DISCRETE AND CONTINUOUS Website: http://AIMsciences.org DYNAMICAL SYSTEMS Volume 8, Number 2, April 2002 pp. 381–397
AN OVERVIEW ON LINEAR AND NONLINEAR DIRAC EQUATIONS
MARIA J. ESTEBAN AND ERIC SER´ E´
Abstract. In this paper we review the variational treatment of linear and nonlinear eigenvalue problems involving the Dirac operator. These problems arise when searching for bound states and bound energies of electrons sub- mitted to external or self-consistent interactions in a relativistic framework. The corresponding energy functionals are totally indefinite and that creates difficulties to define variational principles related to those equations. Here we describe recent works dealing with this kind of problems. In particular we describe the solutions found to linear Dirac equations with external potential, the Maxwell-Dirac equations, some nonlinear Dirac equations with local non- linear terms and the Dirac-Fock equations arising in the description of atoms and molecules.
1. Introduction and basic properties of the dirac operator. In relativistic quantum mechanics [11], the state of a free electron is represented by a wave func- tion Ψ(t, x) with Ψ(t, .) ∈ L2(IR3, CI 4) for any t. This wave satisfies the free Dirac equation:
3 X 2 i ∂t Ψ = H0Ψ, with H0 = −ic~ αk∂k + mc β, (1.1) k=1 where c denotes the speed of light, m > 0, the mass of the electron, and ~ is Planck’s constant. In the Dirac equation, α1, α2, α3 and β are 4×4 complex matrices, whose standard form (in 2 × 2 blocks) is I 0 0 σk β = , αk = (k = 1, 2, 3), 0 −I σk 0 with 0 1 0 −i 1 0 σ = , σ = , σ = . 1 1 0 2 i 0 3 0 −1 One can easily check the following relations α = α∗ , β = β∗, k k (1.2) αkα` + α`αk = 2δk` , αkβ + βαk = 0.
1991 Mathematics Subject Classification. 49S05, 35J60, 35P30, 35P05, 35Q75, 49R05, 49R10, 81Q05, 81V70, 81V45. Key words and phrases. Relativistic quantum mechanics, Dirac operator, eigenvalues, varia- tional methods, indefinite integrals, nonlinear eigenvalue problems, nonlinear elliptic equations, min-max.
381 382 MARIA J. ESTEBAN AND ERIC SER´ E´
These algebraic conditions are here to ensure that H0 is a symmetric operator, such that 2 2 2 4 H0 = −c ∆ + m c . (1.3) Unless otherwise stated, we shall work with a system of physical units such that m = c = ~ = 1. Before going further, let us fix some notations. All throughout these notes, the z1 ∗ . 4 conjugate of z ∈ CIwill be denoted by z . For X = . a column vector in CI , z4 ∗ ∗ ∗ we denote by X the row covector (z1 , ..., z4 ). Similarly, if A = (aij) is a 4 × 4 ∗ ∗ ∗ complex matrix, we denote by A its adjoint, (A )ij = aji. We denote by (X,X0) the Hermitian product of two vectors X, X0 in CI 4, and by 4 4 2 X ∗ |X| , the norm of X in CI , i. e. |X| = XiXi . The usual Hermitian product i=1 in L2(IR3, CI 4) is denoted Z (ϕ, ψ) = ϕ(x), ψ(x) d3x. (1.4) L2 IR3 This course will be devoted to the presentation of various results concerning the existence of stationary solutions of linear and nonlinear Dirac equations. This is not an exhaustive description of the existing literature. Our main purpose is to describe some problems, methods, results and references which will make it possible for graduate students and researchers to be initiated to the study of some linear and nonlinear partial differential equations arising in models of the Relativistic Quantum Mechanics. Linear Dirac equations like
(H0 + V )ψ = λψ (1.5) appear when looking for bound states of an electron subjected to the action of an external electrostatic potential V . Chapter 2 will be devoted to the study of (1.5). Nonlinear Dirac equations appear in different contexts. For instance: • When looking for solutions of (1.5) in special classes of functions ψ. We will see this in detail when describing the Dirac-Fock equations (Chapter 5). • When building relativistic models of extended particles by means of nonlinear Dirac fields. This will be described via the Soler model in Chapter 3. • When analyzing the iteraction of an electron with its own electrostatic po- tential. This problem of self-interaction can be approximately described by the so-called Maxwell-Dirac equations, which are nonlinear Dirac equations, where the nonlinearities are nonlocal (of convolution type). This will be treated in Chapter 4. Let us now list some basic important properties of the free Dirac operator. 2 3 4 1 3 4 (P1) H0 is a self-adjoint operator on L (IR , CI ), with domain D(H0) = H (IR , CI ). Its spectrum is (−∞, −1] ∪ [1, +∞). There are two orthogonal projectors on 2 3 4 + − + L (IR , CI ) , Λ and Λ = 1 2 − Λ , both with infinite rank, and such that L √ √ + + + + H0Λ = Λ H0 = √1 − ∆ Λ = Λ 1 −√∆ − − − − (1.6) H0Λ = Λ H0 = − 1 − ∆ Λ = −Λ 1 − ∆ . AN OVERVIEW ON LINEAR AND NONLINEAR DIRAC EQUATIONS 383
(P2) The coulombic potential V (x) = 1|x| satisfies the following Hardy-type in- equalities: 1 π 2 ϕ, (µ ∗ V ) ϕ ≤ + ϕ, |H0|ϕ , (1.7) L2 2 2 π L2 for all ϕ ∈ Λ+(H1/2) ∪ Λ−(H1/2) and for all probability measures µ on IR3. Moreover,
π 1/2 ϕ, (µ ∗ V ) ϕ ≤ ϕ, |H0|ϕ , ∀ϕ ∈ H (1.8) L2 2 L2
k (µ ∗ V ) ϕk ≤ 2k∇ϕk , ∀ϕ ∈ H1 (1.9) L2 L2 In the particular case where µ is equal to the Dirac mass at the origin δ0, an inequality more precise than (1.8) was proved by Tix ([76]) and by Burenkov-Evans ([14]). This inequality reads as follows αZ H − ϕ, ϕ ≥ ((1 − αZ)ϕ, ϕ) , 0 |x| 2 + 1/2 3 4 for all Z ≤ Zc := π 2 , for all ϕ ∈ Λ (H (IR , CI )), where α ∼ 1/137 is ( 2 + π )α the so-called fine structure constant. The technique used to prove this inequality is based on ideas introduced by Evans-Perry and Siedentop in [34]. We refer to [48], [52] for inequality (1.8) in the case µ = δ0. Thaller’s book [75] gathers many results on the Dirac operator, including (P1) and the standard Hardy inequality (1.9) for µ = δ0, with references. The extension of (1.7), (1.8) and (1.9) from µ = δ0 to a general probability measure µ is immediate, since the projectors Λ±, the gradient ∇ and the free Dirac operator H0 commute with translations. For ϕ ∈ L2(IR3, CI 4), let us denote ϕ+ = Λ+ϕ, ϕ− = Λ−ϕ. Let E = H1/2(IR3, CI 4),E+ = Λ+E,E− = Λ−E. E is a Hilbert space with Hermitian product √ ϕ, ψ = ϕ, 1 − ∆ ψ = ϕ+, ψ+ + ϕ−, ψ− . (1.10) E L2 E E 2. Eigenvalue problems for dirac operators. In this chapter we are interested in perturbed Dirac operators of the form H0 + V , V being a scalar potential. Our goal is to find variational characterizations for the eigenvalues of such operators. One expects the eigenvalues of H0 +V to be critical points (in an appropriate sense) of the Rayleigh quotient ((H + V )ϕ, ϕ) R (ϕ) := 0 , V (ϕ, ϕ) 2 3 where (·, ·) denotes the inner product in L (IR ). If H0 + V were bounded below, like Schr¨odinger operators −∆ − Z/|x| for instance, minimizing the above quotient would give the least eigenvalue (ground state energy). This standard approach fails for Dirac operators: for physically relevant potentials V , the Rayleigh quotient is neither bounded from above nor from below (for instance, [50]). This is the main difficulty we have to face. Many efforts have been devoted to the characterization and computation of Dirac eigenvalues. Some works deal with approximate potentials acting on 2-spinors. The idea is to write Dirac’s equation for the upper spinor in ψ by performing 384 MARIA J. ESTEBAN AND ERIC SER´ E´ some adequate transformation. Then, different procedures have been produced to find approximate Hamiltonians, which “become exact” in the nonrelativistic limit (c → +∞). For references in this direction see, for instance, Durand [27], Durand- Malrieu [28], van Lenthe-Van Leeuwen-Baerends-Snijders [56]. Other approximate Hamiltonians are constructed by using projectors. One of the first attempts in this direction was made by Brown and Ravenhall. In [12], they proposed the following (usually called the Brown-Ravenhall “no-pair”) Hamiltonian + + B = Λ (H0 + V )Λ (2.1) + where Λ = χ(0,∞)(H0) denotes the projection on the positive spectral subspace of the free Dirac operator. If 0 is not in the spectrum of H0 + V , then instead of defining B as above, one + can use the exact positive projector corresponding to H0 + V , denoted by ΛV . The new operator + + BeV = ΛV (H0 + V )ΛV (2.2) coincides with H0 + V in its positive spectral subspace. For attractive potentials V that are not “too strong”, minimizing the Rayleigh quotient for BeV yields the smallest positive eigenvalue of H0 + V . This eigenvalue may be interpreted as + the ground state energy. Unfortunately, the exact projector ΛV is not known a priori. This makes its use difficult in numerical computations. On the contrary, the approximate Brown-Ravenhall Hamiltonian B is known explicitly. Several recent papers have been devoted to the mathematical study of the Brown- Ravenhall Hamiltonian for one-electron atoms with nuclear charge Z > 0. Its precise form is Zα BZ := Λ H − Λ (2.3) + 0 |x| + where α is the fine structure constant. Hardekopf and Sucher [46], on the basis of numerical computations, conjectured that the operator BZ is strictly positive for −1 Z < Zc = 2α /(π/2 + 2/π), that its first eigenvalue vanishes for Z = Zc, and that BZ is unbounded from below for Z > Zc. Evans, Perry and Siedentop [34] proved Z that the operator B is bounded from below if and only if Z ≤ Zc, as predicted in Z [47]. Tix [76], [77] proved that the operator B is strictly positive for Z ≤ Zc (see also Burenkov-Evans [14]). Note that the integer part of Zc is 124, strictly below the critical value Z∗ = 137, for which the first eigenvalue of the Dirac-Coulomb Hamiltonian H0 − Zα/|x| vanishes (see for instance [64], [75]). As we already pointed out above, one the most important properties of the free Dirac operator H0 is that 2 H0 = −∆ + 1 (2.4) Another Hamiltonian having the same property is: √ −∆ + 1 (2.5) This Hamiltonian is positive definite, and so, it does not pose the problems related to the unboundedness of the Dirac operator. But it is nonlocal, which contradicts the Causality Principle in special relativity and does not preserve the symmetry in the t and x variables for the evolution problem (covariance of the equation under the action of Lorentz boosts). Nevertheless, the study of the spectrum of √ −∆ + 1 + V (2.6) AN OVERVIEW ON LINEAR AND NONLINEAR DIRAC EQUATIONS 385 and its use as an approximation model for multi-particle systems is interesting. For references in this direction, let us mention [22], [48], [55], [57] and references therein. In 1994, Hill and Krauthauser ([50]) proposed a variational characterization of eigenvalues of H0 + V based on maximization of quotients like
(ψ, (H0 + V )ψ) 2 . (ψ, (H0 + V ) ψ)
This method works when the spectrum of H0 + V has a gap around 0. The math- ematical study of this approach has still to be done. The characterization of the stable eigenvalues of H0 +V by a minimax method is a quite natural idea and it has been attempted both analytically and numerically: in 1986, Talman [74] proposed a minimax procedure to find the first eigenvalue of H0 + V for negative potentials V (see also [21]). His proposal consisted in first maximizing in the lower component of the spinor ψ, and then minimizing over the upper spinor. Various results concerning the spectrum of H0 + V can be found in [75] and papers quoted therein. Minimax approaches to characterize the lowest eigenvalue of H0 + V have been proposed in the Physics literature (see for instance [26], [54], [74] and [21]). Mathematical rigorous results on this topic can be found in [31], [45] and [24]. In [31] was proved a minimax characterization of all eigenvalues of H0 + V for potentials V which are less singular at the origin than Coulomb potentials −1 1 V = −C|x| , for 0 < C < 2 :
((H0 + V )ϕ, ϕ) λk = inf sup . (2.7) + (ϕ, ϕ) F ⊂ Y ϕ ∈ F ⊕ Y − F vector space ϕ6=0 dim F = k Here, Y ± are the spaces Λ±(H1/2(IR3)). In [45], a minimax approach close to the above one is shown to provide all the eigenvalues of operators with spectral gaps which satisfy a number of conditions. This allows the authors to bound the n-th eigenvalue of the Dirac operator A = H0 + V with Coulomb potential V by the n-th eigenvalue of the Brown-Ravenhall + + + operator Λ (H0 + V )Λ , where Λ = χ(0,∞)(H0). Moreover, for certain bounded non-positive potentials, the number of eigenvalues of H0 + V is shown to be larger than or equal to those of the operator −∆ + V . Finally, also in [45], defining ϕ ψ = , ϕ, χ ∈ IR3 → CI2, the equality χ
((H0 + V )ψ, ψ) λ1(H0 + V ) = min max , (2.8) ϕ6=0 χ (ψ, ψ) and its corresponding counterpart for the n-th eigenvalue, are proved to hold for a particular class of bounded potentials V . The formula (2.8) had been proposed (without proof) by Talman [74] and Datta-Deviah [21] for numerical and compu- tational purposes, without asking ϕ to be different from 0. In [24], two approaches are proposed. The first one characterizes all eigenvalues of H0 + V by a minimax formula for all potentials V which are nonpositive, small at infinity and which have a singularity at 0 less strong than −C |x|−1, C < 0.9. 386 MARIA J. ESTEBAN AND ERIC SER´ E´
To prove this, the authors strongly use an inequality proved in [14] and [76], [77]. Finally, for potentials which behave at 0 less singularly than −|x|−1, a minimization characterization of λ1(H0 + V ) is shown to hold. Moreover, one can prove that Talman’s formula (2.8) (and the corresponding one for the n-th eigenvalue) holds true for all potentials which are nonpositive, small at infinity and which have a singularity at 0 less strong than −|x|−1 (condition which is known to be optimal). One can also show that there is a strong link between Talman’s formula and the minimization procedure proposed in section 4 of [24] to characterize λ1(H0 + V ). All this is done in [25].
3. The nonlinear local dirac equations: the soler model. generalized extended particles models. Nonlinear Dirac equations have been widely used to build relativistic models of extended particles by means of nonlinear Dirac fields. A general form of this equation in the case of an elementary fermion is µ 0 iγ ∂µψ − mψ + γ ∇F (ψ) = 0. (3.1) 4 4 ∂ Here, ψ : IR → CI , ∂µψ = ∂xµ ψ, 0 ≤ µ ≤ 3, we have used Einstein’s convention for summation over µ. Note that x0 plays here the role of time. Throughout this chapter we assume that F satisfies F ∈ C2(CI 4, IR) and F (eiθψ) = F (ψ) for all θ. (3.2) Different functions F have been used to model various types of self-couplings. For a review on this and historical background see for instance [69]. Stationary states of the nonlinear Dirac equation are considered as particle-like solutions. These solutions are in some sense solitons which propagate without changing their shape. Stationary solutions are functions of the type
0 ψ(x0, x) = e−iωx ϕ(x) (3.3) where by x we denote (x1, x2, x3) ∈ IR3, and such that ϕ is a non-zero localized solution of the following stationary nonlinear Dirac equation k 0 0 3 iγ ∂kϕ − mϕ + ωγ ϕ + γ ∇F (ϕ) = 0 in IR . (3.4) Here, we have used the repeated index convention for summation over the three values of k. By “localized”, we mean that ϕ ∈ W 1,q(IR3, CI 4) for any 2 ≤ q < ∞ and F (ϕ) ∈ L1(IR3, IR). The above equation has a variational structure. Indeed it is the Euler-Lagrange equation corresponding to the following functional Z ω 1 0 k m ω 2 I (ϕ) = − (iγ γ ∂kϕ, ϕ) + ϕϕ − |ϕ| − F (ϕ) dx (3.5) IR3 2 2 2 where ( , ) is the usual scalar product in CI 4 and ϕϕ denotes (γ0ϕ, ϕ). Existence of solutions of (3.4), i.e. of critical points of Iω has been proved in [7, 8, 16, 65] in the particular case when 1 F (ϕ) = G(ϕϕ) ,G ∈ C2(IR, IR) ,G(0) = 0 (3.6) 2 AN OVERVIEW ON LINEAR AND NONLINEAR DIRAC EQUATIONS 387 under suitable conditions on G, with ω ∈ (0, m). When (3.6) holds, one can use a particular ansatz for the solutions: 1 v(r) 0 ϕ(x) = (3.7) cos θ iu(r) sin θ eiφ and then equation (3.4) reduces to the O.D.E. system u0 + 2u = vg(v2 − u2) − (m−ω) r (3.8) v0 = ug(v2 − u2) − (m+ω) where g(s) = G0(s). In the above papers the system (3.8) is solved by a shooting method which yields an infinity of localized solutions for (3.8). Below we will state the precise results of [7, 16, 65]. The particular nonlinearity (3.6) corresponds to the so-called Soler model, and has been widely studied (see [69, 70]). There are other models of selfcoupling for which the ansatz (3.7) is no more valid, for instance the following nonlinearity (see [36, 69]) 1 F (ϕ) = |ϕϕ|2 + b |ϕγ5ϕ|2 (3.9) 2 with b 6= 0, ϕ = γ0ϕ and γ5 = γ0γ1γ2γ3. We can prove, by a variational method, the existence of non-zero critical points of Iω, and therefore of localized solutions of (3.4), both in the case when F satisfies (3.6) and in the general case. When F satisfies (3.6), we actually solve the O.D.E. system (3.8) by a variational method to obtain the existence of an infinity of solutions of (3.8) under assumptions slightly less general than those in [7, 16, 65]. The main assumption we have to add is that G satisfies G0(x) · x ≥ θ G(x) , θ > 1 , ∀ x ∈ IR (3.10) This is a superlinearity condition on G. This kind of technical condition often appears as necessary to use variational techniques when solving nonlinear PDE’s and was first introduced by Ambrosetti and Rabinowitz in [3]. If F is a general function which does not necessarily satisfy (3.6), we also obtain an existence result for (3.4) but only for functions F which grow more slowly than |ϕ|3 at infinity. This limitation comes from the fact that we work in the space H1/2(IR3, CI 4). Hence we are not able yet to treat nonlinearities like those satisfying (3.9). Nevertheless we believe that the method used in our paper ([29]) should be useful to extend our results and treat more general nonlinearities. Our results are concerned, for instance, with nonlinearities as 1 3 |ϕϕ|α + b |ϕγ5ϕ|β , 1 < α, β < , b ≥ 0. (3.11) 2 2 Let us now state our main results about equations (3.4) and (3.8). Theorem 3.1 ([29]). Let F : CI 4 → IR satisfy (3.6) with G ∈ C2(IR, IR) . Denoting by g the first derivative of G, we make the following assumptions: (H1) g(x) · x ≥ θ G(x) , θ > 1 , ∀x ∈ IR (H2) G(0) = g(0) = 0. 388 MARIA J. ESTEBAN AND ERIC SER´ E´
(H3) G(x) ≥ 0 (∀x ∈ IR) and G(A0) > 0 for some A0 > 0. (H4) 0 < ω < m. T 1,q 3 4 Then there is an infinity of solutions of equation (3.4) in 2≤q<+∞ W (IR , CI ). Each of them is found by a min-max on the functional Iω. They are of the form (3.7), so they correspond to classical solutions of (3.8) on IR+, decreasing exponen- tially at infinity. For more general nonlinearities F , not compatible with the ansatz (3.7), we have