GFN0/GFN1/GFN2 fails on macOS (but not Linux) for neutral lanthanide complex (Ce, charge=0, multiplicity=1)

[jevandezande]

Running GFN0-xTB, GFN1-xTB, or GFN2-xTB on a neutral cerium complex (charge=0, multiplicity=1, no --charge or --uhf flags) fails on the macOS binary with:

Error: Program stopped due to fatal error
-1- Assigned number of unpaired electrons (flag '--uhf <int>' or <int> in file '.UHF') is not consistent with the total number of electrons

The same input and command succeed on the Linux binary.

Environment:

• xtb version: 6.7.1 (6d44803) compiled by @darwin on 2026/02/25
• Fails on: macOS
• Passes on: Linux

Reproduction

Input file (Ce_18-crown-6.xyz):

43
Cerium 18-crown-6
Ce        -0.0001437629        0.0000290711        0.0000314625
C          1.4048527159        3.0831062543        0.0653267372
C          2.6113949649        2.1594927572       -0.0628305583
O          2.3364127712        0.9762045664        0.6796211324
C          3.3790299350        0.0145917128        0.6249728354
C          3.2625978458       -0.8828393385       -0.6227434392
O          2.0076635562       -1.5436768644       -0.6792740943
C          1.9681497053       -2.7587517478        0.0616929905
C          0.5643899530       -3.3400190761       -0.0677399790
O         -0.3215291148       -2.5117724159        0.6780336161
C         -1.6756881842       -2.9338484301        0.6253399417
C         -2.3970077854       -2.3838388290       -0.6208087612
O         -2.3418075195       -0.9666005156       -0.6769634836
C         -3.3731892235       -0.3249356490        0.0656406480
C         -3.1761313683        1.1814423159       -0.0653740392
O         -2.0144629626        1.5360306187        0.6774313376
C         -1.7030340553        2.9197181486        0.6212067738
C         -0.8651454423        3.2658163879       -0.6253532148
O          0.3348960861        2.5095955647       -0.6784715153
H          1.6491656373        4.0734787978       -0.3444656203
H          1.1434843064        3.1779011677        1.1298840469
H          3.5034900310        2.6532982443        0.3480426379
H          2.7718002915        1.9328905179       -1.1274838434
H          4.3592969145        0.5071368800        0.6396448791
H          3.2688144356       -0.5894180426        1.5328846941
H          4.0830713955       -1.6110827570       -0.6358339519
H          3.3131552181       -0.2710708397       -1.5307155666
H          2.1811391534       -2.5824596549        1.1266423186
H          2.7029716434       -3.4653050821       -0.3497729030
H          0.2874989849       -3.3616521521       -1.1323910389
H          0.5451459869       -4.3608790628        0.3395993030
H         -1.7392152624       -4.0290779383        0.6395520384
H         -2.1420891436       -2.5368647876        1.5342543479
H         -3.4379855058       -2.7301794677       -0.6320427240
H         -1.8942587619       -2.7332851743       -1.5298434253
H         -4.3531507066       -0.6089866202       -0.3435128848
H         -3.3246239002       -0.5965978032        1.1307214780
H         -3.0590751031        1.4313369996       -1.1304691014
H         -4.0500838974        1.7081588386        0.3435929439
H         -2.6198108249        3.5223254795        0.6330113863
H         -1.1267938141        3.1278303619        1.5299879097
H         -0.6448077892        4.3405147059       -0.6396316232
H         -1.4183809242        3.0022034937       -1.5340285941

Commands (all fail on macOS, all pass on Linux):

xtb Ce_18-crown-6.xyz --gfn 0
xtb Ce_18-crown-6.xyz --gfn 1
xtb Ce_18-crown-6.xyz --gfn 2

---

Expected behavior

Ce⁰ has 58 electrons (even count). A singlet state (0 unpaired electrons, --uhf 0) is mathematically valid and should run without error on all platforms, as it does on
Linux.

Actual behavior (macOS only)

Note: The following floating-point exceptions are signalling: IEEE_UNDERFLOW_FLAG
ERROR STOP

Error termination. Backtrace:
#0 0x103cf2103
...
########################################################################
Error: Program stopped due to fatal error
-1- Assigned number of unpaired electrons (flag '--uhf ' or in file '.UHF') is not consistent with the total number of electrons
########################################################################

---

Related issues

A few issues are potentially related:

• Automatic UHF calculation for odd number of electrons #1107 / PR Prevent inconsistent multiplicities #1235: "Automatic UHF for odd electron count" / "Prevent inconsistent multiplicities." Those address the general multiplicity check machinery. PR Prevent inconsistent multiplicities #1235
  (merged March 2025) added the ERROR STOP that produces the error we see, but the underlying f-in-core electron count mismatch on macOS appears to be a separate, prior-stage
  problem.
• GCC 13.4 + MacOS fails on CI #1326 — "GCC 13.4 + macOS fails on CI": GCC-version-specific macOS CI failures on GFN0/1/2. Different symptom ("Condition not fulfilled") but the same pattern of
  macOS-only GFN method failures.

[TyBalduf]

A potential hint at where this comes from:

xtb/src/prog/main.F90

Line 453 in eab23de

if (mol%at(i) > 57 .and. mol%at(i) < 72) mol%z(i) = 3

All lanthanides are treated as if they were La for determining their non-core Z, meaning every even Lanthanide should mistakenly fail this check added in https://github.com/grimme-lab/xtb/pull/1235/changes. I wouldn't think this would be platform dependent, are your Linux and Mac binaries on the same commit?