A new proof of the analyticity of the electronic density of molecules.

Transcription

1 A new proof of the analyticity of the electronic density of molecules. Thierry Jecko AGM, UMR 8088 du CNRS, Université de Cergy-Pontoise, Département de mathématiques, site de Saint Martin, 2 avenue Adolphe Chauvin, F Cergy-Pontoise cédex, France. thierry.jecko@u-cergy.fr web: Abstract We give a new, short proof of the regularity away from the nuclei of the electronic density of a molecule obtained in [FHHS1, FHHS2]. The new argument is based on the regularity properties of the Coulomb interactions underlined in [KMSW] and on well-known elliptic technics. Keywords: Elliptic regularity, analytic elliptic regularity, molecular Hamiltonian, electronic density, Coulomb potential. 1

2 Analyticity of the electronic density, Introduction. For the quantum description of molecules, it is very useful to study the so-called electronic density and, in particular, its regularity properties. This has be done for molecules with fixed nuclei: see [FHHS1, FHHS2, FHHS3] for details and references. The smoothness and the analyticity of the density away from the nuclei are proved in [FHHS1] and [FHHS2] respectively. In this paper, we propose an alternative proof. Let us recall the framework and the precise results of [FHHS1, FHHS2]. We consider a molecule with N moving electrons N 2) and L fixed nuclei. While the distinct vectors R 1,, R L R 3 denote the positions of the nuclei, the positions of the electrons are given by x 1,, x N R 3. The charges of the nuclei are given by the positive Z 1,, Z L and the electronic charge is 1. In this picture, the Hamiltonian of the system is H := where E 0 = N xj j=1 1 k<k L L k=1 Z k Z k R k R k ) Z k x j R k + 1 j<j N 1 x j x j + E 0 1.1) and xj stands for the Laplacian in the variable x j. Setting := N j=1 xj, we define the potential V of the system as the multiplication operator satifying H = + V. Thanks to Hardy s inequality c > 0 ; f W 1,2 R 3 ), t 1 ft) ) 2 dt c ft) 2 dt, 1.2) R 3 R 3 one can show that V is -bounded with relative bound 0 and that H is self-adjoint on the domain of the Laplacian, namely W 2,2 R 3N ) see Kato s theorem in [RS], p ). Let ψ W 2,2 R 3N ) \ {0} and E R such that Hψ = Eψ. Actually E is smaller than E 0 by [FH]. The electronic density associated to ψ is the following L 1 R 3 )-function ρx) := N j=1 R 3N 1) ψx 1,, x j 1, x, x j,, x N ) 2 dx 1 dx j 1 dx j dx N. Here we used N 2. The regularity result is the following Theorem 1.1. [FHHS1, FHHS2]. The density ρ is real analytic on R 3 \ {R 1,, R L }. Remark 1.2. In [FHHS1], it is proved that ρ is smooth on R 3 \{R 1,, R L }. This result is then used in [FHHS2] to derive the analyticity. Now let us sketch the new proof of Theorem 1.1, the complete proof and the notation used are given in Section 2. We consider the almost everywhere defined L 2 -function ψ : R 3 x ψx,,, ) W 2,2 R 3N 1) ) 1.3) and denote by the L 2 R 3N 1) )-norm. By permutation of the variables, it suffices to show that the map R 3 x ψx) 2 belongs to C ω R 3 \ {R 1,, R L }; R), the space of

3 Analyticity of the electronic density, real analytic functions on R 3 \ {R 1,, R L }. We define the potentials V 0, V 1 by V = V 0 + V 1 with V 0 x) = E 0 L k=1 Z k x R k C ω R 3 \ {R 1,, R L }; R). 1.4) We view the function ψ as a distributional solution in D R 3 ; W 2,2 R 3N 1) )) of x ψ + Qx) ψ = 0, 1.5) where the x-dependent operator Qx) B := LW 2,2 R 3N 1) ); L 2 R 3N 1) )) is given by Qx) = x + V 0 E + V 1 with x = N xj. 1.6) j=2 Considering 1.5) in a small enough neighbourhood Ω of some R 3 \ {R 1,, R L }, we pick from [KMSW] a x-dependent unitary operator U x0 x) on L 2 R 3N 1) ) such that W : Ω x U x0 x)v 1 x)u x0 x) 1 B 1.7) belongs to C ω Ω; B). It turns out that P 0 = U x0 x x )Ux 1 0 is an elliptic differential operator in x with analytic, differential coefficients in B. Applying U x0 to 1.5) and setting ϕ = U x0 ψ, we obtain P0 + W x) + V 0 x) E ) ϕ = ) Since U x0 x) is unitary on L 2 R 3N 1) ), ψx) = ϕx). Thus, it suffices to show that ϕ C ω Ω; L 2 R 3N 1) )). Using 1.8), a parametrix of the operator P 0 + W + V 0, we show by induction that, for all k, ϕ W k,2 Ω; W 2,2 R 3N 1) )). Thus ϕ C Ω; W 2,2 R 3N 1) )). Finally we can adapt the arguments in [H1] p to get ϕ C ω Ω; W 2,2 R 3N 1) )), yielding ϕ C ω Ω; L 2 R 3N 1) )). The main idea in the construction of the unitary operator U x0 is to change, locally in x, the variables x 2,, x N in a x-dependent way such that the x-dependent singularities 1/ x x j becomes locally x-independent see Section 2). In [KMSW], where this clever method was introduced, the nuclei positions play the role of the x variable and the x 2,, x N are the electronic degrees of freedom. The validity of the Born-Oppenheimer approximation is proved there for the computation of the eigenvalues and eigenvectors of the molecule. We point out that this method is the core of a recently introduced, semiclassical pseudodifferential calculus adapted to the treatment of Coulomb singularities in molecular systems, namely the twisted h-pseudodifferential calculus h being the semiclassical parameter). This calculus is due to A. Martinez and V. Sordoni in [MS]. As one can see in [KMSW, MS], the above method works in a larger framework. So do Theorem 1.1 and our proof. For instance, we do not need the positivity of the charges Z k, the fact that E < E 0, and the precise form of the Coulomb interaction. We do not use the self-adjointness or the symmetry) of the operator H. We could replace in 1.1) each xj by i xj + Ax) 2, where A is a suitable, analytic, magnetic vector potential. We could also add a suitable, analytic exterior potential.

4 Analyticity of the electronic density, Let us now compare our proof with the one in [FHHS1, FHHS2]. Here we only use almost) classical arguments of elliptic regularity. In [FHHS1, FHHS2], the elliptic regularity is essentially replaced by some Hölder continuity regularity result on ψ. The authors introduced an adapted, smartly chosen variable w.r.t. which they can derivate ψ. Here the x-dependent change of variables produces regularity with respect to x. As external tools, we only exploit basic notions of pseudodifferential calculus, the rest being elementary. In [FHHS1, FHHS2], a general, involved regularity result from the literature on PDE is an important ingredient of the arguments. We believe that, in spirit, the two proofs are similar. The shortness and the relative simplicity of the new proof is due to the clever method borrowed from [KMSW], which transforms the singular potential V 1 in an analytic function with values in B. Acknowledgment: The author is supported by the french ANR grant NONAa and by the european GDR DYNQUA. He thanks Vladimir Georgescu, Sylvain Golénia, Hans- Henrik Rugh, and Mathieu Lewin, for stimulating discussions. 2 Details of the proof. Here we complete the proof of Theorem 1.1, sketched in Section 1. Notation and basic facts. For a function f : R d R n x,y) fx,y) R p, let d x f be the total derivative of f w.r.t. x, by x α f with α N d the corresponding partial derivatives. For α N d and x R d, Dx α := i x ) α := i x1 ) α1 i xd ) α d, D x = i x, x α := x α 1 1 x α d d, α := α 1+ +α d, α! := α 1!) α d!), x 2 = x x 2 d, and x := 1 + x 2 ) 1/2. If A is a Banach space and O an open subset of R d, we denote by Cc O; A) resp. Cb O; A), resp. C ω O; A)) the space of functions from O to A which are smooth with compact support resp. smooth with bounded derivatives, resp. analytic). Let D O; A) denotes the topological dual of Cc O; A). We use the traditional notation W k,2 O; A) for the Sobolev spaces of L 2 O; A)-functions with k derivatives in L 2 O; A) when k N and for the dual of W k,2 O; A) when k N. If A is another Banach space, we denote by LA; A ) the space of the continuous linear maps from A to A and set LA) = LA; A). For A LA) with finite dimensional A, A T denotes the transpose of A and DetA its determinant. By the Sobolev injections, W k,2 O; A) C O; A). 2.1) k N Denoting by A the norm of A, it is well-known cf. [H3]) that a function u C O; A) is analytic if and only if, for any compact K O, there exists some C > 0 such that α N d, Dx α u)x) A C α +1 α!). 2.2) sup x K For convenience, we set W k = W k,2 R 3N 1) ), for k N. Recall that B = LW 2 ; W 0 ). Let B = LW 0 ; W 2 ), B 0 = LW 0 ), and B 2 = LW 2 ). Construction of U x0 see [KMSW, MS]). Let τ C c R 3 ; R) such that τ ) = 1 and τ = 0 near R k, for all k {1; ; L}. For x, s R 3, we set fx,s) = s + τs)x ).

5 Analyticity of the electronic density, Notice that x; s) R 3 ) 2, fx, ) = x and fx,s) = s if s supp τ. 2.3) Since d s f)x,s).s = s + τs), s x ), we can choose a small enough, relatively compact neighborhood Ω of such that x Ω, sup d s f)x,s) I 3 LR 3 ) 1/2, 2.4) s I 3 being the identity matrix of LR 3 ). Thus, for x Ω, fx, ) is a C -diffeomorphism on R 3 and we denote by gx, ) its inverse. By 2.4) and a Neumann expansion in LR 3 ), ds f)x,s) ) 1 = I3 + τs), x x0 ) ) n 1 ) τs), x ), n=1 for x, s) Ω R 3. Notice that the power series converges uniformly w.r.t. s. This is still true for the series of the derivatives β s, for β N 3. Since d s g)x, fx,s)) = d s f)x,s) ) 1 and dx g)x, fx,s)) = d s g)x, fx,s)) d x f)x,s), we see by induction that, for α, β N 3, α x β s g ) x, fx,s)) = γ N 3 x ) γ a αβγ s) 2.5) on Ω R 3, with coefficients a αβγ C R 3 ; LR 3 )). For α = β = 0, this follows from gx, fx,s)) = s. Notice that, except for α, β, γ) = 0,0,0) and for β = 1 with α, γ) = 0,0), the coefficients a αβγ are supported in the compact support of τ. For x R 3 and y = y 2,, y N ) R 3N 1), let F x,y) = fx,y 2 ),, fx,y N )). For x Ω, F x, ) is a C -diffeomorphism on R 3N 1) satisfying the following properties: There exists C 0 > 0 such that, for all α N 3, for all x Ω, for all y, y R 3N 1), C0 1 y y F x,y) F x,y ) C 0 y y, 2.6) x α F x,y) x α F x,y ) C 0 y y, 2.7) and, for α 1, α x F x,y) C ) For x Ω, denote by Gx, ) the inverse diffeomorphism of F x, ). By 2.5), the functions Ω R 3N 1) x,y) α x β y G ) x, F x,y)), for α, β) N 3 N 3N 1), are also given by a power series in x with smooth coefficients in y. Given x Ω, let U x0 x) be the unitary operator on L 2 R 3N 1) ) defined by U x0 x)θ)y) = Detd y F )x,y) 1/2 θf x,y)).

6 Analyticity of the electronic density, Computation of the terms in 1.8) cf. [KMSW, MS]). Consider the functions Ω x J 1 x, ) Cc R 3N 1) ; LR 3N 1) ; R 3 ) ), Ω x J 2 x, ) Cc R 3N 1) ; R 3 ), Ω x J 3 x, ) Cb R 3N 1) ; LR 3N 1) ) ), Ω x J 4 x, ) C c R 3N 1) ; R 3N 1) ), defined by J 1 x,y) = d x Gx,y )) T x, y = F x,y) ), J 2 x,y) = Det dy F x,y) 1/2 Det Dx dy Gx,y ) 1/2 ) y, =F x,y) J 3 x,y) = d y Gx,y )) T x, y = F x,y) ), J 4 x,y) = Det Det d y F x,y) 1/2 D y dy Gx,y ) 1/2) y. =F x,y) Actually, the support of J k x, ), for k 3, is contained in the x-independent, compact support of the function τ cf. 2.3)). So do also the supports of the derivatives α x β y J 3 of J 3, for α + β > 0. Thanks to 2.5), the J k,y) s can also be written as a power series in x with smooth coefficients depending on y. Now U x0 x U 1 = x + J 1 y + J 2 and U x0 x U 1 = J 3 y + J ) In particular, U x0 x) preserves W 2,2 R 3N 1) ), for all x Ω. Furthermore, ) P 0 = U x0 x x U 1 = x + J 1 x; y; D y ) D x + J 2 x; y; D y ), 2.10) where J 2 x; y; D y ) is a scalar differential operator of order 2 and J 1 x; y; D y ) is a column vector of 3 scalar differential operators of order 1. More precisely, the coefficients of J 1 x; y; D y ) and of J 2 x; y; D y ) + y are compactly supported, uniformly w.r.t. x. In particular, these scalar differential operators belong to B. By 2.5), they are given on Ω by a power series of x with coefficients in B and therefore are analytic functions on Ω with values in B cf. [H3]). Next, we look at W defined in 1.7). By 2.3), for j j in {2; ; N}, for k {1; ; L}, and for x Ω, U x0 x) x x j 1) U 1 x) = fx; ) fx; y j ) 1, 2.11) U x0 x) x j R k 1) U 1 x) = fx; y j ) fx; R k ) 1, 2.12) U x0 x) x j x j 1) U 1 x) = fx; y j ) fx; y j ) ) Lemma 2.1. The potential W in 1.7) is an analytic function from Ω to B. Proof: We prove the stronger result: W is analytic from Ω to B := LW 1 ; W 0 ). Notice that W is a sum of terms of the form 2.11), 2.12), and 2.13). We show the regularity of 2.11). Similar arguments apply for the other terms. We first recall the arguments in [KMSW], which proves the C regularity. Using the fact that d x fx, ) fx,y j )) does not depend on x, Dx α fx,x0 ) fx,y j ) 1) = τ ) τy j )) D α α 1 ) fx, ) fx,y j ))

7 Analyticity of the electronic density, for y j. By 2.6) and 2.7), we see that, for all α N 3 and for y j, Dx α fx,x0 ) fx,y j ) 1) 2 α C 0 fx, ) fx,y j ) α D α 1 fx,) fx,y j )) thanks to α N 3, C > 0, y R 3 \ {0}, C 2 α 0 Cα!) fx, ) fx,y j ) 1, D α 1 y) Cα!). 2.14) y α +1 Since x 1 is x -bounded by 1.2) and since U )x) is unitary, fx, ) fx,y j ) 1 is U )x) x U )x)) 1 -bounded with the same bounds. But, by 2.9), U )x) x U )x) 1 y + 1 ) 1/2 is uniformly bounded w.r.t. x. Thus Dx α fx,x0 ) fx,y j ) 1) C B 1C 2 α 0 Cα!), 2.15) uniformly w.r.t. α N 3 and x Ω. Therefore W is a distribution on Ω the derivatives of which belong to L Ω), thus to L 2 Ω). By 2.1), W is smooth. To show the analyticity of W, we just add the following improvement of 2.14), that we prove in appendix below. There exists K > 0 such that α N 3, y R 3 \ {0}, 1 Dα y) K α +1 α!). 2.16) y α +1 Now the l.h.s. of 2.15) is, for α N 3 and x Ω, bounded above by C 1 C 2 α 0 K α +1 α!) K α +1 1 α!), for some K 1 > 0. This yields the result by 2.2). Smoothness. Now we view 1.8) as an elliptic differential equation w.r.t. x with coefficients in B and want to follow usual arguments of elliptic regularity to prove the smoothness of ϕ. In fact, we shall use the basic pseudodifferential calculus in [H2] p ). Since the symbol of P 0 + W + V 0 takes its values in B, which is not an algebra of operators, we verify the validity of the basic calculus in our situation. By 2.10), P 0 + W + V 0 is a differential operator in the x variable the symbol of which px; ξ) = ξ 2 + J 1 x; y; D y ) ξ + J 2 x; y; D y ) + W x) + V 0 x) 2.17) and belongs to the Hörmander class Sm 2, g; B) on Ω R 3 with values in B, where mx; ξ) = ξ ) 1/2 and g = dx 2 + dξ 2 / ξ 2. We can check that the basic calculus of [H2] actually works with the symbols classes Sm, g; B), Sm, g; B ), Sm, g; B 0 ), and Sm, g; B 2 ), for any scalar) order function m on Ω R 3. In particular, we have the following properties: We can give a sense to asymptotic sums of symbols in Sm, g; A), for A {B, B, B 0, B 2 }. The composition of

8 Analyticity of the electronic density, operators a 1 x, D x )a 2 x, D x ) = bx, D x ) with b Sm, g; B 0 ) if a 1, a 2 Sm, g; B 0 ) and if a 1 Sm, g; B) and a 2 Sm, g; B ), also with b Sm, g; B 2 ) if a 1, a 2 Sm, g; B 2 ) and if a 1 Sm, g; B ) and a 2 Sm, g; B). The adjoint of an operator bx, D x ) is b x, D x ) with b Sm, g; B j ) if b Sm, g; B j ), b Sm, g; B ) if b Sm, g; B), and b Sm, g; B) if b Sm, g; B ). For A {B, B, B 0, B 2 }, k Z, and a Sm k, g; A), l Z, ax, D x ) L W l,2 Ω; A); W l k,2 Ω; A) ). 2.18) By the proof of Lemma 2.1, W x) y + 1) 1/2 is uniformly bounded on Ω. By the properties of J 1 and J 2, so are ξ J 1 x; y; D y ) ξ 2 y + 1) 1/2 and J 2 x; y; D y ) + y ) y + 1) 1/2 on Ω R 3. Thus, we can find a 0; 1) and b R such that, for all x, ξ) Ω R 3, for all u W 2, T u W0 a ξ 2 y + 1)u W0 + b u W0 with T = ξ J 1 x; y; D y ) + J 2 x; y; D y ) + y ) + W x) + V 0 x). By Theorem 4.11, p. 291 in [K], there exists C > 0 such that, for all x, ξ) Ω R 3, px, ξ) is bounded below by C + 1. In particular, C + p) 1 is a well-defined symbol in Sm 2, g; B ) on Ω R 3 this is the ellipticity we use). By composition, C + p) 1 x, D x ) p + C)x, D x ) = 1 qx, D x ) with q Sm 1, g; B 2 ). Defining the symbols q k in Sm k, g; B ) by q k x, D x ) = qx, D x ) k C + p) 1 x, D x ) and denoting by q an asymptotic sum of the symbols q k, we have QP 0 + W + V 0 + C) = 1 + R with 2.19) Q := q x, D x ) L W k,2 Ω; B ); W k+2,2 Ω; B ) ), 2.20) R L W k,2 Ω; B 2 ); W k+l,2 Ω; B 2 ) ), 2.21) for all k, l N cf. [H2]). Starting with ϕ W 0,2 Ω; W 2 ) and applying Q to 1.8), we see that ϕ = E +C)Qϕ Rϕ and actually belongs to W 2,2 Ω; W 2 ), by 2.19), 2.20), and 2.21). By induction and by 2.1), we get that ϕ C Ω; W 2 ). We have recovered the result in [FHHS1]. Note that, to get it, we need neither the refined bounds 2.16) nor the power series mentioned above but just use the fact that the functions f, g, F, G are smooth w.r.t. x. Analyticity. To show that ϕ C ω Ω; W 2 ), we adapt the proof of Theorem in [H1] for equation 1.8). So we view the latter as P ϕ = 0 where P = α 2 a α D α x with analytic coefficients a α B cf. Lemma 2.1, 1.4), and 2.10)). Applying D α x to 2.19) with α 2 and using 2.20) and 2.21), we find C > 0 such that, for all v C c Ω; W 2 ) and α N 3, α 2 = D α x v L 2 Ω;W 2 ) C P v L 2 Ω;W 0 ) + C v L 2 Ω;W 2 ). 2.22) For ɛ > 0, let Ω ɛ := {x Ω; dx; R 3 \ Ω) > ɛ} and, for r N, denote the L 2 Ω ɛ ; W r )-norm of v by Nɛ r v). As in [H1] Lemma 7.5.1), we use an appropriate cut-off function, Leibniz formula, and 2.22), to find C > 0 such that, for all v Cc Ω; W 2 ), for all ɛ, ɛ 1 > 0, for all α N 3 such that α 2, ɛ α N 2 ɛ+ɛ 1 D α x v ) Cɛ 2 N 0 ɛ 1 P v ) + C α <2 ɛ α N 2 ɛ 1 D α x v ). 2.23)

9 Analyticity of the electronic density, We used the fact that 2.23) holds true for ɛ large enough since the l.h.s. is zero. Next we show that there exists B > 0 such that, for all ɛ > 0, j N, and α N 3, α < 2 + j = ɛ α N 2 jɛ D α x ϕ ) B α ) This is done by induction on j following the arguments in [H1]. As explained in [H1], ϕ C ω Ω; W 2 ) follows from 2.24) and 2.2). A Appendix Using Cauchy integral formula for analytic functions in several variables cf. [H3]), we prove here the following extension of 2.16). For d N, there exists K > 0 such that α N d, y R d \ {0}, 1 Dα y) K α +1 α!). A.1) y α +1 In dimension d = 1, one can show A.1) with K = 1 by induction. To treat the general case, we use Cauchy inequalities for an appropriate analytic function. Let denote the analytic branch of the square root that is defined on C \ R. Take y R d \ {0}. On the polydisc D = { z = z 1,, z d ) C d ; j, z j < y /4 d) } d, R y j + z j ) 2 7/16) y 2 > 0. j=1 Thus the function u : D C given by uz) = 1 dj=1 y j + z j ) 2 is well defined and analytic. Its modulus is bounded above by y 1 4/ 7. By Cauchy inequalities cf. Theorem 2.2.7, p. 27, in [H3]), α N d, α z u0) 4 y 7 α!) y /4 d) ) α 4 d) α +1 α!) y α +1. A.2) Here zj := 1/2) Rzj + i Izj ) but it can be replaced by Rzj in the formula since u is analytic. Now A.1) follows from A.2) since, for all α, Rzu )0) α = i α D α 1 ) y). References [Fe] H. Federer: Geometric Measure Theory. Springer-Verlag 1969.

10 Analyticity of the electronic density, [FHHS1] S. Fournais, M. Hoffmann-Ostenhof, T. Hoffmann-Ostenhof, T. Østergaard Sørensen: The electron density is smooth away from the nuclei. Comm. Math. Phys. 228, no ), [FHHS2] S. Fournais, M. Hoffmann-Ostenhof, T. Hoffmann-Ostenhof, T. Østergaard Sørensen: Analyticity of the density of electronic wave functions. Ark. Mat. 42, no ), [FHHS3] S. Fournais, M. Hoffmann-Ostenhof, T. Hoffmann-Ostenhof, T. Østergaard Sørensen: Non-isotropic cusp conditions and regularity of the electron density of molecules at the nuclei. Ann. Henri Poincaré ), [FH] R.G. Froese, I. Herbst: Exponential bounds and absence of positive eigenvalues for N-body Schrödinger operators. Comm. Math. Phys. 87, ). [H1] L. Hörmander: Linear partial differential operators. Fourth printing Springer Verlag, [H2] L. Hörmander: The analysis of linear partial differential operators III. Springer Verlag, [H3] L. Hörmander: An introduction to complex analysis in several variables. Elsevier science publishers B.V., [K] T. Kato: Pertubation theory for linear operators. Springer-Verlag [KMSW] M. Klein, A. Martinez, R. Seiler, X.P. Wang: On the Born-Oppenheimer expansion for polyatomic molecules. Comm. Math. Phys. 143, no. 3, ). [MS] A. Martinez, V. Sordoni: Twisted pseudodifferential calculus and application to the quantum evolution of molecules. Preprint 2008, mp arc , to appear in the Memoirs of the AMS. [RS] M. Reed, B. Simon: Methods of Modern Mathematical Physics, Vol. II : Fourier Analysis, Self-adjointness. Academic Press, 1979.