tests/e2e: add Quantum ESPRESSO and GROMACS, completing the CI tier

Four codes now run end to end in CI -- MOPAC, NWChem, Quantum ESPRESSO
and GROMACS -- every one of them Debian-packaged.  ORCA and Gaussian
stay a developer-box tier, being registration-walled and licensed.
ECCE_E2E_REQUIRE names all four, so a code whose binary is not where
the suite looks fails the job rather than skipping quietly.

Quantum ESPRESSO: bulk Si, under a second.  It needs pseudopotentials
to run at all and Debian's quantum-espresso-data-sssp installs 103 of
them where ai.qe already looks.  The assertions are the two things
worth guarding for a plane-wave code: the total energy must be a
Hartree value (QE prints Ry, so ~-22.6 would mean the conversion was
dropped) and STRESS must be a full symmetric 3x3 -- a transposition or
a dropped element shows up in the symmetry check and nowhere else.
This is the code whose units are carried in the section HEADER rather
than beside the numbers, which is how it once produced wrong values
rather than none.

GROMACS: 216 SPC waters, 50 steps of MD.  The deck deliberately runs MD
with temperature coupling rather than a minimisation, because that is
what makes the "Energies (kJ/mol)" table WRAP onto two rows -- TE in
the first, PRESSURE in the second.  CLAUDE.md records that shape as
producing wrong values rather than none, and a minimisation does not
wrap, so a simpler fixture would have tested nothing.  Verified by hand
first: the parser takes Total Energy (not Potential, which is a
plausible number one column away) and reads Pressure from the second
row, and [NULL] does suppress the trailing A V E R A G E S block, which
is the last such table in the file.

Two harness additions this needed:

  setup steps   GROMACS runs grompp before mdrun, so a code definition
                can declare preprocessing commands and extra input
                files to copy alongside the deck.
  stdout        pw.x writes to stdout and ECCE redirects it into
                qe.pwout; the harness captures it rather than making
                the case shell out, so it still runs the binary
                directly.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EFMUNbw2UpRPmgsqviAowb

Author: andy

.github/workflows/test.yml

@@ -97,15 +97,16 @@ jobs:
       - name: Install dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install -y perl python3 mopac nwchem nwchem-openmpi
+          sudo apt-get install -y perl python3 mopac nwchem nwchem-openmpi \
+            quantum-espresso quantum-espresso-data-sssp gromacs
 
       # Declare what this job installed.  Without this a code whose
       # binary is not where the suite looks simply skips, and the job
       # goes green with that coverage quietly gone -- which is exactly
       # what happened to nwchem on its first run here.
       - name: End-to-end suite
         env:
-          ECCE_E2E_REQUIRE: mopac,nwchem
+          ECCE_E2E_REQUIRE: mopac,nwchem,qe,gromacs
         run: tests/e2e/run_tests.py -v
 
   apps:

tests/e2e/README.md

@@ -127,9 +127,9 @@ conversion were dropped, and `ENTHALPY − ETHERM` must equal RT exactly.
 | Code | Tier | Notes |
 |---|---|---|
 | MOPAC | CI | Debian `mopac`; seconds per job |
-| NWChem | CI (todo) | Debian `nwchem-openmpi` |
-| Quantum ESPRESSO | CI (todo) | Debian |
-| GROMACS | CI (todo) | Debian |
+| NWChem | CI | Debian `nwchem-openmpi` |
+| Quantum ESPRESSO | CI | Debian `quantum-espresso` + `-data-sssp` for pseudopotentials |
+| GROMACS | CI | Debian `gromacs`; needs a grompp setup step |
 | ORCA | local only | registration-walled |
 | Gaussian | local only | licensed |
 

tests/e2e/cases.py

@@ -189,6 +189,96 @@ def expect_nwchem_h2o(props, report):
                  '(got %g)' % te)
 
 
+def expect_qe_si(props, report):
+    """Bulk silicon, PBE, 2 atoms -- QE's scalar and tensor results.
+
+    Quantum ESPRESSO is a plane-wave code: no basis set, no orbitals in
+    ECCE's sense.  What it does have that no other code here does is a
+    STRESS tensor, and that is worth guarding because the value carries
+    its units in the section HEADER rather than beside the numbers --
+    the shape that silently produced wrong values rather than none when
+    QE was first integrated.
+    """
+    for key in ('TE', 'STRESS', 'PRESSURE', 'VERSION'):
+        report.check(bool(props.get(key)), '%s extracted' % key)
+    if not (props.get('TE') and props.get('STRESS')):
+        return
+
+    te = float(sect(props['TE'][-1], 'values'))
+    #  Two silicon atoms with this pseudopotential land near -11.3
+    #  Hartree.  QE prints Ry, so a value near -22.6 would mean the
+    #  conversion was dropped -- the range is chosen to tell those apart
+    #  rather than to pin the number.
+    report.check(-15.0 < te < -8.0,
+                 'total energy is a plausible Hartree value for Si2 '
+                 '(got %g; ~-22.6 would mean Ry went unconverted)' % te)
+
+    stress = sect(props['STRESS'][-1], 'values').split()
+    report.check(len(stress) == 9,
+                 'STRESS is a full 3x3 tensor (got %d values)' % len(stress))
+    if len(stress) == 9:
+        #  A stress tensor is symmetric.  This is a real check on the
+        #  parse: a row/column transposition or a dropped element shows
+        #  up here and nowhere else.
+        s = [float(x) for x in stress]
+        sym = (abs(s[1] - s[3]) < 1e-6 and abs(s[2] - s[6]) < 1e-6
+               and abs(s[5] - s[7]) < 1e-6)
+        report.check(sym, 'STRESS tensor is symmetric (got %s)' % stress)
+
+
+def expect_gromacs_water(props, report):
+    """216 SPC waters, 50 steps of MD.
+
+    The point of this case is the WRAPPED ENERGY BLOCK.  With enough
+    terms GROMACS splits its "Energies (kJ/mol)" table over two rows:
+
+        LJ (SR)   Coulomb (SR)    Potential   Kinetic En.  Total Energy
+        2.13e+03  -1.21e+04      -9.97e+03    2.31e+03     -7.66e+03
+      Conserved En.  Temperature  Pressure (bar)
+       -7.32e+03     4.31e+02     2.36e+03
+
+    so TE comes from the first row and PRESSURE from the second.  A
+    parser that reads only the first row silently reports a plausible
+    number for the wrong quantity -- CLAUDE.md records this shape as
+    producing wrong values rather than none.  A short minimisation does
+    not wrap, so the .mdp here deliberately runs MD with temperature
+    coupling to make it.
+
+    It also guards the averages suppression: GROMACS prints one last
+    "Energies (kJ/mol)" block under "A V E R A G E S", and [NULL] exists
+    to stop that being taken as the final step's result.
+    """
+    for key in ('TE', 'PRESSURE'):
+        report.check(bool(props.get(key)), '%s extracted' % key)
+    if not (props.get('TE') and props.get('PRESSURE')):
+        return
+
+    te = float(sect(props['TE'][-1], 'values'))
+    pressure = float(sect(props['PRESSURE'][-1], 'values'))
+
+    #  TE is the Total Energy of 216 waters, a few thousand kJ/mol
+    #  negative.  Loose, because it depends on the random starting
+    #  velocities; what must not happen is picking up the Potential
+    #  (~-1e4) or the Conserved En. from the wrong column.
+    report.check(-9.5e3 < te < -5.0e3,
+                 'TE is the Total Energy, not the Potential '
+                 '(got %g; ~-1.0e4 would be the Potential)' % te)
+
+    #  Pressure lives in the SECOND row.  Reading it from the first
+    #  would pick up a kJ/mol energy of order 1e3-1e4 with the wrong
+    #  sign pattern; a real pressure here is a few thousand bar.
+    report.check(abs(pressure) < 2.0e4,
+                 'PRESSURE came from the wrapped second row (got %g)'
+                 % pressure)
+
+    #  The averages block is the LAST "Energies (kJ/mol)" in the file.
+    #  If [NULL] stopped suppressing it, TE would be the average rather
+    #  than the final step, and the two differ by enough to see.
+    report.check(len(props.get('TEVEC', [])) >= 2,
+                 'TEVEC accumulated per-step energies (got %d)'
+                 % len(props.get('TEVEC', [])))
+
+
 CASES = [
     dict(
         name='mopac-ch4-mos',
@@ -222,13 +312,61 @@ def expect_nwchem_h2o(props, report):
         output='ecce.out',
         expect=expect_nwchem_h2o,
     ),
+    dict(
+        #  Bulk Si, 2 atoms, low cutoff -- under a second.  QE needs
+        #  pseudopotentials to run at all; Debian's
+        #  quantum-espresso-data-sssp installs 103 of them in
+        #  /usr/share/espresso/pseudo, which is where ai.qe looks and
+        #  what this deck names.
+        name='qe-si-scf',
+        code='qe',
+        desc='qe.desc',
+        deck='qe/qe.pwin',
+        parse_args=('.', 'Energy', 'PW', 'PW', '0'),
+        output='qe.pwout',
+        expect=expect_qe_si,
+    ),
+    dict(
+        #  The .mdp is the deck ECCE names; conf.gro and topol.top are
+        #  copied alongside it and consumed by the grompp setup step.
+        name='gromacs-water-md',
+        code='gromacs',
+        desc='gromacs.desc',
+        deck='gromacs/gromacs.mdp',
+        parse_args=('.', 'Energy', 'MD', 'MD', '0'),
+        output='gromacs.log',
+        expect=expect_gromacs_water,
+    ),
 ]
 
 
 #  How to find and invoke each code.  A code that is absent is SKIPPED
 #  with its name reported, never silently passed -- an e2e suite that
 #  quietly tests nothing is worse than no e2e suite.
 CODES = {
+    'gromacs': dict(
+        binary='gmx',
+        search=(),
+        extraInputs=('conf.gro', 'topol.top'),
+        #  grompp first: GROMACS compiles the .mdp, coordinates and
+        #  topology into a .tpr, and only then runs.
+        setup=(lambda exe, deck: [exe, 'grompp', '-f', deck,
+                                  '-c', 'conf.gro', '-p', 'topol.top',
+                                  '-o', 'topol.tpr', '-maxwarn', '5'],),
+        argv=lambda exe, deck: [exe, 'mdrun', '-s', 'topol.tpr',
+                                '-g', 'gromacs.log', '-o', 'traj.trr',
+                                '-c', 'out.gro', '-e', 'ener.edr'],
+        packaged='gromacs',
+    ),
+    'qe': dict(
+        binary='pw.x',
+        search=(),
+        #  pw.x writes its output to stdout; ECCE redirects it into
+        #  qe.pwout, which is what the .edml declares as the parse file.
+        argv=lambda exe, deck: [exe, '-in', deck],
+        stdoutTo=True,
+        packaged='quantum-espresso + quantum-espresso-data-sssp',
+    ),
     'nwchem': dict(
         binary='nwchem',
         #  Debian's nwchem-openmpi ships /usr/bin/nwchem AND