Find appropriate pseudopotentials online
3. Download them to your workspace
4. Reference them in your input file

How to Acquire Pseudopotentials

Step 1: Determine requirements

Element(s): Si, O, Li, etc.
Functional: PBE (most common), LDA, PBEsol
Type: NC (norm-conserving), US (ultrasoft), PAW (projector augmented wave)

Step 2: Find pseudopotentials online

Step 3: Download the files

Use WebFetch or Playwright:

Search for: "silicon PBE pseudopotential SSSP download"
Navigate to SSSP table, find Si row, download .UPF file

Example file names:

• Si.pbe-n-rrkjus_psl.1.0.0.UPF (PBE, ultrasoft)
• Si.pz-vbc.UPF (LDA, norm-conserving)
• O.pbe-n-kjpaw_psl.1.0.0.UPF (PBE, PAW)

Step 4: Save to workspace

workspaces/your-project/pseudo/Si.pbe-n-rrkjus_psl.1.0.0.UPF

Step 5: Note recommended cutoffs

SSSP provides recommended cutoffs. If not available:

• NC: ecutwfc ~40-80 Ry, ecutrho = 4 × ecutwfc
• US: ecutwfc ~30-50 Ry, ecutrho = 8-12 × ecutwfc
• PAW: ecutwfc ~40-60 Ry, ecutrho = 8-12 × ecutwfc

Step 6: Reference in input

&CONTROL
    pseudo_dir = './pseudo'
/
ATOMIC_SPECIES
Si  28.0855  Si.pbe-n-rrkjus_psl.1.0.0.UPF

Complete Agentic Workflow

Example: Silicon Band Structure

Given only: "Calculate the band structure of silicon"

You do:

1. Get crystal structure

   # From Materials Project
   from mp_api.client import MPRester
   import os
   with MPRester(os.environ.get("MP_API_KEY")) as mpr:
       si = mpr.get_structure_by_material_id("mp-149")
       # Note: a = 5.431 Å, diamond structure, Fd-3m
   

2. Find and download pseudopotential

   • Search: "silicon PBE pseudopotential SSSP"
   • Navigate to SSSP table
   • Download Si.pbe-n-rrkjus_psl.1.0.0.UPF
   • Note: recommended ecutwfc = 30 Ry

3. Create SCF input

   &CONTROL
       calculation = 'scf'
       prefix = 'si'
       outdir = './tmp'
       pseudo_dir = './pseudo'
   /
   &SYSTEM
       ibrav = 2                   ! FCC
       celldm(1) = 10.26           ! a in Bohr (5.43 Å)
       nat = 2
       ntyp = 1
       ecutwfc = 40.0              ! From SSSP recommendation
       ecutrho = 320.0             ! 8x for US pseudo
   /
   &ELECTRONS
       conv_thr = 1.0d-8
   /
   ATOMIC_SPECIES
   Si  28.0855  Si.pbe-n-rrkjus_psl.1.0.0.UPF
   
   ATOMIC_POSITIONS crystal
   Si  0.00  0.00  0.00
   Si  0.25  0.25  0.25
   
   K_POINTS automatic
   8 8 8 0 0 0
   

4. Run SCF

   /path/to/pw.x < scf.in > scf.out
   

5. Create bands input

   &CONTROL
       calculation = 'bands'
       prefix = 'si'
       outdir = './tmp'
       pseudo_dir = './pseudo'
   /
   ...
   K_POINTS crystal_b
   5
   0.5  0.5  0.5   20  ! L
   0.0  0.0  0.0   20  ! Gamma
   0.5  0.0  0.5   20  ! X
   0.5  0.25 0.75  20  ! W
   0.5  0.5  0.5   1   ! L
   

6. Run bands and post-process

   pw.x < bands.in > bands.out
   bands.x < bands_pp.in > bands_pp.out
   

7. Analyze

   • Extract band gap from output
   • Si has indirect gap ~0.5 eV (LDA underestimates, exp ~1.1 eV)

Binary Locations

QE binaries are referenced through environment variables, set in config.yaml (keep machine-specific paths out of the repo):

• CPU (MPI): $QE_CPU/pw.x. Parallel runs: mpirun -np N $QE_CPU/pw.x -in input.in (set OMP_NUM_THREADS=1 for multi-rank runs to avoid OpenMP oversubscription).
• GPU: $QE_GPU/pw.x. Note some GPU builds (e.g. NVHPC OpenACC+CUDA) are serial-only by design — do not mpirun them; check your build.
• Sanity-check a new build on a bulk-Si SCF (CPU and GPU energies should agree to ~1e-8 Ry) before trusting production runs.
• The old WSL-era builds under ~/work/archive/gpu-tests-wsl/ remain archive-only and unsupported — do not route runs to them.

Archived WSL-era builds (historical record; do NOT execute): The old dead builds live under ~/work/archive/gpu-tests-wsl/1-GPUTests/dft-qe/ ($QE_CPU/$QE_GPU no longer point there — repointed to the 7.5 builds, see .claude/settings.json env and config.yaml). Known failure signatures, kept for diagnosis:

• CPU build: error while loading shared libraries: libmpi_mpifh.so.40 (exit
  127. — the honest root-cause symptom (OpenMPI-4 runtime rot).
• GPU build: liblapack_lp64.so.0 missing — a red herring; the root cause is the same NVHPC/OpenMPI runtime rot, not LAPACK.
• Do not source .../env/setup_nvhpc.sh and retry: under the current ~/hpc-sdk hpcx env the binaries hang indefinitely in MPI init instead of failing fast.

Alternatives if local QE is unavailable

1. Vast.ai cloud — vast-cloud skill, "QE on VAST" section (build-from-source recipe; pay-per-hour, no queue).
2. CU Alpine — compute-strategy skill → backends/alpine.md (QE via modules on the cluster; owner-sanctioned SSH session required).
3. MLIP screening substitute — for tasks where DFT accuracy is not the point (relative-energy screening, structure triage), mlip-simulation / torch-sim can stand in; state the substitution and its accuracy limits explicitly in your report. (Requires the ML stack installed — check --verify.)

Calculation Types

Crystal Structure Input

Common ibrav Values

From Materials Project or CIF

If you get a structure from Materials Project or a CIF file:

ibrav = 0  ! General cell

CELL_PARAMETERS angstrom
5.431  0.000  0.000
0.000  5.431  0.000
0.000  0.000  5.431

ATOMIC_POSITIONS angstrom
Si  0.000  0.000  0.000
Si  1.358  1.358  1.358
...

Cutoff Selection

Always check pseudopotential recommendations.

If not available, use these guidelines:

Test convergence for production calculations.

K-point Selection

For band structure, use crystal_b with high-symmetry path.

Output Parsing

# Total energy
grep "!" output.out

# Forces
grep -A 20 "Forces acting" output.out

# Fermi energy
grep "Fermi" output.out

# Band gap (for insulators)
grep "highest occupied" output.out

Common Issues

1. "Error reading pseudo file"

   • Check pseudo_dir path
   • Verify file was downloaded correctly
   • Check filename matches ATOMIC_SPECIES

2. Convergence failure

   • Reduce mixing_beta to 0.3-0.5
   • Increase ecutwfc
   • Check structure for overlapping atoms

3. Memory issues

   • Reduce k-points
   • Use disk_io = 'low'

4. Negative frequencies in phonons

   • Structure not fully relaxed
   • Reduce forc_conv_thr and re-relax

Key Principle

Never assume pseudopotentials exist locally.

Every QE calculation requires you to:

1. Identify the elements
2. Choose appropriate functional
3. Find and download pseudopotentials
4. Note recommended cutoffs
5. Reference correctly in input

If a pseudopotential doesn't exist for your element/functional combination, that's important information to report.

Example: Formation Energy Calculation

Task: Calculate formation energy of NaCl

Complete workflow:

1. Get structure

   # From Materials Project
   mp-22862 → NaCl rock salt structure
   

2. Download pseudopotentials

   Na.pbe-spn-kjpaw_psl.1.0.0.UPF (ecutwfc=66 Ry)
   Cl.pbe-n-rrkjus_psl.1.0.0.UPF (ecutwfc=45 Ry)
   

3. Run SCF for NaCl

   • ecutwfc = 66 Ry (use max of both elements)
   • k-points: 6x6x6 automatic
   • Check convergence

4. Run reference calculations

   • Na metal (BCC structure)
   • Cl₂ molecule (in large box)

5. Calculate formation energy

   E_f = E(NaCl) - E(Na_metal) - 0.5*E(Cl₂)
   

6. Compare to literature

   • Expected: ~-4.2 eV/atom
   • If different by >10%, investigate

Common pitfall: Forgetting reference calculations. You need ALL of:

• The compound (NaCl)
• The elemental references (Na metal, Cl₂)

See also: examples/workflows/multi-compound-study.md for multi-compound studies