3.320: Lecture 9 (Mar 3 2005) SUCCESS AND FAILURE FAILURE Photos of soccer players removed for copyright reasons. Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Pseudopotentials Occupation, Eigenvalue Al Z =13 3p 3s 1 -2.7 eV 2 -7.8 eV 2p 2s 6 -69.8 eV 2 -108 eV 1s 2 -1512 eV (- 1 ∆2 + υeff) ψj = εjψj 2 Valence 2p 1s 1 -2.7 eV 2 -7.8 eV Pseudo- Al Z =3 Core-states 1 ∆2 (- 2 (ps) (ps) + υeff ) ψj Figure by MIT OCW. After Pehlke, Eckhard. Lecture on "The Plane-Wave Pseudopotential Method." Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari (ps) = εjψj Bloch Theorem & & & & \ nk& (r ) unk& (r ) exp ik r periodic u is expanded in planewaves, labeled according to the reciprocal lattice vectors & unk& (r ) & & ¦& c exp(i G r ) & G & nk G Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Other possibilities - many • Gaussian basis sets (Hartree-Fock codes) • Real space representations • LCAO (Linear combination of atomic orbitals) • LMTO (Linear muffin-tin orbitals), LAPW (Linearized augmented plane waves), PAW (Projector-augmented wave) Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Brillouin Zone integrations • Sampling at one point (the best – Baldereschi point, or the simplest – Gamma point) • Sampling at regular meshes (Monkhorst-Pack grids) • For metallic systems, integration of the discontinuity is improved introducing a fictitious electronic temperature Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Iterations to Selfconsistency • Construct the external potential (array of non-local psp) • Choose the plane-wave basis set cutoff, k-point sampling • Pick a trial electronic density • Construct the Hamiltonian operator: Hartree and exchange-correlation • Solve Kohn-Sham equations for the given Hamiltonian (e.g. by diagonalization) • Calculate the new charge density • Iterate Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Self-consistent ground state • Iterative diagonalizations (Davidson, Lancsoz, non-scf conjugate gradients + mixing strategy) Hˆ & \ (r) & ªn r º¼ ¬ & E\ (r ) • Direct minimization (Car-Parrinello, conjugate gradients) Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Matrix Diagonalization Hˆ & \ (r ) & ªn r º¼ ¬ & E\ (r ) \ ¦c n Mn n 1,k ¦c n M m Ĥ M n Ecm n 1, k § H11 ¨ ¨ . ¨ . ¨ ¨ . ¨H © k1 Mar 3 2005 ...... ...... H1k · § c1 · ¸ ¨ ¸ . ¸ ¨ .¸ . ¸ ¨ . ¸ ¸ ¨ ¸ . ¸ ¨ .¸ H kk ¸¹ ¨© ck ¸¹ § c1 · ¨ ¸ ¨ . ¸ E ¨ . ¸ ¨ ¸ ¨ . ¸ ¨ c ¸ © k¹ 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Constrained non-linear minimization E >^\ i `@ N & 1 * & 2 & & ³\ i (r ) \ i (r ) dr E H [n(r )] ¦ 2 i 1 & & & & E xc [n(r )] ³ Vext (r ) n(r ) dr & EH [n(r )] Mar 3 2005 & & 1 n(r1 )n(r2 ) & & & & d r1 dr2 ³³ 2 | r1 r2 | 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Constrained non-linear minimization Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Practice and Outlook • Techniques and applications (from k-points to DNA and superconductivity) • Beyond GGA: WDA, TDDFT, QMC • Connection to approximate methods Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Phonons Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Phonon Dispersions (Linear Response Theory) 600 Frequency (cm-1) Si 400 200 0 Γ K X Γ L X W Figure by MIT OCW. Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari L DOS B0(T)/B0(298 K) Thermodynamical Properties (Vibrational Free Energy) 1.002 1 0.998 0 50 100 150 200 250 300 350 400 TEMPERATURE (K) Our ab-initio results Experimental data (McSkimin, H. J., et. al. J Appl Phys 43 (1972): 2944. Figure by MIT OCW. Ab-initio results compared to experimental data of H.J. McSkimin et al., J. Appl. Phys. 43, 2944 -1972. Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Dynamics in Al(110) <110> <001> Surface 2nd layer “Bulk” “Bulk” 2nd layer Surface Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Al(110) Mean Square Displacements Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Computational LEED Graph removed for copyright reasons. Interlayer relaxation vs. Temperature (K). Source: Marzari, N., et al. “Thermal contraction and disordering of the Al(110) surface.” Physical Review Letters 82, no.16 (1999): pp.3296-9. Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari References (theory) • W. Koch, M. C. Holthausen, A Chemist’s Guide to Density Functional Theory • R. G. Parr, W. Yang, Density-Functional Theory of Atoms and Molecules • W. Kohn, Nobel lecture • F. Jensen, Introduction to Computational Chemistry • J. M. Thijssen, Computational Physics • B. H. Bransden and C. J. Joachim, Physics of Atoms and Molecules • K. Burke: The ABC of DFT, http://dft.rutgers.edu/kieron/beta/ Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari References (practice) • Payne, Teter, Allan, Arias, Joannopoulos, Review of Modern Physics 64, 1045 (1992). • Lecture notes from http://www.FHI-Berlin.MPG.DE/th/Meetings/FHImd2001/program.html , (L3 Pehlke, L2 Kratzer, L4 Fuchs) • Hartree-Fock for solids, Dovesi et al., Physica Status Solidi (b) 217, 63 (2000). Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari Software • Gaussian (http://www.gaussian.com) ($$) (chemistry, Hartree-Fock, DFT, correlated approaches) • Gamess-UK (http://www.cse.clrc.ac.uk/qcg/gamess-uk/ ) ($) (chemistry, Hartree-Fock, DFT, correlated approaches) • Materials Studio/Cerius (http://www.accelrys.com) ($$) (DFT, planewave, ultrasoft) • Crystal (http://www.chimifm.unito.it/teorica/crystal) ($) (Hartree-Fock) • VASP (http://cms.mpi.univie.ac.at/vasp) ($) (DFT, planewave, ultrasoft, PAW) • ESPRESSO (http://www.pwscf.org) (free) (DFT, planewave, ultrasoft, linearresponse theory, Car-Parrinello) • ABINIT (http://www.abinit.org) (free) (DFT, planewave, linear-response theory, GW) • CPMD (http://www.cpmd.org) (free) (DFT, planewave, Car-Parrinello, timedependent DFT) • CASINO (http://www.tcm.phy.cam.ac.uk/~mdt26/cqmc.html ) (free) (Quantum Monte Carlo) Mar 3 2005 3.320 Atomistic Modeling of Materials -- Gerbrand Ceder and Nicola Marzari 3.320/SMA5107: Lecture 9b (3/3/05) Applications and Performance of DFT methods 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Spin Polarization: The Local Spin Density Approximation (LSDA) • Electrons have spin +(-) 1/2 µB • Spin is treated as a scalar quantity (this is approximate, as relation to angular momentum makes it a vector quantity) • Two spin states often referred to as “up” and “down” • Up-Up interaction is different from Up-Down 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Exchange Effects: Refresher Hund’s rule: A consequence of Pauling exclusion principle e.g. for atomic d-levels Even in solids where energy levels are split, can get parallel filling if splitting is not too large Why is this important ? Filling of different orbitals may give the atom different chemical properties Materials can carry a magnetic moment ρ↑ ≠ ρ↓ 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Spins in Density Functional Theory In principle Exc[ρ] “knows” about this effect, but in practice it is poorly approximated since only total charge density is variable But in practice, need to help LDA along … Solution: Treat up and down densities separately 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Local Spin Density (LSD) = LDA with different charge density for up and down electrons ↑ veff = v(r) + ↓ veff = v(r) + ρ (r' ) ∫ r − r' ρ (r' ) ∫ r − r' dr' + dr' + [ ∂Exc ρ ↑ , ρ ↓ ] ∂ρ ↑ [ ∂Exc ρ ↑ , ρ ↓ ∂ρ ↓ ] Up and down charge density can be different ρ↑ ≠ ρ b Similar to restricted and unrestricted Hartree Fock Regular LDA can not capture exchange effect well since it is non-local A spin-polarized version of GGA exists as well 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Charge Density: LiCoO2 O Co Li 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Spin Density: filters out unpolarized ions (e.g. O2- ) Ni2+ 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari How good is the output from LDA/GGA ? Computational quantum mechanics provides very basic information: Energy, charge density, band structure, optimized atomic positions, etc. How do I make a faster car from this ? First evaluate accuracy of basic information -> then understands how that propagates into higher order models 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Photo of sports car removed for copyright reasons. Energies of Atoms energy in Ry Li Be GGA: PW91 LDA Expt. 214.928 214.668 214.958 229.296 228.892 229.334 energy in Ry Na Mg GGA: PW91 LDA Expt. energy in Ry GGA: PW91 LDA 2324.514 2322.867 2324.49 2400.12 2398.265 2400.086 B 249.24 248.686 249.308 Al 2484.686 2482.618 2484.672 C 275.561 274.849 275.688 Si 2578.669 2576.384 2578.696 N 2108.926 2108.045 2109.174 O 2149.997 2148.939 2150.126 P 2682.386 2679.88 2682.764 S 2796.152 2793.419 2796.2 2199.433 2198.189 2199.45 Cl 2920.278 2917.313 2920.298 Ne 2257.893 2256.455 2257.856 Ar 21055.077 21051.876 21055.098 K Ca Ga Ge As Se Br Kr 21199.825 21196.383 21355.144 21351.466 23850.018 23843.66 24154.2 24147.583 24471.917 24465.038 24803.334 24796.191 25148.619 25141.209 25507.943 25500.263 LDA underestimates stability of atom, GGA is closer 3/3/05 F 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Small Molecules Binding Energy exp (ev) H2 LiH O2 H20 F2 -4.753 -2.509 -5.230 -10.078 -1.66 LDA -4.913 -2.648 -7.595 -11.567 -3.32 GGA -4.540 -2.322 -6.237 -10.165 Binding energy too high in LDA, GGA is closer but sometimes bound to weak. Pure Hartree Fock without corrections is terrible. 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari HF -3.64 -1.28 1.37 Lattice Parameters in Solids exp Si Ge GaAs Al Cu Ag Ta W Pt Au LDA ∆ GGA ∆ 5.427 5.65 5.65 5.4 5.62 5.62 -0.50% -0.53% -0.53% 5.49 5.74 5.73 1.16% 1.59% 1.42% 4.03 3.60 4.07 3.30 3.16 3.91 4.06 3.98 3.52 4.00 3.26 3.14 3.90 4.05 -1.31% -2.35% -1.69% -1.12% -0.67% -0.41% -0.13% 4.09 3.62 4.17 3.32 3.18 3.97 4.16 1.57% 0.44% 2.47% 0.80% 0.67% 1.49% 2.48% LDA tends to “overbind”, GGA “underbinds” GGA error more variable 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Bulk Modulus in Solids (in GPA) exp Si Ge GaAs Al Cu Ag Ta W Pt Au ∆ LDA 99 77 76 96 78 74 -3.03% 1.30% -2.63% 83 61 65 77 138 102 193 310 283 172 84 192 139 224 337 307 198 9.09% 39.13% 36.27% 16.06% 8.71% 8.48% 15.12% 73 151 85 197 307 246 142 LDA tends to be too stiff. GGA too soft 3/3/05 ∆ GGA 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari -16.16% -20.78% -14.47% -5.19% 9.42% -16.67% 2.07% -0.97% -13.07% -17.44% Oxides exp MgO TiO2 (a) TiO2 (c) Al2O2 BaTiO3 PbTiO3 SnO2 β-MnO2 (a) b-MnO2 (c) 3/3/05 4.21 4.59 2.958 5.128 4 3.9 4.737 4.404 2.876 LDA 4.17 4.548 2.944 5.091 3.94 3.833 4.637 4.346 2.81 ∆ GGA -0.95% -0.92% -0.47% -0.72% -1.50% -1.72% -2.11% -1.32% -2.29% 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari ∆ 4.623 2.987 5.185 0.72% 0.98% 1.11% 3.891 -0.23% 4.444 2.891 0.91% 0.52% Summary of Geometry Prediction LDA under-predicts bond lengths (always ?) GGA error is less systematic though over-prediction is common. errors are in many cases < 1%, for transition metal oxides < 5% 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Predicting Structure: The Energy Scales Accuracy required to predict physical behavior is astonishing V Atomic energy: -1894.074 Ry Fcc V : -1894.7325 Ry Bcc V : -1894.7125 Ry Cohesive energy is 0.638 Ry (0.03% of total E) Fcc/bcc difference is 0.02 Ry (0.001% of total E) Mixing energies are also order 10-6 fraction of total E How can we ever get physical behavior correct ? Large cancellation of errors ! 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Elemental Crystal Structures: GGA pseudopotential experimentally found to be fcc method H experimentally found to be bcc -0.12 Ebcc - Efcc Li Be 0.13 0.11 0.11 2.19 0.04 0.50 Na Mg 0.12 0.05 0.05 1.37 0.50 0.50 K Ca Sc Ti V Cr Mn Fe Co Ni Cu 0.04 -0.05 -0.05 1.64 1.41 0.93 5.80 -3.02 4.79 0.48 -23.95 -7.50 -15.30 -36.76 -6.13 -9.19 7.41 0.78 1.80 -8.45 -7.97 8.36 1.71 4.20 9.23 7.99 7.49 2.84 4.02 4.02 (kJ/mole) B C N O F 34.77 -19.71 -21.12 10.24 -4.53 -6.00 Al Si P S Cl 9.21 10.08 10.08 -1.89 -4.00 -4.00 -16.04 7.95 -17.65 -4.46 Zn Ga Ge As Se Br 5.94 -0.08 6.03 1.48 0.70 0.70 0.70 -1.90 -1.90 -10.71 -14.67 -2.85 VASP-PAW SGTE data Saunders et al. Rb Sr Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn Sb Te I 0.08 -0.20 -0.20 0.43 1.33 0.75 10.02 1.19 3.61 -0.29 -31.20 -13.50 -22.00 -38.74 -15.20 -28.00 19.04 8.00 8.00 48.93 9.00 14.00 32.39 19.00 19.00 3.74 10.50 10.50 2.27 3.40 3.40 4.90 1.02 0.64 0.65 0.99 -1.11 0.25 -8.96 -11.19 -1.26 Po At Cs Ba Hf Ta W Re Os Ir Pt Au Hg Tl Pb Bi 0.10 -0.50 -0.50 -1.62 -1.80 -1.80 10.14 2.38 -4.14 -23.75 -16.00 -26.50 -45.03 -19.30 -33.00 24.87 6.00 18.20 70.92 14.50 30.50 59.39 32.00 32.00 7.85 15.00 15.00 1.90 4.25 4.25 -1.02 -1.40 -0.09 0.07 4.06 2.40 2.40 -4.53 1.40 Fr Ra La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er 12.22 22.40 11.55 11.99 12.55 12.88 -1.61 13.11 12.97 12.73 12.36 11.86 Tm Yb Am Cm Bk Cf Es Fm Md No Ac Th Pa U Np Pu 12.56 13.95 17.09 -10.36 -23.17 11.73 Lu 9.91 Lr data taken from: Y. Wang,*a S. Curtarolo, et al. Ab Initio Lattice Stability in Comparison with CALPHAD Lattice Stability, Computer Coupling of Phase Diagrams and Thermochemistry (Calphad) Vol. 28, Issue 1, March 2004, Pages 79-90. 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Elemental Crystal Structures: GGA pseudopotential experimentally found to be hcp method H -0.01 experimentally found to be fcc Ehcp - Efcc Li Be 0.19 -0.05 -0.05 -7.91 -6.35 -6.35 Na Mg 0.06 -0.05 -0.05 -1.22 -2.60 -2.60 K Ca Sc Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As Se Br 0.26 0.31 0.50 0.50 -4.48 -5.00 -5.51 -6.00 -6.00 0.53 -3.50 --4.80 0.91 -2.85 -1.82 -3.01 -1.00 -1.00 -7.76 -2.24 -1.95 -0.43 -0.43 2.22 2.89 1.50 0.52 0.60 0.60 -0.79 -2.97 0.69 0.70 0.70 -0.28 -1.00 -1.00 -4.83 -35.43 3.00 0.00 (kJ/mole) B C N O F -78.73 -6.18 -34.15 1.00 -14.64 -3.00 VASP-PAW SGTE data Saunders et al. Al Si P S Cl 2.85 5.48 5.48 -3.26 -1.80 -1.80 -3.77 -43.63 -16.81 Rb Sr Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn Sb Te I -0.01 0.38 0.25 0.25 -2.13 -6.00 -3.69 -7.60 -7.60 -3.08 -3.50 -5.00 1.14 -3.65 -5.00 -6.53 -10.00 -10.00 -10.79 -12.50 -12.50 3.26 3.00 3.00 2.50 2.00 2.00 0.29 0.30 0.30 -1.00 -0.89 0.35 0.37 0.65 -0.50 -1.61 -0.25 -3.94 23.40 0.99 0.00 Cs Ba Hf Ta W Re Os Ir Pt Au Hg Tl Pb Bi -0.06 -0.40 0.20 0.20 -6.82 -10.00 -10.00 3.06 -4.00 -6.50 -1.79 -4.55 -6.00 -6.26 -11.00 -11.00 -13.26 -13.00 -13.00 6.55 4.00 4.00 5.02 2.50 2.50 0.08 0.24 0.55 -1.51 -2.07 -1.80 -0.31 -0.31 1.80 0.30 0.30 -4.03 Po At 0.00 Fr Ra La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er 2.63 8.50 2.07 1.94 1.77 1.53 0.24 0.77 0.24 -0.41 -1.18 -1.97 Tm Yb -3.85 Ac Th Pa U Np Pu 0.93 4.00 0.49 -15.79 -14.01 0.69 Am Cm Bk Cf Es Fm Md No Lr Lu data taken from: Y. Wang,*a S. Curtarolo, et al. Ab Initio Lattice Stability in Comparison with CALPHAD Lattice Stability, Computer Coupling of Phase Diagram and Thermochemistry (Calphad) Vol. 28, Issue 1, March 2004, Pages 79-90. 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Summary: Comparing Energy of Structures For most elements, both LDA and GGA predict the correct structure for a material (as far as we know) Notable exceptions: Fe in LDA; materials with substantial electron correlation effects (e.g. Pu) High Throughput studies are now possible. 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Metal Hydrides Hydride formation energy: M + H2 -> MH2 LDA Calculation Experiment Correct structure ! 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Spin Polarization can be Important (in particular for transition metal compounds Pmmn: structure of LiMnO2 300 250 C2/m 200 Pmmn 150 C2/m: other common structure meV 100 50 0 -50 -100 -150 -200 NSP Ferro A-Ferro Spin polarized calculations 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Orbital filling depends on spin polarization 5d orbitals 3/3/05 Mn3+ (high spin) 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Mn3+ (no spin) Reaction Energies Reaction Exp (eV) Li (bcc) + Al (fcc) -> LiAl (compound) 0.5 Cu(fcc) + 0.5 Au(fcc) -> CuAu Li(bcc) + CoO2 -> LiCoO2 3/3/05 -0.2457 -0.053 -4.25 LDA (eV) -0.2234 -0.0193 -3.75 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari ∆ 9.08% 63.58% 11.76% Redox Reactions can be more Problematic Exp 3.5 eV FePO4 + Li -> LiFePO4 GGA 2.8 eV MnO2 + Li -> LiMn2O4 3.6 eV 4.1 eV 3.3 eV 4.6 eV V2(PO4)3 + Li -> LiV2(PO4)3 All these reactions involve the transfer of an electron from a delocalized state in Li metal to a localized state in the transition metal oxide (phosphate) 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Structure of LiV2(PO4)3 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari In standard DFT an electron interacts with the effective potential generated by all the electrons (including itself) H= ∑H i i ρ(r) Metal = Ne Ne ∑∇ + ∑V i =1 2 i i =1 nuclear Ne (ri ) + ∑ Veffective (ri ) i =1 Small self-interaction error Self interaction in DFT is key problem in transition metal oxides r ρ(r) TM Oxide 3/3/05 Large self-interaction error r 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari Li insertion into cathode transfers electron from a delocalized to a localized state ρ(r) Metal Li metal Less selfinteraction r Li+ e- electron is transferred from delocalized state to localized state ρ(r) M4+(O2-)2 TM Oxide More selfinteraction Li+ M3+(O2-)2 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari r Summary (LDA) Lattice constants: 1-3% too small Cohesive Energies: 5-20% too strongly bound Bulk Modulus: 5-20% (largest errors for late TM) Bandgaps: too small GGA gives better cohesive energies. Effect on lattice parameters is more random. GGA important for magnetic systems. 3/3/05 3.320 Atomistic Modeling of Materials G. Ceder and N Marzari