01
How each result is made
Every case is a real run. The untrusted C driver computes a candidate solution and writes a certificate; the proved Lean checker re-checks the certificate in exact rational arithmetic and prints rigorous intervals and error bounds; then an established code in the field solves the same problem independently, and we check that its answer falls inside the Lean intervals.
Every number in a “Certified by Lean” table is printed by the checker, whose soundness theorem guarantees it for the problem the certificate describes (lower bounds rounded down, upper bounds up). The comparison with the reference code is supporting evidence only and is outside the chain of trust. Every archived certificate is re-checked by the current checkers in make verify-all, with one labelled exception: the archived production record of the S1 silicon band gap (one check takes about 8 hours on 32 cores and the certificate is 253 MB, not in the repository), for which only the sha256 of the archived output is verified and the ledger ids are given.
02
Plane-wave DFT in 3D: real silicon, hydrogen, and comparison with experiment
Real silicon: the LDA SCF solution on a 4×4×4 k-point mesh, with bands at every k point DFT
Three-dimensional plane-wave Kohn–Sham LDA: diamond-structure silicon (a = 10.26 bohr) with a real GTH-LDA pseudopotential (CP2K’s Si q4, including the nonlocal part), Slater exchange plus PW92 correlation, zero temperature, a standard-shifted Monkhorst–Pack 4×4×4 mesh (64 k points) and the cutoff |k + G|² ≤ 17/2 in units of (2π/a)² (23–29 plane waves per k point), a 12³ density grid. The untrusted Python driver solves and writes a certificate; the Lean checker reduces the 64 k points to 10 representative blocks with the point group, proves that the discrete Kohn–Sham equations on the full mesh have exactly one zero-temperature SCF solution near the certificate centre, and bounds its chemical potential, the 16 bands at every k point and the band gap on the mesh (DFT-M75–M77, K21–K28, T4; the DFTK alignment DFT-M78 is T0). The cutoff is low, so the numbers are not converged silicon; the point is the whole chain — a real pseudopotential, k points, symmetry and exchange–correlation all inside a proved checker.
Bands at the 10 representative k points
The horizontal axis lists the 10 representative k points left after the point-group reduction (reduced coordinates; ×n is the number of k points each one stands for), the vertical axis ε − μ* (eV). Blue dashes: midpoints of the Lean intervals (about 4×10⁻⁹ Ha wide, invisible here). Orange crosses: DFTK’s eigenvalues for the same discrete problem on the full mesh. Grey band: the Lean-certified gap on the mesh (max ε₄ to min ε₅ over the 64 points).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| zero-temperature LDA SCF solution | exactly one within r = 6.0e-12 of the certificate centre | the discrete Kohn–Sham equations on all 64 k points (weights 1/64); the checker works on the blocks of 10 representative k points and covers the rest with 12 symmetry operations (DFT-M76, T4); the coordinates are Fourier coefficients of the effective potential |
| chemical potential μ* (Ha) | [0.257600113702550, 0.257600116667180] (width 3.0e-09) | of that SCF solution (the G = 0 convention differs from DFTK’s by a constant) |
| occupations | HOMO–LUMO gap ≥ 0.056719202548265 Ha | the lowest 4 bands are doubly occupied at every k point (8 electrons): the zero-temperature Aufbau solution |
| bands ε_i(k) − μ* | 1024 intervals (64 k points × 16 bands), widest 6.5e-11 Ha | a non-representative k point has its representative’s spectrum (DFT-M77, T4) |
| band gap on the mesh | [0.057619508613245, 0.057619508732148] Ha = [1.567906704, 1.567906708] eV | min ε₅ − max ε₄ over the 64 mesh points; an upper bound on the Brillouin-zone gap of this SCF Hamiltonian |
| plane waves | 23–29 per representative k point | |k + G|² ≤ Q in units of (2π/a)², Q = 17/2, a = 10.26 bohr |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| plane waves per k point | 64 k points | the checker’s basis | ✓ equal at every k point |
| distance of DFTK’s SCF solution from the certificate centre (effective-potential coordinates) | 2.55e-12 | r = 6.01e-12 | ✓ inside the ball where Lean proves uniqueness |
| eigenvalues at every k point (plus the G = 0 convention constant) | 1024 values | absolute intervals [lo + μ_lo, hi + μ_hi] | ✓ 1024/1024 inside (864 at non-representative k points); at most 1.6e-12 Ha from the interval centres |
| band gap on the mesh (Ha) | 0.057619508673823 | [0.057619508613245, 0.057619508732148] (width 1.2e-10) | ✓ inside |
Command and Lean checker output
PROBLEM fcc a_bohr=10.26000000000000000000 Q=8.50000000000000000000 T=15225445/536870912 occ=zero(smooth-step,T=scale) Npair=4 psp_Z=4.000000000000000 …
PW3D_LDAPG 64 10 12 89 19.00000000000000000000 reps=0:2,1:6,2:6,3:6,5:6,6:12,7:12,10:6,21:2,22:6
PW3D_LDAPG_PAIR 0 12 0
PW3D_LDAPG_PAIR 1 12 1
PW3D_KBASIS 10 174 175 24 24 26 23 26 29 27 25 28 26
PW3D_LDAPG_OK 0.000000000006006 0.257600113702550 0.257600116667180 0.000000000000099 0.234505791324923 0.000000000002299 1.323395645621417 0.00225545 …
PW3D_LDAPG_ZERO lambda0 2687475158540463/4503599627370496 gap_ge 0.056719202548265
PW3D_LDAPG_KAPPA 0.234506 0.090794 0.080305 0.042244 0.094816 11.983578 3.186814 1.323396 0.000001 0.000001 a 110445.834474
PW3D_LDAPG_BAND 0 1 -0.435100173333547 -0.435100173274702
PW3D_LDAPG_BAND 0 2 -0.075220673835666 -0.075220673776821
PW3D_LDAPG_BANDGAP 0.057619508613245 0.057619508732148
… and PW3D_LDAPG_PAIR × 64, PW3D_LDAPG_BAND × 1024
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
Silicon’s bands bounded over the whole Brillouin zone, and a gap upper bound (with the S1 production record) DFT
The same SCF solution (the previous case’s certificate), followed by a second certificate from the bands driver: an octree divides the Brillouin zone into cubic leaves with a Rayleigh–Ritz trial subspace on each, and the checker proves that band 4 lies below one bound for every κ ∈ ℝ³; it also proves upper bounds on band 5 at 7 points on the Γ–X line and two-sided intervals for bands 4 and 5 at two mesh points, which give a rigorous upper bound on the band gap (DFT-W42–W48, T3–T4). The bands certificate only claims the radius r of the SCF solution; its conclusions hold together with the previous case’s check of the same SCF certificate (as in the S1 production run). The zone-wide bound is built with a deliberate margin η = 0.05 Ha so that the check takes about 2.5 minutes; the production run of the S1 protocol (8×8×8 mesh, Q = 107/2, a = 5.416 Å) is in the archived record below.
Bands along L–Γ–X–W–K–Γ with the proved bounds
Grey lines: the first 8 bands computed by DFTK along the path on the same SCF potential (ε − μ*, eV). Red dashed line: the zone-wide bound proved in Lean — ε₄(κ) lies below it for every κ. Blue triangles: upper bounds on ε₅ at the Ritz points on Γ–X; blue dots: the two-sided intervals at the listed mesh point that lies on the path (equivalent to (7/8, 7/8, 7/8)). The cutoff is low and the plane-wave set |k + G|² ≤ Q jumps along the path, hence the kinks (a property of the model; the checker and DFTK use the same basis).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| SCF certificate | zero-temperature LDA SCF solution unique within 6.0e-12 (≤ the claimed 6.0e-12) | the certificate and check of the previous case (runUGR, DFT-M76); the bands certificate only claims r, and its conclusions hold together with that check (DFT-W47) |
| valence-band ceiling over the whole Brillouin zone | for every κ ∈ ℝ³: ε₄(κ) − μ* < 0.031140052410973 Ha | an octree of 4754 cubic leaves covers [0, 1)³ with a Rayleigh–Ritz count on every leaf, extended to ℝ³ by reciprocal-lattice periodicity (DFT-W44, W47, T4) |
| conduction-band upper bounds at Ritz points | ε₅ − μ* bounded above at 7 points, lowest 0.020201753493342 Ha (κ = (0, 17/40, 17/40)) | (0, t, t) on the Γ–X line, t = 17/40, 2/5, 41/100, 21/50, 43/100, 11/25, 9/20 (reduced coordinates) |
| ε4(7/8, 7/8, 7/8) − μ* (Ha) | [-0.028812754336349, -0.028806754336348] (width 6.0e-06) | two-sided interval at a listed mesh point (DFT-W45, T4) |
| ε5(7/8, 7/8, 7/8) − μ* (Ha) | [0.100347698983233, 0.100353698983234] (width 6.0e-06) | two-sided interval at a listed mesh point (DFT-W45, T4) |
| ε4(7/8, 3/8, 3/8) − μ* (Ha) | [-0.117139478243362, -0.117133478243361] (width 6.0e-06) | two-sided interval at a listed mesh point (DFT-W45, T4) |
| ε5(7/8, 3/8, 3/8) − μ* (Ha) | [0.028806754336347, 0.028812754336348] (width 6.0e-06) | two-sided interval at a listed mesh point (DFT-W45, T4) |
| upper bound on the band gap | ≤ 0.049014507829690 Ha = 1.333753 eV | the smallest upper bound on ε₅ at the Ritz and listed points minus the largest lower bound on ε₄ at the listed points; the Brillouin-zone gap is at most this. The conduction-band minimum is off the mesh (near 0.86 of the way along Γ–X), so this bound is below the gap on the mesh |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| bands 4 and 5 at the listed mesh points | 4 values | absolute PW3D_BZR_POINT intervals | ✓ 4/4 inside, at most 9.6e-13 Ha from the centres; plane-wave counts equal the checker’s at all 14 points |
| ε₅ at the Ritz points | 7 points | PW3D_BZR_RITZ upper bounds | ✓ 7/7 below the bound |
| largest ε₄ at the evaluated points (Ha, absolute) | 0.238740167595 | PW3D_BZR_UP + μ_hi = 0.288740169078 | ✓ 0.0500 Ha below the zone-wide bound (built with a margin η = 0.05 Ha) |
| gap estimate at the same points (Ha): min ε₅ − max ε₄ | 0.049011391831 | ≤ 0.04901450782969 | ✓ 3.1e-06 Ha below the bound; 7 Ritz points, 2 listed points |
Archived production record: the S1 silicon band gap (not re-checked here)
The run above is a reduced one; the production runs of the same two checkers (the LDA point-group mode and the bands mode) on the problem of the S1 protocol are recorded in the ledger (DFT-K29, W51). The production certificate is 253 MB and one check takes about 8 hours on 32 cores, so it is not in the repository and not re-run by make verify-all. The full output of the SCF checker is archived here as record/prod888s1.out and --check only verifies its sha256; for the certificate only its sha256 is given. The output of the bands certificate is not archived here; its numbers below are copied from ledger row DFT-W51.
| Quantity | Result | Source |
|---|---|---|
| problem | real Si (GTH-LDA q4), a = 5.416 Å exactly, Q = 107/2 (399–419 plane waves per k point), unshifted MP 8×8×8 (512 k points, 29 representatives, 48 operations), 24³ grid, Slater + PW92, zero temperature | the problem of the S1 protocol (domains/dft/notes/s1-protocol.md); ledger DFT-K29, T4 |
| zero-temperature LDA SCF solution | exactly one within r = 2.09e-10 | prod888s1.out (DFT-K29) |
| chemical potential μ* (Ha) | [0.235321048732112, 0.235338320741366] (width 1.7e-05) | prod888s1.out (DFT-K29) |
| bands ε_i(k) − μ* | 8192 intervals (512 k points × 16 bands), widest 8.6e-09 Ha | prod888s1.out (DFT-K29) |
| band gap on the mesh | [0.019504197946170, 0.019504214972386] Ha = [0.53073626, 0.53073673] eV | prod888s1.out; an upper bound on the Brillouin-zone gap, not the S1 comparison number |
| valence-band ceiling over the whole Brillouin zone | for every κ ∈ ℝ³: ε₄(κ) − μ* < 0.020247897 Ha | PW3D_BZR_OK of the bands certificate prod3.bzr (η = 0.03, 7,505 leaves); its output is not archived here, the numbers are those of ledger row DFT-W51 (T4) |
| ε₄ at Γ and ε₅ at (0, 5/8, 5/8) (Ha, relative to μ*) | [−0.009755125, −0.009749081] and [0.009749044, 0.009755162] | ledger DFT-W51 |
| S1 Rule U gap upper bound C | C = 0.019510287 Ha = 0.530902 eV | ledger DFT-W51: the S1 summary (fis.s1val, using only the mesh points with two-sided intervals). The ledger records no comparison with experiment yet |
| cost of one check | SCF: 7 h 57 min on 32 cores, 196 CPU hours, 92.8 GB; bands: 44.5 min on 32 cores, 16.0 CPU hours | certify-pw3d f43f379 (SCF, 2026-10-15); not re-run by make verify-all |
sha256: prod888s1.out f4bd0a151e3fa7ff…; prod888s1.cert 8db295db48c5e3fc…
Command and Lean checker output
PROBLEM fcc a_bohr=10.26000000000000000000 Q=8.50000000000000000000 T=15225445/536870912 occ=zero(smooth-step,T=scale) Npair=4 psp_Z=4.000000000000000 …
PROBLEM_BZR basis(k)=|M(k+G)|^2<=Q bands_relative_to_mu prec=120 cube=4 patches=71 up=true ritz=7 points=2
PW3D_BZR_SCF claim 0.000000000006007000000000000000 (valid only together with a PW3D_LDAPG_OK run of the same SCF certificate whose r is <= this value …
PW3D_BZR_UP 0.031140052410973 4754
PW3D_BZR_RITZ 0.00000000000000000000 0.42500000000000000000 0.42500000000000000000 5 0.020201753493342
PW3D_BZR_RITZ 0.00000000000000000000 0.40000000000000000000 0.40000000000000000000 5 0.020744181519140
PW3D_BZR_POINT 0.87500000000000000000 0.87500000000000000000 0.87500000000000000000 4 -0.028812754336349 -0.028806754336348
PW3D_BZR_POINT 0.87500000000000000000 0.87500000000000000000 0.87500000000000000000 5 0.100347698983233 0.100353698983234
PW3D_BZR_POINT 0.87500000000000000000 0.37500000000000000000 0.37500000000000000000 4 -0.117139478243362 -0.117133478243361
PW3D_BZR_POINT 0.87500000000000000000 0.37500000000000000000 0.37500000000000000000 5 0.028806754336347 0.028812754336348
PW3D_BZR_GAP_UPPER 0.049014507829690
PW3D_BZR_OK
… and PW3D_BZR_RITZ × 7
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
S2: an SCF solution proved not to be the ground state DFT
Hydrogen on a simple cubic lattice (lattice constant 7 bohr, described by a two-atom fcc cell with a = 14 bohr), the local part of the GTH-LDA hydrogen pseudopotential, Slater exchange, Fermi–Dirac at T = 1/100 Ha, Γ point, 59 plane waves, a 12³ grid. The spin-restricted (non-magnetic) LDA equations have an SCF solution; the checker proves that it exists and is locally unique, gives a rigorous lower bound on its free energy, and checks an upper bound on the energy of a spin-polarised trial state in the certificate (the two electrons on the two atoms). The difference is positive, so this SCF solution is not the minimiser of the spin-polarised (LSDA) free energy: a converged SCF solution is proved rigorously not to be the ground state (DFT-Z06–Z08, T4).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| spin-restricted LDA SCF solution | exactly one within r = 3.7e-10 of the certificate centre | Slater exchange, Fermi–Dirac T = 1/100 Ha, Γ point, 59 plane waves, 12³ grid (DFT-M26, T4) |
| chemical potential μ (Ha) | [-0.154613711592003, -0.154611685731185] (width 2.0e-06) | of that solution |
| free-energy lower bound F_lo of that solution (Ha) | F_S(γ*, γ*) ≥ -0.362344632501586 | Gibbs identity with Weyl-type eigenvalue bounds (DFT-Z07, T4); without the Ewald energy |
| energy upper bound E_hi of a spin-polarised trial state (Ha) | F_S(p) ≤ -0.374473491402560 | the certificate’s trial state: the spin-up electron on one atom, the spin-down one on the other (DFT-Z06, T4) |
| conclusion (S2) | F_lo − E_hi = 0.012128858900974 Ha = 330.0 meV > 0 | the spin-restricted SCF solution is not a minimiser of the LSDA free energy: inf F_S ≤ F_S(γ*, γ*) − 0.0121 Ha (DFT-Z08, T4) |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| εF of the spin-restricted solution (Ha) | -0.154612698674 | [-0.154613711592003, -0.154611685731185] (width 2.0e-06) | ✓ inside |
| free energy F of the spin-restricted solution (Ha, without Ewald) | -0.362337522239 | ≥ F_lo = -0.362344632502 | ✓ 7.1e-06 Ha above F_lo (the lower bound is nearly tight) |
| free energy F of the antiferromagnetic LSDA SCF solution (Ha) | -0.374487820597 | trial state’s E_hi = -0.374473491403 | ✓ 1.4e-05 Ha below E_hi; ∫|ρ↑ − ρ↓| = 1.47 |
| difference of DFTK’s two SCF solutions | 330.6 meV | proved margin ≥ 330.0 meV | — floating point, recorded only |
Command and Lean checker output
PROBLEM fcc a_bohr=14.00000000000000000000 Q=12.50000000000000000000 T=0.01000000000000000000 Npair=1 smear=fd psp_Z=1.00000000000000000000 psp_alpha= … PW3D_BASIS 59 338 339 PW3D_LDA_OK 0.000000000368373 -0.027934899666825 -0.027934898930078 -0.154613711592003 -0.154611685731185 0.000000000005135 0.000006770556463 0.000000 … PW3D_FREE_LO -0.362344632501586 PW3D_TRIAL_HI -0.374473491402560 PW3D_S2_OK 0.012128858900974
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
Cohen–Bergstresser bands against measured levels: a pre-registered comparison (12 semiconductors) DFT
The 1966 empirical pseudopotential model of Cohen and Bergstresser gives the band structures of diamond and zincblende semiconductors. For Si, Ge and GaAs (round 2) and GaP, GaSb, InP, InAs, InSb, ZnS, ZnSe, ZnTe and CdTe (the round-12 extension) the driver writes certificates for the first 9 bands at Γ, X and L (plus 5 points on the Δ line for Si, Ge, GaAs and GaP), and certify-pw returns rigorous intervals 2×10⁻⁹ Ha wide (DFT-W14, T4). Which levels are compared, against which data set and by which rule was written down as a protocol and committed before the comparison. The certificates are the 12 used by the ledger’s validation claims (domains/dft/validation/), re-checked here by the current checker.
Deviation intervals
Each row is the deviation interval [y − C − u, y − c + u] (eV) of one comparison entry: blue if it contains 0 (agrees), red if not (disagrees); grey dots are entries with unknown uncertainty, not judged (only y minus the centre of the Lean interval is drawn).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| Si(diamond, a = 5.43 Å) | 8 k points (G, X, L, D0.70, D0.75, D0.80, D0.85, D0.90) × 9 bands: 72 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W14, conclusions DFT-W14 (T4) |
| Ge(diamond, a = 5.66 Å) | 8 k points (G, X, L, D0.70, D0.75, D0.80, D0.85, D0.90) × 9 bands: 72 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W14, conclusions DFT-W14 (T4) |
| GaAs(zincblende, a = 5.64 Å) | 8 k points (G, X, L, D0.70, D0.75, D0.80, D0.85, D0.90) × 9 bands: 72 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W14, conclusions DFT-W14 (T4) |
| GaP(zincblende, a = 5.44 Å) | 8 k points (G, X, L, D0.70, D0.75, D0.80, D0.85, D0.90) × 9 bands: 72 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W70, conclusions DFT-W14 (T4) |
| GaSb(zincblende, a = 6.12 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W71, conclusions DFT-W14 (T4) |
| InP(zincblende, a = 5.86 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W72, conclusions DFT-W14 (T4) |
| InAs(zincblende, a = 6.04 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W73, conclusions DFT-W14 (T4) |
| InSb(zincblende, a = 6.48 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W74, conclusions DFT-W14 (T4) |
| ZnS(zincblende, a = 5.41 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W75, conclusions DFT-W14 (T4) |
| ZnSe(zincblende, a = 5.65 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W76, conclusions DFT-W14 (T4) |
| ZnTe(zincblende, a = 6.07 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W77, conclusions DFT-W14 (T4) |
| CdTe(zincblende, a = 6.41 Å) | 3 k points (G, X, L) × 9 bands: 27 intervals, widest 2.0e-09 Ha | PW_OK; cutoff |k + G|² ≤ 67/2; run DFT-W78, conclusions DFT-W14 (T4) |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| Si: first 9 bands at 8 k points | 72 values | PW_BAND intervals | ✓ 72/72 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| Ge: first 9 bands at 8 k points | 72 values | PW_BAND intervals | ✓ 72/72 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| GaAs: first 9 bands at 8 k points | 72 values | PW_BAND intervals | ✓ 72/72 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| GaP: first 9 bands at 8 k points | 72 values | PW_BAND intervals | ✓ 72/72 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| GaSb: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| InP: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| InAs: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| InSb: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| ZnS: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| ZnSe: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| ZnTe: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
| CdTe: first 9 bands at 3 k points | 27 values | PW_BAND intervals | ✓ 27/27 inside, nearest endpoint 1.0e-09 Ha; plane-wave counts equal |
Against experiment (pre-registered)
Physics validation (ledger rule 7): the compared quantities, the data sets, the tolerance rule and the choice of Lean intervals were written down as a protocol and committed before any comparison — round 2 (Si, Ge, GaAs) on 2026-10-08 in commit 82d6278, the extension (nine more zincblende materials) on 2026-10-17 in commit 5fafa3d. The deviation interval is [y − C − u, y − c + u], where [c, C] is the Lean interval, y the measured value minus corrections and u the uncertainty (added linearly); it agrees if it contains 0. The CB 1966 empirical pseudopotentials were fitted to some of the levels; fitted levels and their linear combinations are excluded. The V grade rates the strength of the comparison, not its outcome. 55 entries: 14 agree, 39 disagree, 2 not judged.
| Claim | Quantity | Registered | Grade | Verdict |
|---|---|---|---|---|
| DFT-O01 | valence-band width E(Γ25′) − E(Γ1) (Si, Ge) | 2026-10-08, commit 82d6278; compared 2026-10-08 | V2 | partly consistent (1/2 entries agree) |
| DFT-O02 | top of the valence band at L, E(L3′) − E(Γ25′) (Si, GaAs) | 2026-10-08, commit 82d6278; compared 2026-10-08 | V2 | partly consistent (1/2 entries agree) |
| DFT-O03 | Si conduction level E(Γ2′) − E(Γ25′) | 2026-10-08, commit 82d6278; compared 2026-10-08 | V0 | not judged |
| DFT-O07 | valence-band width E(Γ15v) − E(Γ1) (GaP, GaSb, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (2/5 entries agree) |
| DFT-O08 | lowest valence level at X, E(X1v) − E(Γ15v) (GaP, GaSb, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (1/5 entries agree) |
| DFT-O09 | second valence level at X, E(X3v) − E(Γ15v) (GaP, GaSb, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | inconsistent (0/5 entries agree) |
| DFT-O10 | top valence level at X, E(X5v) − E(Γ15v) (GaSb, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (1/4 entries agree) |
| DFT-O11 | lowest valence level at L, E(L1v) − E(Γ15v) (GaP, GaSb, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (2/5 entries agree) |
| DFT-O12 | top valence level at L, E(L3v) − E(Γ15v) (GaP, InP, InAs, InSb) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (3/4 entries agree) |
| DFT-O13 | valence-band width E(Γ15v) − E(Γ1) (ZnS, ZnSe, ZnTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | inconsistent (0/3 entries agree) |
| DFT-O14 | lowest valence level at X, E(X1v) − E(Γ15v) (ZnS, ZnSe, ZnTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (1/3 entries agree) |
| DFT-O15 | second valence level at X, E(X3v) − E(Γ15v) (ZnS, ZnSe, ZnTe, CdTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | inconsistent (0/4 entries agree) |
| DFT-O16 | top valence level at X, E(X5v) − E(Γ15v) (ZnS, ZnSe, ZnTe, CdTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | inconsistent (0/4 entries agree) |
| DFT-O17 | lowest valence level at L, E(L1v) − E(Γ15v) (ZnS, ZnSe, ZnTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (1/3 entries agree) |
| DFT-O18 | top valence level at L, E(L3v) − E(Γ15v) (ZnS, ZnSe, ZnTe, CdTe) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V2 | partly consistent (1/4 entries agree) |
| DFT-O19 | CdTe lowest valence level at X, E(X1v) − E(Γ15v) (UPS, uncertainty unknown) | 2026-10-17, commit 5fafa3d; compared 2026-10-17 | V0 | not judged |
Entry by entry (55 entries)
| Entry | Lean interval (eV) | measured y ± u (eV) | deviation interval (eV) | Verdict |
|---|---|---|---|---|
| DFT-O01 Si VBW | [12.61339425, 12.61339437] | 12.487 ± 0.302 | [-0.429, +0.175] | ✓ agrees |
| DFT-O01 Ge VBW | [11.96699068, 11.96699080] | 12.503 ± 0.302 | [+0.234, +0.838] | ✗ disagrees |
| DFT-O02 Si L3pv | [-1.25301264, -1.25301253] | -1.187 ± 0.302 | [-0.236, +0.368] | ✓ agrees |
| DFT-O02 GaAs L3v | [-0.91368446, -0.91368434] | -1.298 ± 0.305 | [-0.690, -0.080] | ✗ disagrees |
| DFT-O03 Si G2pc | [3.88924025, 3.88924037] | 4.193(u 未知) | — | — not judged (uncertainty unknown) |
| DFT-O07 GaP VBW | [13.06006032, 13.06006044] | 13.157 ± 0.402 | [-0.305, +0.499] | ✓ agrees |
| DFT-O07 GaSb VBW | [10.60827208, 10.60827219] | 11.333 ± 0.302 | [+0.423, +1.027] | ✗ disagrees |
| DFT-O07 InP VBW | [11.26365407, 11.26365419] | 10.964 ± 0.408 | [-0.708, +0.108] | ✓ agrees |
| DFT-O07 InAs VBW | [11.24777639, 11.24777651] | 12.163 ± 0.402 | [+0.514, +1.318] | ✗ disagrees |
| DFT-O07 InSb VBW | [9.66077702, 9.66077714] | 11.400 ± 0.302 | [+1.437, +2.041] | ✗ disagrees |
| DFT-O08 GaP X1v | [-11.10566423, -11.10566411] | -9.557 ± 0.302 | [+1.247, +1.851] | ✗ disagrees |
| DFT-O08 GaSb X1v | [-9.15989801, -9.15989790] | -9.133 ± 0.202 | [-0.175, +0.229] | ✓ agrees |
| DFT-O08 InP X1v | [-9.50626343, -9.50626331] | -8.864 ± 0.308 | [+0.334, +0.950] | ✗ disagrees |
| DFT-O08 InAs X1v | [-9.98444648, -9.98444636] | -9.663 ± 0.302 | [+0.019, +0.623] | ✗ disagrees |
| DFT-O08 InSb X1v | [-8.53330485, -8.53330473] | -9.200 ± 0.202 | [-0.869, -0.465] | ✗ disagrees |
| DFT-O09 GaP X3v | [-5.71544368, -5.71544356] | -6.857 ± 0.202 | [-1.343, -0.939] | ✗ disagrees |
| DFT-O09 GaSb X3v | [-5.00316205, -5.00316193] | -6.633 ± 0.302 | [-1.932, -1.328] | ✗ disagrees |
| DFT-O09 InP X3v | [-5.41384702, -5.41384690] | -5.864 ± 0.208 | [-0.658, -0.242] | ✗ disagrees |
| DFT-O09 InAs X3v | [-4.89968360, -4.89968349] | -6.163 ± 0.202 | [-1.466, -1.062] | ✗ disagrees |
| DFT-O09 InSb X3v | [-4.26721196, -4.26721184] | -6.100 ± 0.202 | [-2.035, -1.631] | ✗ disagrees |
| DFT-O10 GaSb X5v | [-1.72203002, -1.72202990] | -2.433 ± 0.202 | [-0.913, -0.509] | ✗ disagrees |
| DFT-O10 InP X5v | [-1.94371852, -1.94371840] | -1.964 ± 0.208 | [-0.228, +0.188] | ✓ agrees |
| DFT-O10 InAs X5v | [-1.74602271, -1.74602259] | -2.263 ± 0.302 | [-0.819, -0.215] | ✗ disagrees |
| DFT-O10 InSb X5v | [-1.48405439, -1.48405427] | -2.100 ± 0.402 | [-1.018, -0.214] | ✗ disagrees |
| DFT-O11 GaP L1v | [-11.64465452, -11.64465440] | -10.557 ± 0.302 | [+0.786, +1.390] | ✗ disagrees |
| DFT-O11 GaSb L1v | [-9.57253212, -9.57253200] | -10.033 ± 0.302 | [-0.763, -0.159] | ✗ disagrees |
| DFT-O11 InP L1v | [-10.01829315, -10.01829303] | -9.964 ± 0.308 | [-0.254, +0.362] | ✓ agrees |
| DFT-O11 InAs L1v | [-10.32857810, -10.32857798] | -10.463 ± 0.302 | [-0.437, +0.167] | ✓ agrees |
| DFT-O11 InSb L1v | [-8.84366481, -8.84366469] | -10.200 ± 0.302 | [-1.658, -1.054] | ✗ disagrees |
| DFT-O12 GaP L3v | [-0.90840834, -0.90840822] | -1.157 ± 0.302 | [-0.550, +0.054] | ✓ agrees |
| DFT-O12 InP L3v | [-0.78773439, -0.78773427] | -0.964 ± 0.308 | [-0.484, +0.132] | ✓ agrees |
| DFT-O12 InAs L3v | [-0.69482527, -0.69482515] | -0.763 ± 0.302 | [-0.371, +0.233] | ✓ agrees |
| DFT-O12 InSb L3v | [-0.59255912, -0.59255901] | -1.100 ± 0.302 | [-0.809, -0.205] | ✗ disagrees |
| DFT-O13 ZnS VBW | [14.76762839, 14.76762851] | 13.467 ± 0.402 | [-1.703, -0.899] | ✗ disagrees |
| DFT-O13 ZnSe VBW | [13.72960069, 13.72960081] | 15.050 ± 0.602 | [+0.718, +1.922] | ✗ disagrees |
| DFT-O13 ZnTe VBW | [11.70692371, 11.70692383] | 12.697 ± 0.402 | [+0.588, +1.392] | ✗ disagrees |
| DFT-O14 ZnS X1v | [-13.99726712, -13.99726700] | -11.967 ± 0.302 | [+1.729, +2.333] | ✗ disagrees |
| DFT-O14 ZnSe X1v | [-13.03081658, -13.03081646] | -12.350 ± 0.402 | [+0.279, +1.083] | ✗ disagrees |
| DFT-O14 ZnTe X1v | [-11.05152454, -11.05152443] | -11.297 ± 0.302 | [-0.547, +0.057] | ✓ agrees |
| DFT-O15 ZnS X3v | [-3.58941485, -3.58941473] | -5.467 ± 0.202 | [-2.079, -1.675] | ✗ disagrees |
| DFT-O15 ZnSe X3v | [-3.46041352, -3.46041340] | -5.450 ± 0.302 | [-2.292, -1.688] | ✗ disagrees |
| DFT-O15 ZnTe X3v | [-3.16328825, -3.16328813] | -5.197 ± 0.202 | [-2.235, -1.831] | ✗ disagrees |
| DFT-O15 CdTe X3v | [-2.47320230, -2.47320218] | -4.830 ± 0.202 | [-2.559, -2.155] | ✗ disagrees |
| DFT-O16 ZnS X5v | [-1.49451591, -1.49451579] | -2.467 ± 0.302 | [-1.274, -0.670] | ✗ disagrees |
| DFT-O16 ZnSe X5v | [-1.38161218, -1.38161206] | -1.950 ± 0.302 | [-0.870, -0.266] | ✗ disagrees |
| DFT-O16 ZnTe X5v | [-1.20555340, -1.20555328] | -2.097 ± 0.202 | [-1.093, -0.689] | ✗ disagrees |
| DFT-O16 CdTe X5v | [-0.94246932, -0.94246920] | -1.530 ± 0.202 | [-0.790, -0.386] | ✗ disagrees |
| DFT-O17 ZnS L1v | [-14.18880278, -14.18880266] | -12.367 ± 0.302 | [+1.520, +2.124] | ✗ disagrees |
| DFT-O17 ZnSe L1v | [-13.20450642, -13.20450630] | -12.950 ± 0.302 | [-0.047, +0.557] | ✓ agrees |
| DFT-O17 ZnTe L1v | [-11.21574725, -11.21574714] | -11.697 ± 0.302 | [-0.783, -0.179] | ✗ disagrees |
| DFT-O18 ZnS L3v | [-0.53207596, -0.53207585] | -1.367 ± 0.402 | [-1.237, -0.433] | ✗ disagrees |
| DFT-O18 ZnSe L3v | [-0.50072031, -0.50072019] | -1.150 ± 0.302 | [-0.951, -0.347] | ✗ disagrees |
| DFT-O18 ZnTe L3v | [-0.44933716, -0.44933704] | -0.797 ± 0.302 | [-0.649, -0.045] | ✗ disagrees |
| DFT-O18 CdTe L3v | [-0.34359955, -0.34359943] | -0.630 ± 0.302 | [-0.588, +0.016] | ✓ agrees |
| DFT-O19 CdTe X1v-UPS | [-11.37980292, -11.37980280] | -8.530(u 未知) | — | — not judged (uncertainty unknown) |
Measured values, uncertainties and corrections come from the pre-registered data sets cb-validation and cb-validation-ext (data/experimental/, sha256 recorded in domains/dft/validation.json). This table is generated from validation.json and this case’s Lean output; --check recomputes it and requires the two to agree. The disagreements are about the 1966 empirical model (fitted to a few levels, no spin–orbit coupling), not about the proofs or the checker: the Lean intervals only guarantee that the eigenvalues of the discrete model lie inside them.
Command and Lean checker output
The certificates were written by bench/pw_cb_runs.py (driver kernels/pw/mini_pw.py, Q = 67/2) and archived in domains/dft/validation/ (the runs of DFT-W14 and DFT-W70–W78); the driver is not re-run here, the current checker re-checks the same certificates.
PROBLEM material=Si structure=diamond a_angstrom=5.43000000000000000000 Q=33.50000000000000000000 vs3=-0.21000000000000000000 vs8=0.040000000000000000 … PW_KPOINT 0 0 0 0 G PW_KPOINT 1 1 0 0 X PW_BAND 0 1 -0.079212128305701 -0.079212126305717 PW_BAND 0 2 0.384321562929469 0.384321564929453 PW_VBM 0.384321562929469 0.384321564929453 PW_CBM 0.414453504126865 0.414453506126850 PW_GAP 0.030131939197413 0.030131943197380 PW_DIRECT_GAP 0 0.125828955758147 0.125828959758115 PW_DIRECT_GAP 1 0.145317865167267 0.145317869167234 PW_OK 8 … and PW_KPOINT × 8, PW_BAND × 72, PW_DIRECT_GAP × 8 … 11 more certificates
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
03
1D mini-DFT, against DFTK
An electron pair in a double well: rHF ground state, bands and DOS DFT
The 1D periodic reduced Hartree–Fock (rHF) model: two Gaussian wells (V₀ = 2, σ = 0.5) in a cell of length L = 4, one electron pair, plane-wave cutoff K = 16, zero temperature, Γ point. After the self-consistent solve come non-self-consistent bands at 11 k-points on Γ→X and a Gaussian-broadened DOS (σ = 0.05 Ha) on 8 k-points. The Lean checker returns rigorous intervals for the ground-state energy, its truncation limit, all 33 eigenvalues, the gap, the energy components, the density error, and every band and DOS value.
Bands and density of states
Left: the driver’s bands (grey lines), the midpoints of the Lean intervals (blue dots; the intervals are under 4×10⁻⁴ wide, invisible at this scale) and DFTK (orange crosses). Right: the Lean intervals for the DOS (pointwise lower and upper bounds, about 10⁻³ of D wide).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| ground-state energy E_K (Ha) | [-2.592240050095254, -2.592240045480463] (width 4.6e-09) | minimum of the discrete rHF problem at cutoff K |
| truncation limit E_∞ (Ha) | [-2.592240050095254, -2.592240045480463] (width 4.6e-09) | E_K − E_∞ ≤ 0.000000004614790 |
| lowest eigenvalue ε₀ (Ha) | [-1.287640682708198, -1.287251130721002] (width 3.9e-04) | eigenvalue 0 of the ground-state Hamiltonian |
| gap | [-1.287251130721002, -0.319289528687437] (width 9.7e-01) | no eigenvalue in this interval |
| kinetic, external, Hartree energy | [0.067506335949670, 0.067718129885493] (width 2.1e-04) / [-2.677288516959576, -2.677112517792824] (width 1.8e-04) / [0.017330341593619, 0.017366131747901] (width 3.6e-05) | energy components of the ground state |
| density error | Coulomb norm ≤ 0.000067932245089, pointwise ≤ 0.003219614873737 | driver’s density minus the true ground-state density |
| bands, DOS | 627 band values (widest 3.9e-04), 201 DOS values (widest 5.2e-02) | a rigorous interval for every band on the Γ→X path and the DOS k-grid, and for every DOS energy |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| ground-state energy E_K (Ha) | -2.592240045484882 | [-2.592240050095254, -2.592240045480463] (width 4.6e-09) | ✓ inside |
| eigenvalues at Γ | 33 values | one by one | ✓ 33/33 inside |
| bands ε_i(k) on Γ→X (without the zone edge X) | 330 values | one by one | ✓ 330/330 inside; DFTK has a different number of plane waves at X |
| density of states D(E) | 201 values | one by one | ✓ 201/201 inside |
Command and Lean checker output
PROBLEM POT gauss V0 2 SIGMA 0.5 R 1
PROBLEM L 4
PROBLEM K 16
PROBLEM NPAIR 1
PROBLEM KPOINTS 1
POT_OK gauss 0.00000000000000052658
CERT_OK -2.592240050095254 -2.592240045480463
GAUSS_OK -2.592240050095255 -2.592240045480462
LIMIT_GAUSS_OK -2.592240050095255 -2.592240045480462
LIMIT_OK -2.592240050095254 -2.592240045480463 0.000000004614790
RHO_OK 0.000067932245089
RHOPT_OK 0.003219614873737
VH_OK 0.000194775931639
EIG_OK 0 -1.287640682708198 -1.287251130721002
EIG_OK 1 -0.319289528687437 -0.318899976700241
EIG_OK 2 0.262544983439772 0.262934535426968
EIG_OK 3 3.683647263204016 3.684036815191212
EIG_OK 4 3.701649338727949 3.702038890715144
… and BAND_OK × 627, DOS_OK × 201
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
A metal at finite temperature: free energy, entropy and energy DFT
The same model with L = 10 and two electron pairs, Fermi–Dirac smearing at T = 0.01 Ha; there is no gap (a metal). At finite temperature the rHF free energy has a variational principle, so the checker bounds it directly; the driver writes two more certificates at T ∓ 10⁻⁵, and the concavity of the free energy in T gives intervals for the entropy and the energy.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| free energy F = E − 2TS (Ha) | [-1.957329595295100, -1.957329595259820] (width 3.5e-11) | minimum of the finite-temperature discrete rHF problem |
| entropy S | [1.234288349887705, 1.234329046099463] (width 4.1e-05) | from the certificates at T ∓ h (concavity of F in T) |
| energy E (Ha) | [-1.932643828297345, -1.932643014337831] (width 8.1e-07) | E = F + 2TS |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| free energy F = E − 2TS (Ha) | -1.957329595265296 | [-1.957329595295100, -1.957329595259820] (width 3.5e-11) | ✓ inside |
| entropy S | 1.234308767025132 | [1.234288349887705, 1.234329046099463] (width 4.1e-05) | ✓ inside |
| energy E (Ha) | -1.932643419924793 | [-1.932643828297345, -1.932643014337831] (width 8.1e-07) | ✓ inside |
Command and Lean checker output
PROBLEM POT gauss V0 2 SIGMA 0.5 R 1 PROBLEM L 10 PROBLEM K 16 PROBLEM NPAIR 2 PROBLEM KPOINTS 1 PROBLEM T 0.00999000000000000061561866715464930166490375995635986328125 0.01000000000000000020816681711721685132943093776702880859375 0.01000999999999 … PROBLEM SMEAR fd POT_OK gauss 0.00000000000000133884 CERT_OK -1.957329595295100 -1.957329595259820 ENTROPY_OK 1.234288349887705 1.234329046099463 ENERGY_OK -1.932643828297345 -1.932643014337831 GAUSS_OK -1.957329595295105 -1.957329595259815 GAUSS_ENTROPY 1.234288349352170 1.234329046634999 GAUSS_ENERGY -1.932643828308061 -1.932643014327115
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
LDA (Slater exchange + PW92 correlation): the SCF solution DFT
Local density approximation: Slater exchange plus Perdew–Wang 92 correlation, as DFTK’s Xc([:lda_x, :lda_c_pw]). Marzari–Vanderbilt cold smearing at T = 0.01 Ha has no variational principle; the checker uses an interval Newton–Kantorovich argument to prove that exactly one exact solution of the SCF equations lies near the driver’s, and bounds its chemical potential, energy and free energy.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| solution of the SCF equations | exactly one within radius 5.1e-13 of the driver’s | local existence and uniqueness for the nonlinear SCF equations |
| chemical potential μ (Ha) | [-1.118102778468797, -1.118102778467769] (width 1.0e-12) | of that exact solution |
| energy E (Ha) | [-4.395711224027123, -4.395711222450693] (width 1.6e-09) | of that exact solution |
| free energy F (Ha) | [-4.400909924054979, -4.400909924054954] (width 2.5e-14) | E − 2T Σ s(x_i) of that exact solution |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| Fermi level εF / chemical potential μ (Ha) | -1.118102778468322 | [-1.118102778468797, -1.118102778467769] (width 1.0e-12) | ✓ inside |
| energy E (Ha) | -4.395711223238898 | [-4.395711224027123, -4.395711222450693] (width 1.6e-09) | ✓ inside |
| free energy F (Ha) | -4.400909924054980 | [-4.400909924054979, -4.400909924054954] (width 2.5e-14) | ✓ 8.9e-16 outside, within the double-precision rounding tolerance 1e-13 (as in the bench) |
Command and Lean checker output
PROBLEM POT gauss V0 2 SIGMA 0.5 R 1 PROBLEM L 10 PROBLEM K 8 PROBLEM NPAIR 2 PROBLEM KPOINTS 1 PROBLEM T 0.01000000000000000020816681711721685132943093776702880859375 PROBLEM SMEAR mv PROBLEM XC lda PROBLEM M 33 POT_OK gauss 0.00000000000000133884 LDA_OK 0.00000000000051321805 -1.118102778468797 -1.118102778467769 0.00000000000000077290 0.00000000000297779268 0.00000000000025660903 1.067402 LDA_ENERGY -4.395711224027123 -4.395711222450693 -4.400909924054979 -4.400909924054954 GAUSS_LDA 0.00000000002701177832 -1.118102778495295 -1.118102778441271 GAUSS_LDA_EF -4.395711264730940 -4.395711181746877 -4.400909924062062 -4.400909924047871
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
LDA with third-order Methfessel–Paxton smearing DFT
LDA (Slater + PW92) with L = 4 and two electron pairs, third-order Methfessel–Paxton smearing at T = 0.05 Ha. Higher-order MP occupations can exceed 1 or go negative; the SCF-solution certificate again gives local existence and uniqueness and intervals for μ, E and F.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| solution of the SCF equations | exactly one within radius 2.0e-13 of the driver’s | local existence and uniqueness for the nonlinear SCF equations |
| chemical potential μ (Ha) | [-1.066087578689828, -1.066087578689420] (width 4.1e-13) | of that exact solution |
| energy E (Ha) | [-6.296405131092821, -6.296405125236783] (width 5.9e-09) | of that exact solution |
| free energy F (Ha) | [-6.297134431845237, -6.297134431845201] (width 3.6e-14) | E − 2T Σ s(x_i) of that exact solution |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| Fermi level εF / chemical potential μ (Ha) | -1.066087578689641 | [-1.066087578689828, -1.066087578689420] (width 4.1e-13) | ✓ inside |
| energy E (Ha) | -6.296405128164800 | [-6.296405131092821, -6.296405125236783] (width 5.9e-09) | ✓ inside |
| free energy F (Ha) | -6.297134431845214 | [-6.297134431845237, -6.297134431845201] (width 3.6e-14) | ✓ inside |
Command and Lean checker output
PROBLEM POT gauss V0 2 SIGMA 0.5 R 1 PROBLEM L 4 PROBLEM K 8 PROBLEM NPAIR 2 PROBLEM KPOINTS 1 PROBLEM T 0.05000000000000000277555756156289135105907917022705078125 PROBLEM SMEAR mp3 PROBLEM XC lda PROBLEM M 33 POT_OK gauss 0.00000000000000052658 LDA_OK 0.00000000000020368478 -1.066087578689828 -1.066087578689420 0.00000000000000048911 0.00000000015720902543 0.00000000000010184239 1.956090 LDA_ENERGY -6.296405131092821 -6.296405125236783 -6.297134431845237 -6.297134431845201 GAUSS_LDA 0.00000000000557525180 -1.066087578695200 -1.066087578684048 GAUSS_LDA_EF -6.296405208311149 -6.296405048018455 -6.297134431846363 -6.297134431844076
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
An insulator with k-point sampling DFT
L = 4, two electron pairs, K = 12, three Γ-centred k-points, zero temperature. The k-point problem is equivalent to the Γ-point problem on a supercell three cells long; the checker proves the energy bounds on the supercell, divides by the number of cells, and also bounds the truncation limit.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| ground-state energy per cell (Ha) | [-3.564739489620383, -3.564739477201193] (width 1.2e-08) | discrete rHF minimum on the equivalent supercell, per cell |
| truncation limit (Ha) | [-3.564739489620383, -3.564739477201193] (width 1.2e-08) | truncation error ≤ 0.000000012419189 |
Against DFTK 0.8.0
| Quantity | DFTK | Lean | Agreement |
|---|---|---|---|
| ground-state energy per cell (Ha) | -3.564739477203016 | [-3.564739489620383, -3.564739477201193] (width 1.2e-08) | ✓ inside |
Command and Lean checker output
PROBLEM POT gauss V0 2 SIGMA 0.5 R 1 PROBLEM L 4 PROBLEM K 12 PROBLEM NPAIR 2 PROBLEM KPOINTS 3 POT_OK gauss 0.00000000000000052658 CERT_OK -3.564739489620383 -3.564739477201193 GAUSS_OK -3.564739489620385 -3.564739477201191 LIMIT_OK -3.564739489620383 -3.564739477201193 0.000000012419189
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
04
mini-FEM, against scikit-fem
Poisson with the manufactured solution sin πx FEM
−u″ = f on (0, 1) with u = 0 at both ends and f = c·sin πx, where c is the double nearest π², so the true solution is (c/π²) sin πx. P1 elements on a uniform mesh of 32 cells. The checker gives a rigorous interval for the true solution at every node, the nodal error, and the largest error between the true solution and the piecewise-linear finite-element solution over the whole interval.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| sup|u − V_h| | ≤ 0.001207687544605 | largest error of the finite-element solution over the whole interval |
| nodal error max|u(x_i) − V_i| | ≤ 0.00000000000000242231 | solver’s floating-point nodal values against the true solution |
| true solution u(0.5) | [0.999999999999999, 1.000000000000000] (width 1.0e-15) | one such interval at each of the 33 nodes (NODE_OK) |
Against scikit-fem 12.0.2
| Quantity | scikit-fem | Lean | Agreement |
|---|---|---|---|
| true solution u(x_i) (Green representation, 40-digit mpmath quadrature) | 33 values | one by one | ✓ 33/33 inside |
| scikit-fem nodal values V_i (ElementLineP1, 8-point Gauss) | 33 nodes | against the driver’s V_i | ✓ largest difference 3.3e-16 |
Command and Lean checker output
PROBLEM EQUATION -u''=f
PROBLEM L 1
PROBLEM BC0 DIRICHLET A 1 B 0 G 0
PROBLEM BCL DIRICHLET A 1 B 0 G 0
PROBLEM F ((9.869604401089357992304940125904977321624755859375)*sin(1*pi*x))
NODE_OK 0 0.00000000000000000000 0.000000000000000 0.000000000000000
NODE_OK 1 0.03125000000000000000 0.098017140329560 0.098017140329561
NODE_ERR 0.00000000000000242231
LINF_OK 0.001207687544605
… and NODE_OK × 33
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
Neumann at both ends: compatibility and the mean-zero normalization FEM
−u″ = e^{2x} on (0, 1) with u′(0) = 0, u′(1) = −(e² − 1)/2 (the decimal −3.194528049465325 on the command line), normalized by ∫u = 0. A solution exists only if ∫f + g₀ + g_L = 0; once the decimals are rounded to doubles this no longer holds exactly. The checker proves the defect c̄ to be about −9·10⁻¹⁷ (the interval excludes 0, so strictly the original problem has no solution) and certifies the precisely stated projected problem −u″ = e^{2x} − c̄. P1 elements, 24 cells; the discrete solution (the same as scikit-fem’s Lagrange-multiplier solution) differs from the true nodal values by one constant s, which the checker also encloses.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| sup|u − V_h| | ≤ 0.001603527799249 | largest error of the finite-element solution over the whole interval |
| nodal error max|u(x_i) − V_i| | ≤ 0.00000000000000097984 | solver’s floating-point nodal values against the true solution |
| compatibility defect c̄ | [-0.00000000000000008986, -0.00000000000000008985] (width 1.0e-20) | c̄ = (∫f + g₀ + g_L)/L; the certified problem is −u″ = f − c̄ (the original one when c̄ = 0) |
| constant s | [-0.00046211782024677687, -0.00046211782024677686] (width 0.0e+00) | u(x_i) = W_i + s, W the discrete solution with ∫W_h = 0 (SHIFT_OK) |
| true solution u(0.5) | [0.119061555251569, 0.119061555251570] (width 1.0e-15) | one such interval at each of the 25 nodes (NODE_OK) |
Against scikit-fem 12.0.2
| Quantity | scikit-fem | Lean | Agreement |
|---|---|---|---|
| true solution u(x_i) (projected problem, ∫u = 0, 40-digit mpmath quadrature) | 25 values | one by one | ✓ 25/25 inside |
| compatibility defect c̄ = (∫f + g₀ + g_L)/L | -8.98557e-17 (mpmath) | [-0.00000000000000008986, -0.00000000000000008985] (width 1.0e-20) | ✓ inside the interval |
| scikit-fem discrete solution (ElementLineP1, Lagrange multiplier for ∫u_h = 0) | differs from the driver’s V_i by the constant -4.621178e-04 | [-0.00046211782024677687, -0.00046211782024677686] (width 0.0e+00) | ✓ constant inside SHIFT_OK (float tolerance 1e-12), spread 4.1e-15 |
Command and Lean checker output
PROBLEM EQUATION -u''=f
PROBLEM L 1
PROBLEM BC0 NEUMANN A 0 B 1 G 0
PROBLEM BCL NEUMANN A 0 B 1 G -3.19452804946532520347091121948324143886566162109375
PROBLEM F exp(2*x)
COMPAT_OK -0.00000000000000008986 -0.00000000000000008985
SHIFT_OK -0.00046211782024677687 -0.00046211782024677686
NODE_OK 0 0.00000000000000000000 0.298632012366331 0.298632012366332
NODE_OK 1 6004799503160661/144115188075855872 0.297739333319357 0.297739333319358
NODE_ERR 0.00000000000000097984
LINF_OK 0.001603527799249
… and NODE_OK × 25
Generated 2026-10-05 · Lean toolchain v4.33.1 · status of this domain →
Adaptive refinement of a cantilever FEM
The beam equation u⁗ = f on (0, 1), clamped at the left end and free at the right, under a load f = 100·e^{10x} that grows sharply towards the free end. Cubic Hermite elements, starting from 4; each round refines by the per-element error bounds proved in Lean until sup|u − V_h| ≤ 10⁻⁶ on the whole interval.
Deflection and per-element error bounds
Top: the finite-element solution and the element end points. Bottom: the bound on sup|u − V_h| that the checker proves for each element (log scale); the dashed line is the tolerance 10⁻⁶. Refinement concentrates at the heavily loaded free end.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| sup|u − V_h| | ≤ 0.000000977151322 | largest error of the finite-element solution over the whole interval |
| nodal error max|u(x_i) − V_i| | ≤ 0.00000000005795191529 | solver’s floating-point nodal values against the true solution |
| slope error max|u'(x_i) − V_i'| | ≤ 0.00000000008093920381 | rotation of the beam |
| true solution u(1) | [62628.474409900415250, 62628.474409900446002] (width 2.9e-11) | one such interval at each of the 147 nodes (NODE_OK) |
| true slope u'(1) | [90308.409758707522338, 90308.409758707553090] (width 2.9e-11) | rotation at the free end (SLOPE_OK) |
| per-element bounds | 146 elements, largest 9.77e-07, smallest 6.10e-08 | ELEM_OK |
Against scikit-fem 12.0.2
| Quantity | scikit-fem | Lean | Agreement |
|---|---|---|---|
| true solution u(x_i) (fourfold integral of f, 40-digit mpmath quadrature) | 147 values | one by one | ✓ 147/147 inside |
| true slope u'(x_i) | 147 values | one by one | ✓ 147/147 inside |
| scikit-fem nodal values (ElementLineHermite, 8-point Gauss, same mesh) | 2.3e-02 from the true solution | NODE_ERR ≤ 0.00000000005795191529 | — solution of size 6.3e+04; the driver is 4.3e-11 from the true solution (banded LU with mixed-precision iterative refinement), scikit-fem (double-precision sparse direct solve) 2.3e-02 |
Command and Lean checker output
PROBLEM EQUATION u''''=f
PROBLEM L 1
PROBLEM END0 CLAMPED G 0 0
PROBLEM ENDL FREE G 0 0
PROBLEM B0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
PROBLEM BL 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1
PROBLEM G 0 0 0 0
PROBLEM F ((100)*exp(10*x))
NODE_OK 0 0.00000000000000000000 -0.000000000015376 0.000000000015376
NODE_OK 1 0.03125000000000000000 95.676211657212808 95.676211657243561
SLOPE_OK 0 0.00000000000000000000 -0.000000000015376 0.000000000015376
SLOPE_OK 1 0.03125000000000000000 6087.429086069786353 6087.429086069817105
ELEM_OK 0 0.000000339473588
ELEM_OK 1 0.000000463999827
NODE_ERR 0.00000000005795191529
SLOPE_ERR 0.00000000008093920381
LINF_OK 0.000000977151322
… and NODE_OK × 147, SLOPE_OK × 147, ELEM_OK × 146
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
First eigenvalue of the L-shaped membrane: an interval proved on both sides FEM
The first eigenvalue λ₁ of the Dirichlet Laplacian on the L-shaped domain ((0, 2)² minus the lower-right unit square), defined for the continuous problem (infimum of the Rayleigh quotient over H¹₀(Ω)). The eigenfunction is singular at the re-entrant corner, and ordinary finite elements only approximate λ₁. Upper bound: the discrete eigenvector of conforming Q1 elements as trial function, whose Rayleigh quotient the checker computes exactly. Lower bound: rotated Q1 nonconforming elements give σ; the checker verifies in exact rationals that K − σM is positive semidefinite and adds the correction from the Wirtinger constant on each cell. A 64 × 64 grid (3,072 cells) checks in about 40 s; the 128 × 128 grid gives [9.632674112571737, 9.645135475992192] (about 8 minutes).
The Q1 discrete eigenfunction
scikit-fem’s first Q1 eigenfunction on the same grid (contours 0.1, 0.2, …, 0.9, maximum normalised to 1); the gradient is singular at the re-entrant corner (1, 1). The Lean interval is about the eigenvalue of the continuous problem, not about this discrete function.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| upper bound on λ₁ | λ₁ ≤ 9.655274511290479 | the discrete first eigenvector of conforming Q1 elements as trial function; the checker computes its Rayleigh quotient exactly (FEM-B25, T4) |
| lower bound on λ₁ | λ₁ ≥ 9.620536290476300 (σ = 9.629702978944) | rotated Q1 nonconforming elements: the checker verifies in exact rationals that K − σM is positive semidefinite, and the Wirtinger constant h/π on each cell gives λ₁ ≥ σ/(1 + σh²/π²) (FEM-B26, T4) |
| first eigenvalue λ₁ | [9.620536290476300, 9.655274511290479] (width 3.5e-02), relative width 3.6e-03 | the eigenvalue of the continuous problem (infimum of the Rayleigh quotient over H¹₀(Ω)), not of a discrete matrix |
Against scikit-fem 12.0.2
| Quantity | scikit-fem | Lean | Agreement |
|---|---|---|---|
| scikit-fem Q1 discrete eigenvalue (same 64 × 64 mesh) | 9.655274511290 | upper bound 9.655274511290 | ✓ Lean’s upper bound is the Rayleigh quotient of the driver’s trial vector, 1.0e-13 above the discrete eigenvalue |
| scikit-fem Q2 discrete eigenvalue (same mesh; conforming, hence also an upper bound) | 9.641720693217 | [9.620536290476300, 9.655274511290479] (width 3.5e-02) | ✓ inside |
| literature value (Betcke–Trefethen 2005, numerical, not a proof) | 9.6397238440219 | [9.620536290476300, 9.655274511290479] (width 3.5e-02) | ✓ inside |
Command and Lean checker output
PROBLEM EQUATION -laplace(u) = lambda u in Omega, u = 0 outside Omega (first Dirichlet eigenvalue, min-max over Lipschitz functions) PROBLEM DOMAIN LSHAPE a 2 b 2 P 1 Q 1 PROBLEM GRID 64 64 32 32 UPPER_OK 9.655274511290479 LOWER_OK sigma 9.629702978943896596319973468780517578125 lo 9.620536290476300 EIG_INTERVAL 9.620536290476300 9.655274511290479
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
05
mini-MD, against ASE
Velocity Verlet trajectory of a Lennard-Jones chain MD
A 1D Lennard-Jones chain of 8 atoms (ε = σ = m = 1, free ends, all pairs), starting from the equilibrium lattice with random momenta at T₀ = 0.01, integrated by velocity Verlet with h = 0.001 for 1000 steps. The checker examines the whole trajectory in exact rationals: the initial energy, the energy and momentum drift, the Verlet residual, the local error of every step, and the global error between the data and the true solution from the initial state (the global bound grows like e^{Kt} and covers the first half).
Certified global error bound against the actual error
Position error between the true solution from the initial state and the computed trajectory: Lean’s TRUE_OK bound (blue) and the actual error measured against a high-accuracy DOP853 reference (orange), log scale. The global bound grows like e^{Kt}, so it covers the first 500 steps; the per-step local bound is almost exact (actual error / bound ≈ 0.99).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| initial energy H(Q₀, P₀) | [-7.186568898605097, -7.186568898605096] (width 8.9e-16) | also the energy of the true solution in the global tube |
| energy drift | ≤ 0.00000046826354893195 | max|H(t_k) − H(0)| on the computed trajectory (exact) |
| total momentum drift | ≤ 0.00000000000000029057 | exact |
| Verlet residual | ≤ 0.00000000000000044406 | the computed trajectory satisfies both velocity Verlet equations |
| one-step local error | position ≤ 0.00000000279045942382, momentum ≤ 0.00000004228626446212 | true solution from each data point against the next data point |
| global error (t = 0.5) | position ≤ 0.00099933951185418332, momentum ≤ 0.01514386287974667751 | true solution from the initial state against the data, for the first 500 steps (TRUE_HORIZON) |
Against ASE 3.29.0
| Quantity | ASE | Lean | Agreement |
|---|---|---|---|
| ASE VelocityVerlet trajectory (1000 steps) | 8 atoms × 1000 steps | against the trajectory in the certificate | ✓ max|q − q_ASE| = 4.6e-16, max|p − p_ASE| = 7.4e-15 |
| true solution minus data (scipy DOP853, t ≤ 0.5) | 500 times | TRUE_OK position and momentum bounds | ✓ actual error / Lean bound ≤ 9.9e-01 |
| one-step local error (DOP853, 50 sampled steps) | 50 steps | per-step LOCAL bounds | ✓ actual error / Lean bound ≤ 0.99 |
Command and Lean checker output
PROBLEM N 8
PROBLEM STEPS 1000
PROBLEM EPS 1
PROBLEM SIGMA 1
PROBLEM H 0.001000000000000000020816681711721685132943093776702880859375
PROBLEM MASS 0 1
PROBLEM MASS 1 1
PROBLEM MASS 2 1
PROBLEM MASS 3 1
PROBLEM MASS 4 1
PROBLEM MASS 5 1
PROBLEM MASS 6 1
PROBLEM MASS 7 1
PROBLEM Q0 0 0
PROBLEM Q0 1 1.1224620483093730172186042182147502899169921875
PROBLEM Q0 2 2.244924096618746034437208436429500579833984375
PROBLEM Q0 3 3.3673861449281190516558126546442508697509765625
PROBLEM Q0 4 4.48984819323749206887441687285900115966796875
… and LOCAL × 1000, TRUE_OK × 500
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
Lennard-Jones atoms in a periodic box: intervals for the true solution MD
32 atoms (an fcc 2³ lattice with small random displacements and momenta at T₀ = 0.01, seed 7) in a cubic periodic box of side L ≈ 3.17 with the minimum image, Lennard-Jones smoothly cut by a C⁴ switch between r_s = 1.1 and r_c = 1.5, velocity Verlet with h = 0.002 for 100 steps. The checker examines the whole trajectory in exact rationals, proves that the true solution of the Newton equations exists and is unique, and bounds the distance between the true solution and the computed trajectory — a statement about the continuous-time solution, not about the integrator’s discrete orbit. The global bound grows like e^(Kt) and covers the first part of the run.
Certified global error bound against the actual error
Position error between the true solution from the initial state and the computed trajectory: Lean’s TRUE_OK bound (blue) and the actual error measured against a high-accuracy DOP853 reference (orange), log scale.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| existence and uniqueness of the true solution | the Newton equations of the 32 atoms have a solution on [0, ∞) from the initial state, unique on every [0, T] | periodic box, minimum image, C⁴-smoothly cut LJ (MD-C12, B17) |
| initial energy H(Q₀, P₀) | [-189.972908475945618, -189.972908475945617] (width 0.0e+00) | also the energy of the true solution in the global tube |
| energy drift | ≤ 0.00031469304111286990 | max|H(t_k) − H(0)| on the computed trajectory (exact) |
| total momentum drift | ≤ 0.00000000000000118656 | exact |
| Verlet residual | ≤ 0.00000000000000022205 | the computed trajectory satisfies both velocity Verlet equations |
| one-step local error | position ≤ 0.00000021838735379189, momentum ≤ 0.00000639056145983239 | true solution from each data point against the next data point |
| global error (t = 0.144) | position ≤ 0.00094883561665340393, momentum ≤ 0.02776530883413702509 | true solution from the initial state against the data, for the first 72 steps (TRUE_HORIZON) |
| length of the global tube | 72 steps (t ≤ 0.144) | of 100; the global bound grows like e^(Kt) with K ≤ 47.9 (MD-B16) |
Against ASE 3.29.0
| Quantity | ASE | Lean | Agreement |
|---|---|---|---|
| ASE VelocityVerlet trajectory (100 steps, the same potential wrapped as an ASE Calculator) | 32 atoms × 100 steps | against the trajectory in the certificate | ✓ max|q − q_ASE| = 4.4e-16, max|p − p_ASE| = 1.9e-15 |
| true solution minus data (scipy DOP853, t ≤ 0.144) | 72 times | TRUE_OK position and momentum bounds | ✓ actual error / Lean bound ≤ 9.2e-01 |
Command and Lean checker output
PROBLEM N 32
PROBLEM STEPS 100
PROBLEM EPS 1
PROBLEM SIGMA 1
PROBLEM RS 1.100000000000000088817841970012523233890533447265625
PROBLEM RC 1.5
PROBLEM L 3.174802103936399166883575162501074373722076416015625
PROBLEM H 0.00200000000000000004163336342344337026588618755340576171875
PROBLEM MASS 0 1
PROBLEM MASS 1 1
PROBLEM MASS 2 1
PROBLEM MASS 3 1
PROBLEM MASS 4 1
PROBLEM MASS 5 1
PROBLEM MASS 6 1
PROBLEM MASS 7 1
PROBLEM MASS 8 1
PROBLEM MASS 9 1
… and LOCAL × 100, TRUE_OK × 72
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →
06
mini-CALPHAD, against pycalphad
The fcc miscibility gap in Al–Zn CALPHAD
Al–Zn (the alzn_mey.tdb database shipped with pycalphad) at 600 K and x(Zn) = 0.3, where fcc splits into two phases of different composition. The checker proves the global equilibrium over the whole composition range: rigorous intervals for the equilibrium Gibbs energy, the chemical potentials and both phase compositions; that liquid and hcp are absent; the driving force of each phase; and that an equilibrium exists.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| equilibrium Gibbs energy G* (J/mol) | [-22985.126672267, -22985.126672266] (width 1.0e-09) | every feasible configuration ≥ lower bound; the certificate’s configuration ≤ upper bound |
| chemical potential μ(AL) (J/mol) | [-20590.725231961, -20590.725231960] (width 1.0e-09) | any equilibrium |
| chemical potential μ(ZN) (J/mol) | [-28572.063366314, -28572.063366313] (width 1.0e-09) | any equilibrium |
| FCC_A1 composition x(ZN) | [0.220126163802, 0.220126409545] (width 2.5e-07) | in any equilibrium, this phase’s composition lies here |
| FCC_A1 composition x(ZN) | [0.491533032729, 0.491533329798] (width 3.0e-07) | in any equilibrium, this phase’s composition lies here |
| LIQUID | absent | in no equilibrium (ABSENT) |
| HCP_A3 | absent | in no equilibrium (ABSENT) |
| driving force of LIQUID (J/mol) | [-896.512026734, -896.512026733] (width 1.0e-09) | max_y (μ·X(y) − G(y)), pycalphad’s DF |
| driving force of FCC_A1 (J/mol) | [-0.000000001, 0.000000001] (width 2.0e-09) | max_y (μ·X(y) − G(y)), pycalphad’s DF |
| driving force of HCP_A3 (J/mol) | [-379.321840947, -379.321840946] (width 1.0e-09) | max_y (μ·X(y) − G(y)), pycalphad’s DF |
| existence | exists | EXISTS |
Against pycalphad 0.11.1
| Quantity | pycalphad | Lean | Agreement |
|---|---|---|---|
| equilibrium Gibbs energy G (J/mol) | -22985.126672247 | [-22985.126672267, -22985.126672266] (width 1.0e-09) | — outside (1.9e-08 away); see the exact-model rows below |
| chemical potential μ(AL) (J/mol) | -20590.725231960 | [-20590.725231961, -20590.725231960] (width 1.0e-09) | ✓ inside |
| chemical potential μ(ZN) (J/mol) | -28572.063366314 | [-28572.063366314, -28572.063366313] (width 1.0e-09) | ✓ inside |
| G of pycalphad’s own equilibrium configuration (exact model, 200 bits) | -22985.126672247446571 | [-22985.126672267, -22985.126672266] (width 1.0e-09) | ✓ 1.9e-08 above the upper bound; every feasible configuration must be ≥ the Lean lower bound |
| G* (exact model, 200 bits, common tangent by Newton iteration) | -22985.126672266436731 | [-22985.126672267, -22985.126672266] (width 1.0e-09) | ✓ inside |
| μ(AL) (same) | -20590.725231960394963 | [-20590.725231961, -20590.725231960] (width 1.0e-09) | ✓ inside |
| μ(ZN) (same) | -28572.063366313867524 | [-28572.063366314, -28572.063366313] (width 1.0e-09) | ✓ inside |
| FCC_A1 composition x(ZN) (same) | 0.22012628667395587827 | [0.220126163802, 0.220126409545] (width 2.5e-07) | ✓ inside |
| FCC_A1 composition x(ZN) (same) | 0.49153318126357613364 | [0.491533032729, 0.491533329798] (width 3.0e-07) | ✓ inside |
| FCC_A1 composition x(ZN) | 0.220126287 | [0.220126163802, 0.220126409545] (width 2.5e-07) | ✓ inside |
| FCC_A1 composition x(ZN) | 0.491533181 | [0.491533032729, 0.491533329798] (width 3.0e-07) | ✓ inside |
| driving force of LIQUID (J/mol) | -896.512026734 | [-896.512026734, -896.512026733] (width 1.0e-09) | ✓ inside |
| driving force of FCC_A1 (J/mol) | -0.000000000 | [-0.000000001, 0.000000001] (width 2.0e-09) | ✓ inside |
| driving force of HCP_A3 (J/mol) | -379.321840947 | [-379.321840947, -379.321840946] (width 1.0e-09) | ✓ inside |
pycalphad works in double precision, which is too coarse to compare its G and μ directly with Lean intervals 1e-9 wide; its magnetic term also uses τ = T/(T_c + 10⁻⁹), whereas the certificate uses the exact continuous form (a documented convention difference in the CALPHAD ledger). The rows marked “exact model, 200 bits” take pycalphad’s own symengine model expression, set that 10⁻⁹ to zero and evaluate in 200-bit MPFR: first G of pycalphad’s own equilibrium configuration (any feasible configuration must lie above the Lean lower bound), then the common tangent solved by Newton iteration from pycalphad’s equilibrium — a high-precision equilibrium of the same model. Rows marked “—” are pycalphad’s raw double-precision values, recorded only.
Command and Lean checker output
PROBLEM MODEL CERT PROBLEM T 600 PROBLEM X0 0.3 PROBLEM R 8.3145 PROBLEM PHASES LIQUID FCC_A1 HCP_A3 G_OK -22985.126672267 -22985.126672266 TANGENT cert EPS -0.000000000002425 FRAC FCC_A1 0.220126286673 0.705704921583 FRAC FCC_A1 0.491533181263 0.294295078416 MU_OK A -20590.725231961 -20590.725231960 MU_OK B -28572.063366314 -28572.063366313 KAPPA 0.000000000022 ABSENT LIQUID WINDOW FCC_A1 0.220126163802 0.220126409545 WINDOW FCC_A1 0.491533032729 0.491533329798 ABSENT HCP_A3 DF_OK LIQUID -896.512026734 -896.512026733
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
Cr–Fe: a bcc miscibility gap caused by magnetism CALPHAD
The bcc phase of Cr–Fe (crfe_bcc_magnetic.tdb, with the Inden–Hillert–Jarl magnetic term) at 600 K and x(Fe) = 0.5. Ferromagnetism on the iron side splits bcc into two phases; the chromium end is antiferromagnetic. The checker proves the global equilibrium, including bounds on the magnetic term.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| equilibrium Gibbs energy G* (J/mol) | [-19793.346508744, -19793.346508743] (width 1.0e-09) | every feasible configuration ≥ lower bound; the certificate’s configuration ≤ upper bound |
| chemical potential μ(CR) (J/mol) | [-19642.803579468, -19642.803579467] (width 1.0e-09) | any equilibrium |
| chemical potential μ(FE) (J/mol) | [-19943.889438020, -19943.889438019] (width 1.0e-09) | any equilibrium |
| BCC_A2 composition x(FE) | [0.432097736118, 0.432097836657] (width 1.0e-07) | in any equilibrium, this phase’s composition lies here |
| BCC_A2 composition x(FE) | [0.903631494701, 0.903631563012] (width 6.8e-08) | in any equilibrium, this phase’s composition lies here |
| driving force of BCC_A2 (J/mol) | [-0.000000001, 0.000000001] (width 2.0e-09) | max_y (μ·X(y) − G(y)), pycalphad’s DF |
| existence | exists | EXISTS |
Against pycalphad 0.11.1
| Quantity | pycalphad | Lean | Agreement |
|---|---|---|---|
| equilibrium Gibbs energy G (J/mol) | -19793.346508745 | [-19793.346508744, -19793.346508743] (width 1.0e-09) | — outside (8.8e-10 away); see the exact-model rows below |
| chemical potential μ(CR) (J/mol) | -19642.803579461 | [-19642.803579468, -19642.803579467] (width 1.0e-09) | — outside (5.9e-09 away); see the exact-model rows below |
| chemical potential μ(FE) (J/mol) | -19943.889438029 | [-19943.889438020, -19943.889438019] (width 1.0e-09) | — outside (8.9e-09 away); see the exact-model rows below |
| G of pycalphad’s own equilibrium configuration (exact model, 200 bits) | -19793.346508743741646 | [-19793.346508744, -19793.346508743] (width 1.0e-09) | ✓ inside; every feasible configuration must be ≥ the Lean lower bound |
| G* (exact model, 200 bits, common tangent by Newton iteration) | -19793.346508743835235 | [-19793.346508744, -19793.346508743] (width 1.0e-09) | ✓ inside |
| μ(CR) (same) | -19642.803579467780949 | [-19642.803579468, -19642.803579467] (width 1.0e-09) | ✓ inside |
| μ(FE) (same) | -19943.889438019889521 | [-19943.889438020, -19943.889438019] (width 1.0e-09) | ✓ inside |
| BCC_A2 composition x(FE) (same) | 0.43209778638735056038 | [0.432097736118, 0.432097836657] (width 1.0e-07) | ✓ inside |
| BCC_A2 composition x(FE) (same) | 0.90363152885665802856 | [0.903631494701, 0.903631563012] (width 6.8e-08) | ✓ inside |
| BCC_A2 composition x(FE) | 0.432097786 | [0.432097736118, 0.432097836657] (width 1.0e-07) | ✓ inside |
| BCC_A2 composition x(FE) | 0.903631529 | [0.903631494701, 0.903631563012] (width 6.8e-08) | ✓ inside |
| driving force of BCC_A2 (J/mol) | -0.000000000 | [-0.000000001, 0.000000001] (width 2.0e-09) | ✓ inside |
pycalphad works in double precision, which is too coarse to compare its G and μ directly with Lean intervals 1e-9 wide; its magnetic term also uses τ = T/(T_c + 10⁻⁹), whereas the certificate uses the exact continuous form (a documented convention difference in the CALPHAD ledger). The rows marked “exact model, 200 bits” take pycalphad’s own symengine model expression, set that 10⁻⁹ to zero and evaluate in 200-bit MPFR: first G of pycalphad’s own equilibrium configuration (any feasible configuration must lie above the Lean lower bound), then the common tangent solved by Newton iteration from pycalphad’s equilibrium — a high-precision equilibrium of the same model. Rows marked “—” are pycalphad’s raw double-precision values, recorded only.
Command and Lean checker output
PROBLEM MODEL CERT PROBLEM T 600 PROBLEM X0 0.5 PROBLEM R 8.3145 PROBLEM PHASES BCC_A2 G_OK -19793.346508744 -19793.346508743 TANGENT cert EPS 0.000000000000000 FRAC BCC_A2 0.432097786387 0.855997127041 FRAC BCC_A2 0.903631528856 0.144002872958 MU_OK A -19642.803579468 -19642.803579467 MU_OK B -19943.889438020 -19943.889438019 KAPPA 0.000000000023 WINDOW BCC_A2 0.432097736118 0.432097836657 WINDOW BCC_A2 0.903631494701 0.903631563012 DF_OK BCC_A2 -0.000000001 0.000000001 EXISTS
Generated 2026-10-03 · Lean toolchain v4.33.1 · status of this domain →
The Al–Zn phase diagram CALPHAD
Al–Zn from 400 to 900 K in steps of 50 K. At each temperature the driver divides the composition range into single-phase regions and two-phase regions (common tangents) and writes one certificate per region. The checker proves each one: the global equilibrium, windows for the phase-boundary compositions, and the absent phases. Each temperature also gets a partition certificate: the checker proves the phases of every equilibrium at every composition in [0, 1], so no region is missed.
Lean-certified phase boundaries on pycalphad’s phase diagram
Lines: pycalphad’s binplot on the same database and phases. Dots: the centres of the Lean windows at both ends of each common tangent; the widest window is 4×10⁻⁶, a point at this scale.
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| regions | 35 (12 common tangents), T = 400…900 K | one certificate per region |
| equilibrium Gibbs energy, chemical potentials | G_OK and MU_OK in all 35 certificates | at a representative point of each region |
| phase-boundary compositions | 47 WINDOWs, widest 3.7e-06 | compositions at both ends of each tangent |
| existence | EXISTS in 35/35 certificates | |
| partition of [0, 1] | 11/11 temperatures: 23 single-phase and 12 two-phase regions cover [0, 1] | every equilibrium at every composition of a region has its phases (CAL-B05) |
| phase-boundary uncertainty | 24 unresolved strips, the widest 4.0e-09 | only the phases of the two neighbouring regions occur |
Against pycalphad 0.11.2
| Quantity | pycalphad | Lean | Agreement |
|---|---|---|---|
| equilibrium phases and boundary compositions of each region | 35 regions | WINDOW, ABSENT | ✓ 35/35 regions agree (same phases; two-phase compositions within the Lean windows ±1e-6) |
| equilibrium phases at every point of an x grid | 550 points (x = 0.01, 0.03, …, 0.99) | REGION | ✓ 550/550 points lie in a certified region and agree (two-phase compositions within the windows ±1e-6) |
Command and Lean checker output
PROBLEM MODEL CERT
PROBLEM TDB_LINE_IGNORED AL ZN LIQUID FCC_A1 HCP_A3
PROBLEM T 400
PROBLEM X0 0.0128700596
PROBLEM R 8.3145
PROBLEM PHASES LIQUID FCC_A1 HCP_A3
G_OK -11852.667456941 -11852.667456940
TANGENT checker
EPS -0.000000000000001
FRAC FCC_A1 0.012870059600 1.000000000000
MU_OK A -11759.040272152 -11759.040272149
MU_OK B -19033.846116203 -19033.846116023
KAPPA 0.000000089579
ABSENT LIQUID
WINDOW FCC_A1 0.012869187494 0.012870931728
ABSENT HCP_A3
DF_OK LIQUID -4854.208679780 -4854.208679598
DF_OK FCC_A1 -0.000000090 0.000000090
… 45 more certificates
Generated 2026-10-05 · Lean toolchain v4.33.1 · status of this domain →
The Pb–Sn eutectic temperature, proved for every real temperature in an interval CALPHAD
Pb–Sn (pbsn.tdb shipped with pycalphad: liquid, FCC_A1, BCT_A5), a simple eutectic near the overall composition x(Sn) = 0.73735. Temperature is not sampled on a grid: the driver splits 434.55–474.57 K into 34 temperature-interval certificates, and on each one the checker’s conclusions hold at every real temperature — no liquid at any composition on the low side, liquid only at x₀ on the high side, and windows for the three phase compositions in the segment that straddles T_E. Together they put the temperature at which liquid first appears, T_E, in an interval 0.02 K wide (CAL-B15, T4).
The Pb–Sn phase diagram and the proved eutectic
Grey lines: pycalphad’s binplot on the same database and phases. Blue line: Lean’s T_E interval (0.02 K wide, a line at this scale); blue dots: centres of the three phase-composition windows. Orange bars at x₀: the temperature ranges proved to have no liquid (lower, at every composition) and liquid only at x₀ (upper).
Certified by Lean
| Quantity | Lean result | Meaning |
|---|---|---|
| eutectic temperature T_E (K) | [454.552100, 454.572100] (width 2.0e-02) | the temperature at which liquid first appears: the next two rows together (CAL-B15, T4) |
| no liquid (K) | [434.552100, 454.552100] (width 2.0e+01) | at every real temperature in this range, every composition x ∈ (0, 1) and every equilibrium, the liquid fraction is 0 |
| liquid only (K) | [454.572100, 474.572100] (width 2.0e+01) | at every temperature in this range, every equilibrium at x(Sn) = 0.73735 is liquid only |
| three-phase equilibrium: α (FCC_A1, Pb side) composition x(Sn) | [0.252700414305, 0.273906167874] (width 2.1e-02) | in every equilibrium at x₀ inside the T_E interval |
| three-phase equilibrium: liquid composition x(Sn) | [0.727251726177, 0.747253079665] (width 2.0e-02) | in every equilibrium at x₀ inside the T_E interval |
| three-phase equilibrium: β (BCT_A5, Sn side) composition x(Sn) | [0.972920291847, 0.978022551434] (width 5.1e-03) | in every equilibrium at x₀ inside the T_E interval |
| model and existence | MODEL_OK、EXISTS | every segment’s model is the TDB model of that temperature range; an equilibrium exists at every temperature in [T_a, T_b] |
Against pycalphad 0.11.2
| Quantity | pycalphad | Lean | Agreement |
|---|---|---|---|
| pycalphad’s eutectic temperature (steps of 0.002 K) | 454.5611 | [454.552100, 454.572100] (width 2.0e-02) | ✓ last without liquid 454.560 K, first with liquid 454.562 K |
| the no-liquid range: 5 temperatures × 25 compositions | 125 equilibria | SOLID_BELOW | ✓ no liquid in any of them |
| the liquid-only range: 7 temperatures at x₀ | 7 equilibria | ONLY_ABOVE | ✓ liquid only in all of them |
| FCC_A1 composition x(Sn) at T = 454.552100 K | 0.263189016 | [0.252700414305, 0.273906167874] (width 2.1e-02) | ✓ inside |
| BCT_A5 composition x(Sn) at T = 454.552100 K | 0.975522173 | [0.972920291847, 0.978022551434] (width 5.1e-03) | ✓ inside |
| LIQUID composition x(Sn) at T = 454.572100 K | 0.737350000 | [0.727251726177, 0.747253079665] (width 2.0e-02) | ✓ inside |
Command and Lean checker output
PROBLEM R 8.3145 PROBLEM TDB pbsn.tdb 4469 PROBLEM TDB_SHA256 6bbbd02a49fd711e03277b6be1bc1595de77d71a3fcc26e733fab3c417f8c74d PROBLEM COMPS PB SN PROBLEM PHASES LIQUID FCC_A1 BCT_A5 MODEL_OK PROBLEM X0 0.73735 BOXES 4 1 29 SOLID_BELOW LIQUID 434.552100 454.552100 ONLY_ABOVE LIQUID 454.572100 474.572100 EUTECTIC_T 454.552100 454.572100 THREE alpha FCC_A1 0.252700414305 0.273906167874 THREE liquid LIQUID 0.727251726177 0.747253079665 THREE beta BCT_A5 0.972920291847 0.978022551434 EXISTS
Generated 2026-10-17 · Lean toolchain v4.33.1 · status of this domain →