Density-functional theory in strong magnetic fields Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, E. Sagvolden) Center for Theoretical and Computational Chemistry, Department of Chemistry, University of Oslo, Norway CMA-CTCC workshop on computational quantum mechanics, December 19-20, 2012 DFT in strong fields. . . 2/35 Outline 1 Review of standard DFT DFT and the Hohenberg–Kohn theorem Constrained-search formulation of DFT Lieb’s formulation of DFT 2 Magnetic fields and DFT BDFT Paramagnetic CDFT 3 Practical use of the Lieb formulation The Adiabatic Connection Some initial BDFT calculations 4 Vorticity and approximate CDFT functionals The VRG functional(s) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 2/35 Overview 1 Review of standard DFT DFT and the Hohenberg–Kohn theorem Constrained-search formulation of DFT Lieb’s formulation of DFT 2 Magnetic fields and DFT BDFT Paramagnetic CDFT 3 Practical use of the Lieb formulation The Adiabatic Connection Some initial BDFT calculations 4 Vorticity and approximate CDFT functionals The VRG functional(s) DFT in strong fields. . . 3/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK mapping Restrict attention to N-particle Hamiltonians of the form H[v ] = 12 p 2 + W + v (1) P where W = i<j rij−1 is fixed and v is an arbitrary one-body potential. There exists a one-to-one mapping between potentials and non-degenerate ground states1 AN = {ρ|ρ comes from g.s. ψ of some H[v ]} VN = {v |H[v ] has a non-deg. g.s.} Bijection between VN and AN ρ ∈ AN determines v and H[v ], which in turn determines ψ and all properties Restriction to non-deg. g.s. is not essential 1 P. Hohenberg and W. Kohn, Phys. Rev. 136:B864 (1964) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 3/35 DFT in strong fields. . . 3/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK mapping Restrict attention to N-particle Hamiltonians of the form H[v ] = 12 p 2 + W + v (1) P where W = i<j rij−1 is fixed and v is an arbitrary one-body potential. There exists a one-to-one mapping between potentials and non-degenerate ground states1 AN = {ρ|ρ comes from g.s. ψ of some H[v ]} VN = {v |H[v ] has a non-deg. g.s.} Bijection between VN and AN ρ ∈ AN determines v and H[v ], which in turn determines ψ and all properties Restriction to non-deg. g.s. is not essential 1 P. Hohenberg and W. Kohn, Phys. Rev. 136:B864 (1964) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 3/35 DFT in strong fields. . . 3/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK mapping Restrict attention to N-particle Hamiltonians of the form H[v ] = 12 p 2 + W + v (1) P where W = i<j rij−1 is fixed and v is an arbitrary one-body potential. There exists a one-to-one mapping between potentials and non-degenerate ground states1 AN = {ρ|ρ comes from g.s. ψ of some H[v ]} VN = {v |H[v ] has a non-deg. g.s.} Bijection between VN and AN ρ ∈ AN determines v and H[v ], which in turn determines ψ and all properties Restriction to non-deg. g.s. is not essential 1 P. Hohenberg and W. Kohn, Phys. Rev. 136:B864 (1964) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 3/35 DFT in strong fields. . . 3/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK mapping Restrict attention to N-particle Hamiltonians of the form H[v ] = 12 p 2 + W + v (1) P where W = i<j rij−1 is fixed and v is an arbitrary one-body potential. There exists a one-to-one mapping between potentials and non-degenerate ground states1 AN = {ρ|ρ comes from g.s. ψ of some H[v ]} VN = {v |H[v ] has a non-deg. g.s.} Bijection between VN and AN ρ ∈ AN determines v and H[v ], which in turn determines ψ and all properties Restriction to non-deg. g.s. is not essential 1 P. Hohenberg and W. Kohn, Phys. Rev. 136:B864 (1964) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 3/35 DFT in strong fields. . . 3/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK mapping Restrict attention to N-particle Hamiltonians of the form H[v ] = 12 p 2 + W + v (1) P where W = i<j rij−1 is fixed and v is an arbitrary one-body potential. There exists a one-to-one mapping between potentials and non-degenerate ground states1 AN = {ρ|ρ comes from g.s. ψ of some H[v ]} VN = {v |H[v ] has a non-deg. g.s.} Bijection between VN and AN ρ ∈ AN determines v and H[v ], which in turn determines ψ and all properties Restriction to non-deg. g.s. is not essential 1 P. Hohenberg and W. Kohn, Phys. Rev. 136:B864 (1964) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 3/35 DFT in strong fields. . . 4/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK functional Let E [v ] be the energy of the g.s. of H[v ] E : VN → R Denote the direct interaction between potential and density by Z (ρ|v ) = ρ(r) v (r) dr (2) Define the intrinsic energy as FHK [ρ? ] = E [v ? ] − (ρ? |v ? ) = hψ ? | 12 p 2 + W |ψ ? i where ρ? determines the potential v ? FHK : AN → R FHK is universal, i.e. independent of v Problem: in general not easy to tell whether ρ ∈ AN or ρ∈ / AN for a given ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 4/35 (3) DFT in strong fields. . . 4/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK functional Let E [v ] be the energy of the g.s. of H[v ] E : VN → R Denote the direct interaction between potential and density by Z (ρ|v ) = ρ(r) v (r) dr (2) Define the intrinsic energy as FHK [ρ? ] = E [v ? ] − (ρ? |v ? ) = hψ ? | 12 p 2 + W |ψ ? i where ρ? determines the potential v ? FHK : AN → R FHK is universal, i.e. independent of v Problem: in general not easy to tell whether ρ ∈ AN or ρ∈ / AN for a given ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 4/35 (3) DFT in strong fields. . . 4/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK functional Let E [v ] be the energy of the g.s. of H[v ] E : VN → R Denote the direct interaction between potential and density by Z (ρ|v ) = ρ(r) v (r) dr (2) Define the intrinsic energy as FHK [ρ? ] = E [v ? ] − (ρ? |v ? ) = hψ ? | 12 p 2 + W |ψ ? i where ρ? determines the potential v ? FHK : AN → R FHK is universal, i.e. independent of v Problem: in general not easy to tell whether ρ ∈ AN or ρ∈ / AN for a given ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 4/35 (3) DFT in strong fields. . . 4/35 Review of standard DFT DFT and the Hohenberg–Kohn theorem The HK functional Let E [v ] be the energy of the g.s. of H[v ] E : VN → R Denote the direct interaction between potential and density by Z (ρ|v ) = ρ(r) v (r) dr (2) Define the intrinsic energy as FHK [ρ? ] = E [v ? ] − (ρ? |v ? ) = hψ ? | 12 p 2 + W |ψ ? i where ρ? determines the potential v ? FHK : AN → R FHK is universal, i.e. independent of v Problem: in general not easy to tell whether ρ ∈ AN or ρ∈ / AN for a given ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 4/35 (3) DFT in strong fields. . . 5/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search Let IN = {ρ|ρ comes from valid wave fun. ψ} The constrained-search2 is a very transparent formulation of DFT: E [v ] = inf hψ|H[v ]|ψi = inf hψ| 12 p 2 + W + v |ψi ψ ψ 1 2 = inf (ρ|v ) + inf hψ| 2 p + W |ψi (4) ψ7→ρ ρ∈IN Universal intrinsic energy: Fcs [ρ] = inf hψ| 12 p 2 + W |ψi (5) ψ7→ρ Fcs : IN → R; arbitrary potentials can be allowed Easy generalization: replace pure states ψ with mixed states above 2 M. Levy Proc. Natl. Acad. Sci. USA 76:6062 (1979) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 5/35 DFT in strong fields. . . 5/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search Let IN = {ρ|ρ comes from valid wave fun. ψ} The constrained-search2 is a very transparent formulation of DFT: E [v ] = inf hψ|H[v ]|ψi = inf hψ| 12 p 2 + W + v |ψi ψ ψ 1 2 = inf (ρ|v ) + inf hψ| 2 p + W |ψi (4) ψ7→ρ ρ∈IN Universal intrinsic energy: Fcs [ρ] = inf hψ| 12 p 2 + W |ψi (5) ψ7→ρ Fcs : IN → R; arbitrary potentials can be allowed Easy generalization: replace pure states ψ with mixed states above 2 M. Levy Proc. Natl. Acad. Sci. USA 76:6062 (1979) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 5/35 DFT in strong fields. . . 5/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search Let IN = {ρ|ρ comes from valid wave fun. ψ} The constrained-search2 is a very transparent formulation of DFT: E [v ] = inf hψ|H[v ]|ψi = inf hψ| 12 p 2 + W + v |ψi ψ ψ 1 2 = inf (ρ|v ) + inf hψ| 2 p + W |ψi (4) ψ7→ρ ρ∈IN Universal intrinsic energy: Fcs [ρ] = inf hψ| 12 p 2 + W |ψi (5) ψ7→ρ Fcs : IN → R; arbitrary potentials can be allowed Easy generalization: replace pure states ψ with mixed states above 2 M. Levy Proc. Natl. Acad. Sci. USA 76:6062 (1979) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 5/35 DFT in strong fields. . . 5/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search Let IN = {ρ|ρ comes from valid wave fun. ψ} The constrained-search2 is a very transparent formulation of DFT: E [v ] = inf hψ|H[v ]|ψi = inf hψ| 12 p 2 + W + v |ψi ψ ψ 1 2 = inf (ρ|v ) + inf hψ| 2 p + W |ψi (4) ψ7→ρ ρ∈IN Universal intrinsic energy: Fcs [ρ] = inf hψ| 12 p 2 + W |ψi (5) ψ7→ρ Fcs : IN → R; arbitrary potentials can be allowed Easy generalization: replace pure states ψ with mixed states above 2 M. Levy Proc. Natl. Acad. Sci. USA 76:6062 (1979) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 5/35 DFT in strong fields. . . 6/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search The HK formulation shows that ρ ∈ AN determines all properties. The constrained-search formulation shows that ρ ∈ IN determines the intrinsic energy. ρ ∈ IN is easier to work with than ρ ∈ AN Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 6/35 DFT in strong fields. . . 6/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search The HK formulation shows that ρ ∈ AN determines all properties. The constrained-search formulation shows that ρ ∈ IN determines the intrinsic energy. ρ ∈ IN is easier to work with than ρ ∈ AN Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 6/35 DFT in strong fields. . . 6/35 Review of standard DFT Constrained-search formulation of DFT Constrained-search The HK formulation shows that ρ ∈ AN determines all properties. The constrained-search formulation shows that ρ ∈ IN determines the intrinsic energy. ρ ∈ IN is easier to work with than ρ ∈ AN Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 6/35 DFT in strong fields. . . 7/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT A function is convex if linear interpolation is always an overestimate, F [µρ1 + (1 − µ)ρ2 ] ≤ µF [ρ1 ] + (1 − µ)F [ρ2 ] (6) for 0 ≤ µ ≤ 1. A function is concave if linear interpolation is always an underestimate, E [µv1 + (1 − µ)v2 ] ≤ µE [v1 ] + (1 − µ)E [v2 ] Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 7/35 (7) DFT in strong fields. . . 7/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT A function is convex if linear interpolation is always an overestimate, F [µρ1 + (1 − µ)ρ2 ] ≤ µF [ρ1 ] + (1 − µ)F [ρ2 ] (6) for 0 ≤ µ ≤ 1. A function is concave if linear interpolation is always an underestimate, E [µv1 + (1 − µ)v2 ] ≤ µE [v1 ] + (1 − µ)E [v2 ] Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 7/35 (7) Convexity and concavity 8 6 4 convex neither concave y 2 0 −2 −4 −6 −3 −2 −1 0 x 1 2 3 DFT in strong fields. . . 9/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT E [v ] is a concave functional Fcs [ρ] (mixed state-version) is a convex functional Results from convex analysis are then available.3 Legendre–Fenchel transformations:4 E [v ] = inf ((ρ|v ) + FL [ρ]) (8) FL [ρ] = sup (E [v ] − (ρ|v )) (9) ρ v ρ is now an element of a Banach space, v is an element of the dual space; for very unphysical densities, FL [ρ] = +∞ Lieb suggested: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) and v ∈ L3/2 (R3 ) + L∞ (R3 ); other choices conceivable too 3 4 E. H. Lieb Int. J. Quantum Chem. 24:243 (1983) Technical point: lower/upper semicontinuity Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 9/35 DFT in strong fields. . . 9/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT E [v ] is a concave functional Fcs [ρ] (mixed state-version) is a convex functional Results from convex analysis are then available.3 Legendre–Fenchel transformations:4 E [v ] = inf ((ρ|v ) + FL [ρ]) (8) FL [ρ] = sup (E [v ] − (ρ|v )) (9) ρ v ρ is now an element of a Banach space, v is an element of the dual space; for very unphysical densities, FL [ρ] = +∞ Lieb suggested: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) and v ∈ L3/2 (R3 ) + L∞ (R3 ); other choices conceivable too 3 4 E. H. Lieb Int. J. Quantum Chem. 24:243 (1983) Technical point: lower/upper semicontinuity Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 9/35 DFT in strong fields. . . 9/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT E [v ] is a concave functional Fcs [ρ] (mixed state-version) is a convex functional Results from convex analysis are then available.3 Legendre–Fenchel transformations:4 E [v ] = inf ((ρ|v ) + FL [ρ]) (8) FL [ρ] = sup (E [v ] − (ρ|v )) (9) ρ v ρ is now an element of a Banach space, v is an element of the dual space; for very unphysical densities, FL [ρ] = +∞ Lieb suggested: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) and v ∈ L3/2 (R3 ) + L∞ (R3 ); other choices conceivable too 3 4 E. H. Lieb Int. J. Quantum Chem. 24:243 (1983) Technical point: lower/upper semicontinuity Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 9/35 DFT in strong fields. . . 9/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT E [v ] is a concave functional Fcs [ρ] (mixed state-version) is a convex functional Results from convex analysis are then available.3 Legendre–Fenchel transformations:4 E [v ] = inf ((ρ|v ) + FL [ρ]) (8) FL [ρ] = sup (E [v ] − (ρ|v )) (9) ρ v ρ is now an element of a Banach space, v is an element of the dual space; for very unphysical densities, FL [ρ] = +∞ Lieb suggested: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) and v ∈ L3/2 (R3 ) + L∞ (R3 ); other choices conceivable too 3 4 E. H. Lieb Int. J. Quantum Chem. 24:243 (1983) Technical point: lower/upper semicontinuity Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 9/35 DFT in strong fields. . . 9/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT E [v ] is a concave functional Fcs [ρ] (mixed state-version) is a convex functional Results from convex analysis are then available.3 Legendre–Fenchel transformations:4 E [v ] = inf ((ρ|v ) + FL [ρ]) (8) FL [ρ] = sup (E [v ] − (ρ|v )) (9) ρ v ρ is now an element of a Banach space, v is an element of the dual space; for very unphysical densities, FL [ρ] = +∞ Lieb suggested: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) and v ∈ L3/2 (R3 ) + L∞ (R3 ); other choices conceivable too 3 4 E. H. Lieb Int. J. Quantum Chem. 24:243 (1983) Technical point: lower/upper semicontinuity Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 9/35 DFT in strong fields. . . 10/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT Usefulness of convex formulation: Solutions of convex optimization problems are either unique or form a convex set Compare restrictions on ρ: HK: ρ is g.s. density of some v , i.e. ρ ∈ AN CS: ρ comes from a valid pure/mixed state, i.e. ρ ∈ IN Lieb: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) Would like to identify functional derivatives: v = −δF [ρ]/δρ, ρ = δE [v ]/δv Differentiability of FHK [ρ]? Differentiability Fcs [ρ]? In a convex formulation, differentiability may be set aside. Instead, the notion of sub-gradients and sub-differentials of a convex functional is available. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 10/35 DFT in strong fields. . . 10/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT Usefulness of convex formulation: Solutions of convex optimization problems are either unique or form a convex set Compare restrictions on ρ: HK: ρ is g.s. density of some v , i.e. ρ ∈ AN CS: ρ comes from a valid pure/mixed state, i.e. ρ ∈ IN Lieb: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) Would like to identify functional derivatives: v = −δF [ρ]/δρ, ρ = δE [v ]/δv Differentiability of FHK [ρ]? Differentiability Fcs [ρ]? In a convex formulation, differentiability may be set aside. Instead, the notion of sub-gradients and sub-differentials of a convex functional is available. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 10/35 DFT in strong fields. . . 10/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT Usefulness of convex formulation: Solutions of convex optimization problems are either unique or form a convex set Compare restrictions on ρ: HK: ρ is g.s. density of some v , i.e. ρ ∈ AN CS: ρ comes from a valid pure/mixed state, i.e. ρ ∈ IN Lieb: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) Would like to identify functional derivatives: v = −δF [ρ]/δρ, ρ = δE [v ]/δv Differentiability of FHK [ρ]? Differentiability Fcs [ρ]? In a convex formulation, differentiability may be set aside. Instead, the notion of sub-gradients and sub-differentials of a convex functional is available. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 10/35 DFT in strong fields. . . 10/35 Review of standard DFT Lieb’s formulation of DFT Convexity and DFT Usefulness of convex formulation: Solutions of convex optimization problems are either unique or form a convex set Compare restrictions on ρ: HK: ρ is g.s. density of some v , i.e. ρ ∈ AN CS: ρ comes from a valid pure/mixed state, i.e. ρ ∈ IN Lieb: ρ ∈ L1 (R3 ) ∩ L3 (R3 ) Would like to identify functional derivatives: v = −δF [ρ]/δρ, ρ = δE [v ]/δv Differentiability of FHK [ρ]? Differentiability Fcs [ρ]? In a convex formulation, differentiability may be set aside. Instead, the notion of sub-gradients and sub-differentials of a convex functional is available. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 10/35 Overview 1 Review of standard DFT DFT and the Hohenberg–Kohn theorem Constrained-search formulation of DFT Lieb’s formulation of DFT 2 Magnetic fields and DFT BDFT Paramagnetic CDFT 3 Practical use of the Lieb formulation The Adiabatic Connection Some initial BDFT calculations 4 Vorticity and approximate CDFT functionals The VRG functional(s) DFT in strong fields. . . 11/35 Magnetic fields and DFT Generalizations of DFT When generalizing DFT, one has to choose which properties to retain: framework DFT TD-DFT SDFT Paramagnetic CDFT Physical CDFT dens.-pot. map yes yes no no maybe CS yes no? yes yes no (not universal) convex yes no? yes yes no Blue answers are recent results.5 Other aspects: TD-DFT derivation from action functional, non-zero temperature, partial Legendre–Fenchel transforms, . . . 5 Tellgren, Kvaal, Sagvolden, Ekström, Teale & Helgaker, Phys. Rev. A 86:062506 (2012) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 11/35 DFT in strong fields. . . 12/35 Magnetic fields and DFT BDFT Magnetic-field DFT (BDFT) Consider now Hamiltonians H[v , A] = 12 (p + A)2 + v + W (10) The Hamiltonians under consideration now differ by more than a multiplicative potential. One approach6 : fix a gauge A = A[B], imagine a family of Density Functional Theories parametrized by the magnetic field B For each particular choice of B, one obtains a HK-type mapping, a constrained-search, and even a convex formulation involving only ρ and v . E [v ; B] is concave in v F [ρ; B] is convex in ρ, “half-universal” (i.e. independent of v but not B and A) 6 C. J. Grayce and R. A. Harris, Phys. Rev. A 50:3089 (1994) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 12/35 DFT in strong fields. . . 13/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic Current-Density Functional Theory Energy of trial ψ: hψ|H[v , A]|ψi = hψ| 12 p 2 + 21 {p, A} + 21 A2 + v + W |ψi = (ρψ |v + 21 A2 ) + (jp;ψ |A) + hψ| 12 p 2 + W |ψi Paramagnetic and physical current density: jp (r0 ) = hψ| 21 {−i∇1 , δ(r0 − r1 )}|ψi (11) j(r0 ) = hψ| 12 {−i∇1 + A, δ(r0 − r1 )}|ψi (12) jp is (a) gauge-dependent and (b) determined by ψ j is (a) gauge-invariant and (b) not determined by ψ alone Paramagnetic CDFT uses (ρ, jp ) as the density variables7 7 G. Vignale and M. Rasolt, Phys. Rev. Lett. 59:2360 (1987) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 13/35 DFT in strong fields. . . 14/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic Current-Density Functional Theory Constrained-search remains straightforward in Vignale & Rasolt’s formulation: E [v , A] = inf hψ|H[v , A]|ψi ψ = inf hψ| 12 p 2 + 12 {p, A} + 12 A2 + v + W |ψi ψ 1 2 1 2 = inf (ρ|v + 2 A ) + (jp |A) + inf hψ| 2 p + W |ψi ψ7→ρ,jp ρ,jp The intrinsic energy is universal (independent of (v , A)): FVR [ρ, jp ] = inf hψ| 12 p 2 + W |ψi ψ7→ρ,jp Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 14/35 (13) DFT in strong fields. . . 15/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic CDFT and convexity FVR [ρ, jp ] (mixed state version) is convex in (ρ, jp ) E [v , A] is not concave densities and potentials are not paired linearly → not the form of a Legendre–Fenchel transformation Solution: a change of variables to A0 = A (14) u = v + 12 A2 (15) Ē [u, A] = E [v + 12 A2 , A] (16) The resulting functional is concave in (u, A) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 15/35 DFT in strong fields. . . 15/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic CDFT and convexity FVR [ρ, jp ] (mixed state version) is convex in (ρ, jp ) E [v , A] is not concave densities and potentials are not paired linearly → not the form of a Legendre–Fenchel transformation Solution: a change of variables to A0 = A (14) u = v + 12 A2 (15) Ē [u, A] = E [v + 12 A2 , A] (16) The resulting functional is concave in (u, A) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 15/35 DFT in strong fields. . . 15/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic CDFT and convexity FVR [ρ, jp ] (mixed state version) is convex in (ρ, jp ) E [v , A] is not concave densities and potentials are not paired linearly → not the form of a Legendre–Fenchel transformation Solution: a change of variables to A0 = A (14) u = v + 12 A2 (15) Ē [u, A] = E [v + 12 A2 , A] (16) The resulting functional is concave in (u, A) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 15/35 DFT in strong fields. . . 15/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic CDFT and convexity FVR [ρ, jp ] (mixed state version) is convex in (ρ, jp ) E [v , A] is not concave densities and potentials are not paired linearly → not the form of a Legendre–Fenchel transformation Solution: a change of variables to A0 = A (14) u = v + 12 A2 (15) Ē [u, A] = E [v + 12 A2 , A] (16) The resulting functional is concave in (u, A) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 15/35 DFT in strong fields. . . 16/35 Magnetic fields and DFT Paramagnetic CDFT Paramagnetic CDFT and convexity Legendre–Fenchel transformations: Ē [u, A] = inf ((ρ|u) + (jp |A) + FVR [ρ, jp ]) , ρ,jp FVR [ρ, jp ] = sup Ē [u, A] − (ρ|u) − (jp |A) (17) (18) u,A Care is required when identifying dual Banach spaces for densities and potentials. Gauge transformations can make A grow arbitrarily fast as r → ∞, (jp |A) may be lowered without bound through gauge transformations, etc. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 16/35 Overview 1 Review of standard DFT DFT and the Hohenberg–Kohn theorem Constrained-search formulation of DFT Lieb’s formulation of DFT 2 Magnetic fields and DFT BDFT Paramagnetic CDFT 3 Practical use of the Lieb formulation The Adiabatic Connection Some initial BDFT calculations 4 Vorticity and approximate CDFT functionals The VRG functional(s) DFT in strong fields. . . 17/35 Practical use of the Lieb formulation The Adiabatic Connection AC Consider Hamiltonians with scaled electron repulsion, H[v , A] = 12 (p + A)2 + v + λW , (19) with 0 ≤ λ ≤ 1. Can define a family of concave functionals Ēλ [u, A] . . . . . . and convex functionals FVR,λ [ρ, jp ] λ = 1 (full electron repulsion) is the case of actual interest λ = 0 (no repulsion) corresponds to the fictional Kohn–Sham system Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 17/35 DFT in strong fields. . . 17/35 Practical use of the Lieb formulation The Adiabatic Connection AC Consider Hamiltonians with scaled electron repulsion, H[v , A] = 12 (p + A)2 + v + λW , (19) with 0 ≤ λ ≤ 1. Can define a family of concave functionals Ēλ [u, A] . . . . . . and convex functionals FVR,λ [ρ, jp ] λ = 1 (full electron repulsion) is the case of actual interest λ = 0 (no repulsion) corresponds to the fictional Kohn–Sham system Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 17/35 DFT in strong fields. . . 17/35 Practical use of the Lieb formulation The Adiabatic Connection AC Consider Hamiltonians with scaled electron repulsion, H[v , A] = 12 (p + A)2 + v + λW , (19) with 0 ≤ λ ≤ 1. Can define a family of concave functionals Ēλ [u, A] . . . . . . and convex functionals FVR,λ [ρ, jp ] λ = 1 (full electron repulsion) is the case of actual interest λ = 0 (no repulsion) corresponds to the fictional Kohn–Sham system Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 17/35 DFT in strong fields. . . 17/35 Practical use of the Lieb formulation The Adiabatic Connection AC Consider Hamiltonians with scaled electron repulsion, H[v , A] = 12 (p + A)2 + v + λW , (19) with 0 ≤ λ ≤ 1. Can define a family of concave functionals Ēλ [u, A] . . . . . . and convex functionals FVR,λ [ρ, jp ] λ = 1 (full electron repulsion) is the case of actual interest λ = 0 (no repulsion) corresponds to the fictional Kohn–Sham system Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 17/35 DFT in strong fields. . . 18/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection FVR,0 [ρ, jp ] = inf hψ| 12 p 2 |ψi (20) ψ7→ρ,jp is essentially the Kohn–Sham kinetic energy8 , denoted Ts [ρ, jp ] Remaining intrinsic energy of actual (λ = 1) system: Z 1 d FVR,1 [ρ, jp ] − FVR,0 [ρ, jp ] = FVR,λ [ρ, jp ] dλ 0 dλ (21) Z 1 Z 1 λ λ = hψρ,jp |W |ψρ,jp idλ = W (λ)dλ 0 0 λ minimizes hψ| 1 p 2 + λW |ψi subject to ψ 7→ ρ, j where ψρ,j p 2 p When analyzing and constructing approximate functionals, it is useful to model the integrand W (λ) 8 If the Fermi level is non-degenerate, there is no distinction. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 18/35 DFT in strong fields. . . 18/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection FVR,0 [ρ, jp ] = inf hψ| 12 p 2 |ψi (20) ψ7→ρ,jp is essentially the Kohn–Sham kinetic energy8 , denoted Ts [ρ, jp ] Remaining intrinsic energy of actual (λ = 1) system: Z 1 d FVR,1 [ρ, jp ] − FVR,0 [ρ, jp ] = FVR,λ [ρ, jp ] dλ 0 dλ (21) Z 1 Z 1 λ λ = hψρ,jp |W |ψρ,jp idλ = W (λ)dλ 0 0 λ minimizes hψ| 1 p 2 + λW |ψi subject to ψ 7→ ρ, j where ψρ,j p 2 p When analyzing and constructing approximate functionals, it is useful to model the integrand W (λ) 8 If the Fermi level is non-degenerate, there is no distinction. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 18/35 DFT in strong fields. . . 19/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The constraint ψ 7→ ρ, jp is difficult to enforce directly in a computer implementation. A Lieb-type convex formulation allows an alternative treatment. For each λ, one may obtain an approximation to FVR,λ [ρ, jp ] = sup Ēλ [u, A] − (ρ|u) − (jp |A) (22) u,A The exact energy Ēλ [u, A] is replaced by, say, a FCI energy evaluated in some finite orbital basis The optimization over all (u, A) in a large space is replaced by optimization over a space spanned by practically convenient basis functions. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 19/35 DFT in strong fields. . . 19/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The constraint ψ 7→ ρ, jp is difficult to enforce directly in a computer implementation. A Lieb-type convex formulation allows an alternative treatment. For each λ, one may obtain an approximation to FVR,λ [ρ, jp ] = sup Ēλ [u, A] − (ρ|u) − (jp |A) (22) u,A The exact energy Ēλ [u, A] is replaced by, say, a FCI energy evaluated in some finite orbital basis The optimization over all (u, A) in a large space is replaced by optimization over a space spanned by practically convenient basis functions. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 19/35 DFT in strong fields. . . 19/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The constraint ψ 7→ ρ, jp is difficult to enforce directly in a computer implementation. A Lieb-type convex formulation allows an alternative treatment. For each λ, one may obtain an approximation to FVR,λ [ρ, jp ] = sup Ēλ [u, A] − (ρ|u) − (jp |A) (22) u,A The exact energy Ēλ [u, A] is replaced by, say, a FCI energy evaluated in some finite orbital basis The optimization over all (u, A) in a large space is replaced by optimization over a space spanned by practically convenient basis functions. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 19/35 DFT in strong fields. . . 19/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The constraint ψ 7→ ρ, jp is difficult to enforce directly in a computer implementation. A Lieb-type convex formulation allows an alternative treatment. For each λ, one may obtain an approximation to FVR,λ [ρ, jp ] = sup Ēλ [u, A] − (ρ|u) − (jp |A) (22) u,A The exact energy Ēλ [u, A] is replaced by, say, a FCI energy evaluated in some finite orbital basis The optimization over all (u, A) in a large space is replaced by optimization over a space spanned by practically convenient basis functions. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 19/35 DFT in strong fields. . . 19/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The constraint ψ 7→ ρ, jp is difficult to enforce directly in a computer implementation. A Lieb-type convex formulation allows an alternative treatment. For each λ, one may obtain an approximation to FVR,λ [ρ, jp ] = sup Ēλ [u, A] − (ρ|u) − (jp |A) (22) u,A The exact energy Ēλ [u, A] is replaced by, say, a FCI energy evaluated in some finite orbital basis The optimization over all (u, A) in a large space is replaced by optimization over a space spanned by practically convenient basis functions. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 19/35 DFT in strong fields. . . 20/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The resulting implementation is computationally demanding, but theoretically very well founded as it becomes an instance of a convex optimization problem Previous work by A. Teale et al. in CTCC have compared AC curves from high-quality ab inition calculations against AC curves corresponding to approximate DFT functionals We now wish to study BDFT and CDFT in the same way. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 20/35 DFT in strong fields. . . 20/35 Practical use of the Lieb formulation The Adiabatic Connection The Adiabatic Connection The resulting implementation is computationally demanding, but theoretically very well founded as it becomes an instance of a convex optimization problem Previous work by A. Teale et al. in CTCC have compared AC curves from high-quality ab inition calculations against AC curves corresponding to approximate DFT functionals We now wish to study BDFT and CDFT in the same way. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 20/35 He2 molecule 3.8 B=0 B=0.5 3.78 3.76 3.74 W 3.72 3.7 3.68 3.66 3.64 3.62 0 0.1 0.2 0.3 0.4 0.5 lambda 0.6 0.7 0.8 0.9 1 LiH molecule 4.06 B=0 B = 0.1 au B = 0.5 au 4.05 χxc = 1/2 d2 Wxc/dB2 [au] W xc [hartree] 4.04 0.5 4.03 4.02 4.01 4 0.45 0.4 0.35 3.99 0.3 3.98 3.97 0.25 0.1 0.2 0.3 0.4 0.5 λ 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 λ 0.6 0.7 0.8 0.9 Strategy: construct simple analytical model functions Wxc (λ, B) 2W xc and χxc (λ) = ∂∂B , try to relate to practical approximate 2 B=0 density functionals Overview 1 Review of standard DFT DFT and the Hohenberg–Kohn theorem Constrained-search formulation of DFT Lieb’s formulation of DFT 2 Magnetic fields and DFT BDFT Paramagnetic CDFT 3 Practical use of the Lieb formulation The Adiabatic Connection Some initial BDFT calculations 4 Vorticity and approximate CDFT functionals The VRG functional(s) DFT in strong fields. . . 23/35 Vorticity and approximate CDFT functionals Gauge-invariance and jp The paramagnetic current is not gauge-invariant. FVR [ρ, jp ] = Ts [ρ, jp ] + FHartree [ρ] + F xc [ρ, jp ] (23) All gauge-dependence on the RHS should be contained in Ts . Vignale and Rasolt therefore introduced the vorticity ν =∇× jp j − ρA j =∇× =∇× −B ρ ρ ρ (24) Tempting to stipulate that F xc [ρ, jp ] = F xc [ρ, ν] (25) to ensure gauge-invariance. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 23/35 DFT in strong fields. . . 24/35 Vorticity and approximate CDFT functionals CDFT: what is vorticity? In fluid mechanics, vorticity is the curl of the velocity field ∇× j ρv = ∇× = ∇×v (26) ρ ρ vector indicating volume element’s tendency to rotate and its rotation axis In CDFT, vorticity is slightly different ν =∇× jp j − ρA j =∇× =∇× −B ρ ρ ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 24/35 (27) DFT in strong fields. . . 24/35 Vorticity and approximate CDFT functionals CDFT: what is vorticity? In fluid mechanics, vorticity is the curl of the velocity field ∇× j ρv = ∇× = ∇×v (26) ρ ρ vector indicating volume element’s tendency to rotate and its rotation axis In CDFT, vorticity is slightly different ν =∇× jp j − ρA j =∇× =∇× −B ρ ρ ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 24/35 (27) DFT in strong fields. . . 24/35 Vorticity and approximate CDFT functionals CDFT: what is vorticity? In fluid mechanics, vorticity is the curl of the velocity field ∇× j ρv = ∇× = ∇×v (26) ρ ρ vector indicating volume element’s tendency to rotate and its rotation axis In CDFT, vorticity is slightly different ν =∇× jp j − ρA j =∇× =∇× −B ρ ρ ρ Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 24/35 (27) BH molecule, HF/u-aug-cc-pVTZ, field along z (out of the screen) log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 x 2 3 4 5 −3 −4 −4 −0.08 −0.1 −3 −2 −1 0 1 2 3 4 5 log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 2 3 4 5 −3 −4 −4 x −0.08 −0.1 −3 −2 −1 0 1 2 3 4 5 log10(ρ) (in plane), B = 0.10 4 curl(j/ρ) proj. on B (in plane) 1 4 3 3 0.1 0 2 2 0.05 −1 1 y 1 0 −2 0 0 −1 −3 −1 −0.05 −2 −4 −2 −0.1 −3 −3 −5 −4 −4 −3 −2 −1 0 1 x 2 3 4 5 −0.15 −4 −4 −3 −2 −1 0 1 2 3 4 5 log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 2 3 4 5 −3 −4 −4 x −0.08 −0.1 −3 −2 −1 0 1 2 3 4 5 log10(ρ) (in plane), B = 0.15 4 curl(j/ρ) proj. on B (in plane) 1 4 0.15 3 3 0 0.1 2 2 −1 1 0.05 y 1 −2 0 0 0 −1 −3 −0.05 −1 −2 −0.1 −4 −2 −3 −0.15 −3 −5 −4 −4 −3 −2 −1 0 1 x 2 3 4 5 −4 −4 −0.2 −3 −2 −1 0 1 2 3 4 5 log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 2 3 4 5 −3 −4 −4 x −0.08 −0.1 −3 −2 −1 0 1 2 3 4 5 log10(ρ) (in plane), B = 0.20 4 curl(j/ρ) proj. on B (in plane) 1 4 0.2 3 0 3 0.15 2 2 0.1 −1 1 0.05 y 1 −2 0 0 0 −1 −0.05 −3 −1 −0.1 −2 −4 −2 −0.15 −3 −5 −3 −4 −4 −3 −2 −1 0 1 x 2 3 4 5 −4 −4 −0.2 −3 −2 −1 0 1 2 3 4 5 log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 2 3 4 5 −3 −4 −4 x −0.08 −0.1 −3 −2 −1 0 1 2 3 4 5 log10(ρ) (in plane), B = 0.25 4 curl(j/ρ) proj. on B (in plane) 1 4 0.25 3 0 0.2 3 0.15 2 2 −1 0.1 1 1 0.05 y −2 0 0 0 −0.05 −3 −1 −1 −2 −0.1 −4 −0.15 −2 −3 −0.2 −5 −3 −0.25 −4 −4 −3 −2 −1 0 1 x 2 3 4 5 −4 −4 −3 −2 −1 0 1 2 3 4 5 log (ρ) (in plane), B = 0.05 10 4 curl(j/ρ) proj. on B (in plane) 1 4 0.08 3 3 0.06 0 2 2 0.04 1 0.02 −1 y 1 −2 0 0 0 −0.02 −1 −3 −1 −0.04 −2 −4 −2 −0.06 −3 −5 −4 −4 −3 −2 −1 0 1 2 3 4 5 −3 −4 −4 x −0.08 −0.1 −3 −2 −1 log10(ρ) (in plane), B = 0.30 0 1 2 3 4 5 curl(j/ρ) proj. on B (in plane) 4 1 4 3 0 3 0.2 2 2 −1 0.1 1 1 y −2 0 0 0 −3 −1 −0.1 −1 −4 −2 −2 −0.2 −5 −3 −3 −4 −4 −0.3 −6 −3 −2 −1 0 1 x 2 3 4 5 −4 −4 −3 −2 −1 0 1 2 3 4 5 DFT in strong fields. . . 26/35 Vorticity and approximate CDFT functionals The VRG functional(s) VRG functionals Vignale–Rasolt–Geldart9 suggested the functional form Z me kF (ρ) χ xc FVRG [ρ, ν] = − 1 ν 2 dr 48π 2 χ0 (28) where χ is magnetic susceptibility (and χ0 is the susceptibility of a noninteracting electron gas). Taking the susceptibility ratio to be a function of ρ only, one may write Z xc FVRG [ρ, ν] = C g (ρ)ν 2 dr (29) g was fitted to RPA results for a uniform electron gas, rs = (4πρ/3)−1/3 < 10 bohr. 9 G. Vignale et al. Phys. Rev. B 37:2502 (1988) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 26/35 Different fits proposed: LCH, OMC, TP10 1.4 1.2 g 1.0 æ æ æ æ æ æ æ æ æ 0.8 æ 1.01 0.6 g 1.00 0.4 0.99 0.98 0.2 0.0 0 0.97 0.0 æ 0.5 10 1.0 rs 1.5 æ 2.0 æ 20 30 rs 10 40 50 æ A. M. Lee, S. M. Colwell, N. C. Handy Chem. Phys. Lett. 229:225 (1994); E. Orestes, T. Marcasso, K. Capelle Phys. Rev. A 68:022105 (2003); J. Tao and J. P. Perdew Phys. Rev. Lett. 95:196403 (2005) DFT in strong fields. . . 28/35 Vorticity and approximate CDFT functionals The VRG functional(s) VRG functionals All forms of g (ρ) proposed to date have problems. No fitting data for rs > 10 bohr −→ unreliable low-density limit. All forms feature an integrand with the asymptotic behavior ρ1/3 ν 2 (30) for small ρ ∼ 0. Formally, the integral over all space seems well-defined. The integral is evaluated numerically on a grid. Very sensitive to noise. ν is in practice computed as ν =∇× jp ρ∇ × jp − ∇ρ × jp = ρ ρ2 (31) Numerical noise in the numerator gets amplified when ρ ∼ 0. Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 28/35 DFT in strong fields. . . 29/35 Vorticity and approximate CDFT functionals The VRG functional(s) VRG functionals Let η = ρ2 ν = ρ∇ × jp − ∇ρ × jp . Example: LCH form −1/3 xc FVRG-LCH [ρ, ν] = C −1 + e −b0 ρ (1 + b2 ρ−1/3 ) η2 dr 4 + ρ4 (32) Regularization parameter not in the original; LCH used a hard cut-off at rs = 10 bohr, claiming it does not affect results Computed nuclear shielding constants, σαβ = ∂ 2 E /∂µ∂Bµ=B=0 , do depend cut-offs (hard or soft) Cut-off dependence is a “barrier to any really useful results”11 11 W. Zhu and S. B. Trickey J. Chem. Phys. 125:094317 (2006) Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 29/35 NMR Isotropic Shielding Constants [ppm] VRG corrections tend to reduce the shieldings on average – worsening the results Mol HF CO N2 H2 O NH3 CH4 AlF FCN H2 S HFCO LiF LiH ME MAE Nuc H F C O N O H N H C H Al F F C N S H O C F H Li F H Li LDA 29.44 416.89 −20.74 −87.05 −90.13 335.08 30.99 267.55 31.57 193.86 31.25 532.13 141.25 344.69 64.91 91.91 727.14 30.44 −138.75 17.35 84.48 22.94 85.41 341.83 25.76 86.40 −21.16 21.19 LDA-VRG 29.46 416.58 −25.05 −94.03 −95.66 335.27 30.99 268.44 31.57 193.95 31.25 528.29 135.41 344.55 62.87 89.99 726.50 30.44 −146.91 15.27 77.99 22.85 82.28 281.82 25.44 81.70 −25.60 25.62 KT3 30.24 412.22 5.60 −54.52 −59.88 327.84 31.65 263.16 32.08 193.49 31.68 566.43 174.71 337.46 83.17 122.15 724.58 31.05 −89.27 37.06 116.34 23.58 89.88 359.32 26.76 92.19 −9.15 10.25 KT3-VRG 30.26 411.86 1.80 −60.92 −64.91 328.02 31.65 264.22 32.08 193.62 31.68 562.82 169.17 337.19 81.27 120.63 723.91 31.05 −96.83 35.17 110.02 23.49 85.67 295.89 26.54 89.58 −13.54 13.91 CCSD(T) 29.18 419.75 3.76 −51.02 −59.07 337.45 31.04 270.63 31.74 200.03 31.48 576.70 225.57 379.12 85.21 118.40 740.04 30.72 −92.87 43.24 178.18 24.14 90.06 386.99 26.54 89.87 Exp 28.51 409.6 0.9 −62.74 −61.6 323.6 30.05 264.54 30.68 195 30.61 Vib −0.33 −11.8 −2.42 −5.75 −4.33 −14.23 −0.52 −8.71 −0.61 −3.74 −0.63 Emp. Eq. 28.84 421.4 3.32 −56.99 −57.27 337.83 30.57 273.25 31.29 198.74 31.24 30.53 −0.41 30.94 147.7 −12.32 160.02 87.5 374.3 25.7 90.6 0.09 1.14 −0.1 0.13 87.41 373.16 25.8 90.47 Summary Convex formulation of paramagnetic CDFT available if one works with u = v + 12 A2 . Enables the calculation of AC curves through convex optimization. FCI/CCSD quality approximations to exact AC curves can be computed FCI level calculations within BDFT framework implemented CCSD and CDFT framework nearly implemented Comparison of model AC curves from existing DFT and CDFT functionals will hopefully lead to a better understanding of why they fail. . . and how to improve them Existing current-dependent VRG functionals show a poor low-density limit and degrade the accuracy of magnetic properties for molecules. DFT in strong fields. . . Acknowledgments 32/35 Thanks to. . . CDFT work: Trygve Helgaker Andrew M. Teale & his student James Furness Ulf Ekström Simen Kvaal Espen Sagvolden Other: Kai Kaarvann Lange, Mark R. Hoffmann (FCI/CASSSCF) Stella Stopkowicz (CC), Jon Austad (Laplace MP2)—will soon be useful for AC studies CTCC The audience Erik Tellgren (and T. Helgaker, A. M. Teale, S. Kvaal, U. Ekström, DFT E. Sagvolden) in strong fields. . . 32/35 More slides. . . Paramagnetic CDFT: counter-example to presumed HK theorem Helium atom in uniform magnetic field B = Bez , and symmetric gauge A = 21 B × r. . . For fields not strong enough to induce a level-crossing, the Hamiltonians H[v , A] = H[v + 12 A2 , 0] (33) have the same ground state (with vanishing canonical angular momentum). No HK-type mapping between potentials and ground states Some limited use of HK ideas still possible: (ρ, jp ) does determine the unique ground state wave function (alt. the subspace of deg. wave functions), and therefore all A-independent properties Paramagnetic CDFT: counter-example to presumed HK theorem Helium atom in uniform magnetic field B = Bez , and symmetric gauge A = 21 B × r. . . For fields not strong enough to induce a level-crossing, the Hamiltonians H[v , A] = H[v + 12 A2 , 0] (33) have the same ground state (with vanishing canonical angular momentum). No HK-type mapping between potentials and ground states Some limited use of HK ideas still possible: (ρ, jp ) does determine the unique ground state wave function (alt. the subspace of deg. wave functions), and therefore all A-independent properties Paramagnetic CDFT and convexity Example of Banach spaces (S. Kvaal): w 2 ρ ∈ L1 (R3 ), w −2 ∞ 3 u ∈ L (R ), w jp ∈ L1 (R3 )3 , w −1 ∞ (34) 3 3 A ∈ L (R ) (35) where, e.g., w (r) = 1 + r n√ (require well-defined moments) w (r) = 1 + e 1+r (require near-exponential decay) Pairings (ρ|u) and (jp |A) are now always well-defined. Only allow gauge transformations jp 7→ jp + ρ∇χ, A 7→ A + ∇χ satisfying w −1 ∇χ ∈ L∞ (R3 )3 √ Matching constraint on the wave function: w ψ ∈ H 1/2 , with H 1/2 a fractional-order Sobolev space.12 12 Follows if |∇w (r)| ≤ C · w (r), which is true for the two examples. (36)