Theory Notes
Contents
[hide]- 1 Helium Calculations
- 2 The Hartree Fock Method
- 3 Configuration Interaction
- 4 Hylleraas Coordinates
- 5 Completeness
- 6 Solutions of the Eigenvalue Problem
- 7 Matrix Elements of H
- 8 Radial Integrals and Recursion Relations
- 9 Graphical Representation
- 10 Matrix Elements of H
- 11 General Hermitean Property
- 12 Optimization of Non-linear Parameters
- 13 The Screened Hydrogenic Term
- 14 Small Corrections
Helium Calculations
[−ℏ22m(∇21+∇22)−Ze2r1−Ze2r2+e2r212]ψ=Eψ
Define ρ=Zra0 where a0=ℏ2me2 (Bohr radius). Then
[−ℏ22mZ2(me2ℏ2)2(∇2ρ1+∇2ρ2)−Z2e2a0ρ−11−Z2e2a0ρ−12+e2a0Zρ−112]ψ=Eψ
But ℏ2m(me2ℏ2)2=e2a0 is in atomic units (au) of energy. Therefore
[−12(∇2ρ1+∇2ρ2)−1ρ1−1ρ2+Z−1ρ12]ψ=εψ where ε=Ea0Z2e2
The problem to be solved is thus [12(∇21+∇22)−1r1−1r2+Z−1r12]ψ=εψ
The Hartree Fock Method
Assume that ψ(r1,r2) can be written in the form
ψ(r1,r2)=1√2[u1(r1)u2(r2)±u2(r1)u1(r2)]
for the 1S21S ground state
[−12(∇21+∇22)−1r1−1r2+Z−1r12]ψ(r1,r2)=Eψ(r1,r2)
Substitute into <ψ|H−E|ψ> and require this expression to be stationary with respect to arbitrary infinitesimal variations δu1 and δu2 in u1 and u2. ie
12<δu1(r1)u2(r2)±u2(r1)δu1(r2)|H−E|u1(r1)u2(r2)±u2(r1)u1(r2)>
=∫δu1(r1)dr1{∫dr2u2(r2)(H−E)[u1(r1)u2(r2)±u2(r1)u1(r2)]}
=0 for arbitrary δu1(r1).
Therefore {∫dr2…}=0.
Similarly, the coefficient of δu2 would give
∫dr1u1(r1)(H−E)[u1(r1)u2(r2)±u2(r1)u1(r2)]=0
Define
I12=∫dru1(r)u2(r),
I21=∫dru1(r)u2(r),
Hij=∫drui(−12∇−1r)uj(r),
Gij(r)=∫dr′ui(r′)1|r−r′|uj(r′)
Then the above equations become the pair of integro-differential equations
[H0−E+H22+G22(r)]u1(r)=∓[I12(H0−E)+H12+G12(r)]u2(r)
[H0−E+H11+G11(r)]u2(r)=∓[I12(H0−E)+H12+G12(r)]u1(r)
These must be solved self-consistently for the "constants" I12 and Hij and the function Gij(r).
The H.F. energy is E≃−2.87⋯a.u. while the exact energy is E=−2.903724⋯a.u.
The difference is called the "correlation energy" because it arises from the way in which the motion of one electron is correlated to the other. The H.F. equations only describe how one electron moves in the average field provided by the other.
Configuration Interaction
Expand ψ(r1,r2)=C0u(s)1(r1)u(s)1(r2)+C1u(P)1(r1)u(P)1(r2)Υ01,1,0(ˆr1,ˆr2)+C2u(d)1(r1)u(d)2(r2)Υ02,2,0(ˆr1,ˆr2)+...± exchange where ΥMl1,l2,L(ˆr1,ˆr2)=Σm1,m2Υm1l1(r1)Υm2l2(r2)×<l1l2m1m2∣LM>.
This works, but is slowly convergent, and very laborious. The best CI calculations are accurate to 10−7 a.u.
Hylleraas Coordinates
[E.A. Hylleraas, Z. Phys. 48,469(1928) and 54,347(1929)] suggested using the co-ordinates r1, r2 and r12 or equivalently
s=r1+r2,t=r1−r2,u=r12
and writing the trial functions in the form
Ψ(r1,r2)=∑1+j+k≤Ni,j,kci,j,kri+l11rj+l22rk12e−αr1−βr2YMl1,l2,L(ˆr1,ˆr2)±exchange
Diagonalizing H in this non-orthogonal basis set is equivalent to solving ∂E∂ci,j,k=0 for fixed α and β.
The diagonalization must be repeated for different values of α and β in order to optimize the non-linear parameters.
Completeness
The completeness of the above basis set can be shown by first writing r212=r21+r22−2r1r2cos(Θ12) and cos(Θ12)=4π3∑1m=−1Ym∗l(θ1,φ1)Yml(θ2,φ2) consider first S-states. The r012 terms are like the ss terms in a CI calculation. The r212 terms bring in p-p type contributions, and the higher powers bring in d-d, f-f etc type terms. In general Pl(cos(θ12)=4π2l+1∑lm=−lYml∗(θ1,φ1)Yml(θ2,φ2)
For P-states, one would have similarly
r012 (sp)Pr212 (pd)Pr412 (df)P⋮ ⋮
For D-states
r012 (sp)D (pp′)Dr212 (pd)D (dd′)Dr412 (df)D (ff′)D⋮ ⋮ ⋮
In this case, since there are two lowest-order couplings to form a D-state, both must be present in the basis set. ie
Ψ(r2,r2)=∑cijkri1rj+22rk12e−αr1−βr−2YM022(ˆr1,ˆr2)+∑dijkri+11rj+12r12e−α′r1−β′r2YM112(ˆr1,ˆr2)
For F-states, one would need (sf)F and (pd)F terms.
For G-states, one would need (sg)G, (pf)G and (dd′)G terms.
Completeness of the radial functions can be proven by considering the Stern-Liouville problem
(−12∇2−λrs−E)ψ(r)=0 or (−121r2(r2∂∂r)−l(l+1)2r2−λr−E)u(r)=0. For fixed E and variable λ (nuclear charge).
The eigenvalues are λn=(E/En)1/2, where En=−12n2
unl(r)=1(2l+1)!((n+l)!(n−l−1)2!)1/2(2α)3/2e−αr,
with α=(−2E)1/2 and n≥l+1.
Unlike the hydrogen spectrum, which has both a discrete part for E<0 and a
continuous part for E>0, this forms an entirely discrete set of finite
polynomials, called Sturmian functions. They are orthogonal with respect to
the potential
-ie ∫∞0r2drun′l(r)1runl(r)=δn,n′
Since they become complete in the limit n→∞, this assures the
completeness of the variational basis set.
[see B Klahn and W.A. Bingel Theo. Chim. Acta (Berlin) 44,9 and 27(1977)].
Solutions of the Eigenvalue Problem
For convenience, write
Ψ(r1,r2)=∑Nm=1cmφm
where m=m′th combinations of i,j,k
φijk=ri1rj2rk12e−αr1−βr2YMl1,l2,L(ˆr1,ˆr2)±exchange.
(cos(θ)sin(θ)−sin(θ)cos(θ))(H11H12H12H22)(cos(θ)−sin(θ)sin(θ)cos(θ))=(cH11+sH12cH12+sH22−sH11+cH12−sH12+cH22)(c−ssc)=(c2H11+s2H22+2csH12(c2−s2)H12+cs(H22−H11)(c2−s2)H12+cs(H22−H11)s2H11+c2H22−2csH12)
Therefore (cos2(θ)−sin2(θ))H12=cos(θ)sin(θ)(H11−H22) and tan(2θ)=2H12H11−H22
ie
cos(θ)=(r+ω2r)1/2,sin(θ)=−sgn(H12)(r−ω2r)1/2
where
ω=H22−H11r=(ω2+4H212)1/2E1=12(H11+H22−r)E2=12(H11+H22+r)
Brute Force Method
-Gives all the eigenvalues and eigenvectors, but it is slow.
-First orthonormalize the basis set - ie form linear combinations
Φm=∑nφnRnm such that <Φm|Φn>=δm,n. This can be done by finding an orthogonal tranformation, T, such that
TTOT=I=( I1 0…00 I2…0 ⋮ I3… 00 0 0 ⋱); Omn=<φm|φn> and then applying a scale change matrix S=(1I1/210…001I1/22…0⋮ 1I1/2300 0 0 ⋱)=ST
Then STTTOTS=1. ie RTOR=1 with R=TS.
If H is the matrix with elements Hmn=<φm|φn>, then H expressed in the Φm basis set is H′=RTHR.
We next diagonalize H′ by finding an orthogonal transformation W such that WTH′W=λ=(λ10…00 λ2…0 ⋮ ⋱00 …λN)
The q'th eigenvector is
Ψ(q)=∑nΦnWn,q=∑n,n′φn′Rn′,nWn,q. ie. c(q)nprime=∑nRn′nWn,q.
The Power Method
-Based on the observation that if H has one eigenvalue, λM, much bigger than all the rest, and χ=(a1a2⋮) is an arbitrary starting vector, then χ=∑qxqΨ(q).
(H)nχ=∑qxqλnqΨ(q)→xMλnMΨ(M) provided xM≠0.
To pick out the eigenvector correspondng to any eigenvalue, with the original problem in the form
HΨ=λOΨ (H−λ)qO)Ψ=(λ−λq)OΨ
Therefore, GΨ=1λ−λqΨ where G=(H−λqO)−1O with eigenvalues 1λn−λq.
By picking λq close to any one of the λn, say λn′, then 1λn−λq is much larger for n=n′ than for any other value. The sequence is then
χ1=Gχχ2=Gχ1χ3=Gχ2⋮
until the ratios of components in χn stop changing.
- To avoid matrix inversion and multiplication, note that the sequence is equivalent to
Fχn=(λ−λq)Oχn−1
where F=H−λqO. The factor of (λ−λq) can be dropped because this only affects the normalization of χn. To find χn, solve Fχn=Oχn−1 (N equations in N unknowns). Then
λ=<χn|H|χn><χn|χn>
Matrix Elements of H
H=−12∇21−12∇22−1r1−1r2+Z−1r12
Taking r1,r2 and r12 as independent variables,
∇21=1r21∂∂r1(r21∂∂r1)+1r212∂∂r12(r212∂∂r12)−l1(l1+1)r21+2(r1−r2cos(θ))1r12∂2∂r1∂r−2(∇1⋅r2)1r∂∂r
where
The complete set of 6 independent variables is r1,r2,r12,θ1,φ1,χ.
If r12 were not an independent variable, then one could take the column element to be dτ=r21dr1sin(θ1)dθ1dφ1r22dr2sin(θ2)dθ2dφ2.
However, θ2 and φ2 are no longer independent variables. To eliminate them, take the point r1 as the
origin of a new polar co-ordinate system, and write dτ=−r21dr1sin(θ1)dθ1dφ1r212dr12sin(ψ)dψdχ
and use r22=r21+r212+2r1r12cos(ψ).
Then for fixed r1 and r12, 2r2dr2=−2r1r12sin(ψ)dψ
Thus dτ=r1dr1r2dr2r12dr12sin(θ1)dθ1dφ1dχ
The basic type of integral to be calculated is
I(l1,m1,l2,m2;R)=∫sin(θ1)dθ1dφ1dχYm1l1(θ1,φ1)∗Ym2l2(θ2,φ2)×∫r1dr1r2dr2r12dr12R(r1,r2,r12)
Consider first the angular integral. Ym2l2(θ2,φ2) can be expressed in terms of the independent variables θ1,φ1,χ by use of the rotation matrix relation
Ym2l2(θ2,φ2)=∑mD(l2)m2,m(φ1,θ1,χ)∗Yml2(θ,φ)
where θ,φ are the polar angles of r2 relative to r1. The angular integral is then
Iang=∫2π0dχ∫2π0dφ1∫π0sin(θ1)d=theta1Ym1l1(θ1,φ1)∗×∑mD(l2)m2,m(φ1,θ1,χ)∗Yml2(θ,φ)
Use
Ym1l1(θ1,φ1)∗=√2l1+14πD(l1)m1,0(φ1,θ1,χ)
together with the orthogonality property of the rotation matrices (Brink and
Satchler, p 147)
D(j)∗m,m′D(J)M,M′sin(θ1)dθ1dφ1dχ=8π22j+1δjJδmMδm′M′
to obtain
Iang=√2l1+14π8π22l1+1δl1,l2δm1,m2Y0l2(θ,φ)=2πδl1,l2δm1,m2Pl2(cosθ)
since
Y0l2(θ,φ)=√2l1+14πPl2(cos(θ)).
Note that Pl2(cosθ) is just a short hand expression for a radial function because cosθ=r21+r22−r2122r1r2.
The original integral is thus
I(l1,m1,l2,m2;R)=2πδl1l2δm1m2×∫∞0r1dr1∫∞0r2dr2∫r1+r2|r1−r2|r12dr12R(r1,r2,r12)Pl2(cosθ)
where cosθ=(r21+r22−r212)/(2r1r2) is a purely radial function.
The above would become quite complicated for large l2 because Pl2(cosθ) contains terms up to (cosθ)l2. However, recursion relations exist which allow any integral containing Pl(cosθ) in terms of those containing just P0(cosθ)=1
and P1(cosθ)=cosθ.
Radial Integrals and Recursion Relations
- The basic radial integral is I0(a,b,c)=∫∞0r1dr1∫∞r1r2dr2∫r1+r2r2−r1r12dr12ra1rb2rc12e−αr1−βr2+∫∞0r2dr2∫∞r2r1dr1∫r1+r2r1−r2r12dr12ra1rb2rc12e−αr1−βr2=2c+2∑[(c+1)/2]i=0(c+22i+1){q!βq+1(α+β)P+1∑j=0q(p+j)!j!(βα+β)j+q′!αq′+1(α+β)p′+1∑q′j=0(p′+j)!j!(αα+β)j}
where p=a+2i|2p′=b+2i+2q=b+c−2i+2q′=a+c−2i+2
The above applies for a,b ≥−2c≥−1. [x] means greatest integer in x.
Then I1(a,b,c)=∫dτrra1rb2rc12e−αr1−βr2P1(cosθ)=12(I0(a+1,b−1,c)+I0(a−1,b+1,c)−I0(a−1,b−1,c+2)).
The Radial Recursion Relation
Recall that the full integral is I(l1m1,l2m2;R)=2πδl1l2δm1m2Il2(R) where Il2(R)=∫dτrR(r1,r2,r12)Pl2(cosθ).
To obtain the recursion relation, use
Pl(x)=[P′l+1(x)−P′l−1(x)]2l+1 with P′l+1(x)=ddxPl+1(x).
Here x=cosθ and ddcosθ=dr12dcosθddr12=−r1r2r12ddr12.
Then, Il(R)=−∫dτrRr1r2r12ddr12[Pl+1(cosθ)−Pl−1(cosθ)]2l+1.
The r12 part of the integral is ∫r1+r2|r1−r2|r12dr12Rr1r2r12ddr12[Pl+1−Pl−1]=Rr1r2[Pl+1−Pl−1]r1+r2|r1−r2|−∫r1+r2|r1−r2|r12dr12(ddr12R)r1r2r12[Pl+1−Pl−1]2l+1
The integrated term vanishes because cosθ=r21+r22−r2122r1r2 =−1when r212=(r1+r2)2= 1when r212=(r1−r2)2 and Pl(1)=1, Pl(−1)=(−1)l.
Thus, Il+1(r1r2r12R′)=(2l+1)Il(R)−Il−1(r1r2r12R′)
If R=ra−11rb−12rc+212e−αr1−βr2, then Il+1(ra1rb2rc12)=2l+1c+2Il(ra−11rb−22rc+212)+Il−1(ra1rb2rc12)
For the case c=-2, take R=ra−11rb−12lnr12e−αr1−βr2. Then Il+1(ra1rb2rc12)=Il(ra−11rb−12lnr12)(2l+1)−Il−1(ra1rb2r−212). This allows all Il integrals to be calculated from tables of I0 and I1 integrals.
The General Integral
-The above results fro the angular and raidal integrals can now be combined into a general formula for integrals of the type I=∫∫dr1dr2R1YM′l′1l′2L′(ˆr1,ˆr2)TQk1k2K(r1,r2)R2YMlll2L(ˆr1,ˆr2)
where YMl1l2L(ˆr1,ˆr2)=∑m1,m2<l1l2m1m2|LM>Ym1l1(ˆr1)Ym2l2(ˆr2) and TQk1k2K(r1,r2)=∑q1,q2<k1k2q1q2|KQ>Yq1k1(ˆr1)Yq2k2(ˆr2)
The basic idea is to make repeated use of the formula Ym1l1(ˆr1)Ym2l2(ˆr2)=∑lm((2l1+1)(2l2+1)(2l+1)4π)1/2 ×(l1l2lm1m2m)(l1l2l000)Ym∗l(ˆr1)
where Ym∗(ˆr)=(−1)mY−ml(ˆr) and (l1l2lm1m2m)=(−1)l1−l2−m(2l+1)1/2(l1l2m1m2|l,−m) is a 3−j symbol.
In particular, write
Ym′2l′1(ˆr1)∗Yq1k1(ˆr1)Ym1l1(ˆr1)⏟=∑ΛM(…)YMΛ(ˆr1) ∑λ1μ1Yμ1λ1(ˆr1)Ym′l′2(ˆr2)Yq2k2(ˆr2)Ym2l2(ˆr2)⏟=∑Λ′M′(…)YM′∗Λ′(ˆr2) ∑λ2μ2Yμ2λ2(ˆr2)
The angular integral then gives a factor of 2πδΛ,Λ′δM,M′PΛ(cosθ12). The total integral therefore reduces to the form I=∑λCΛIΛ(R1R2), where cΛ=∑λ1λ2Cλ1,λ2,Λ.
Graphical Representation
Matrix Elements of H
Recall that H=−12∇21−12∇22−1r1−1r2+Z−1r12.
Consider matrix elements of ∇21=1r12(∂∂r1)+1r2(∂∂r)−(→ly1)2r21+2(r1−r2cosθ)r∂2∂r1∂r−2(∇y1⋅r2)1r∂∂r. where r≡r12 and cosθ≡cosθ12.
Also ˆ∇y1≡r1∇y1 operates only on the spherical harmonic part of the wavefunction.
A general matrix element is <ra′1rb′2rc′12e−α′r1− beta′r2YM′l′1l′2L′(ˆr1,ˆr2)|∇21|ra1rb2rc12e−αr1−βr2YMl1l2L(ˆr1,ˆr2)>.
Since ∇21 is rotationally invariant, this vanishes unless L=L′, M=M′. Also ∇21 is Hermitean, so that the result must be the same whether it operates to the right or left, even though the results look very different. In fact, requiring the results to be the same yields some interesting and useful integral identites as follows:
Put F=FYMl1l2L(ˆr1,^r2), and F=ra1rb2rc12e−αr1−βr2.
Then ∇21F=[1r21[a(a+1)−l1(l1+1)]+c(c+1)r2+α22α(a+1)r1+2(r1−r2cosθ)r1r2c[a−αr1]−2cr2(ˆ∇y1⋅ˆr2)r2r1]F
and
<F′|∇21|F>=∑Λ∫dτrF′CΛ(1)PΛ(cosθ){1r21[a(a+1)−l1(l1+1)]−2α(a+1)r1+c(c+1)r2+α2+2(r2−r2cosθ)r1r2c(a−αr1)}F+∑Λ∫dτrF′CΛ(ˆ∇y1⋅ˆr2)PΛ(cosθ)(−2cr2r1r2)F
where ∫dτr=∫∞0r1dr1∫infty0r2dr2∫r1+r2|r1−r2|rdr.
For brevity, let the sum over Λ and the raidal integrations be
understood, and let (∇21) stand for the terms which appear in the
integrand. Then operating to the right gives
(∇21)R=1r21[a(a+1)−l1(l1+1)]+c(c+1)r2+α2−2α(a+1)r1+2(r1−r2cosθ)r1r62c(a−αr1)−2cr−2r1r2(ˆ∇y1⋅ˆr2).
Operating to the left gives
(∇21)L=1r21[a′(a′+1)−l′1(l′1+1)]+c′(c′+1)r2+α2′−2α′(a′+1)r1+2(r1−r2cosθ)r1r2c′(a′−αr1)−2c′r−2r1r2(ˆ∇y′1⋅ˆr2).
Now put a+=a+a′,ˆ∇+1=ˆ∇y1+ˆ∇y′1 and a−=a−a′ ˆ∇−1=ˆ∇y1−ˆ∇y′1 etc.
and substitute a′=a+−a, c′=c+−c, and α′=α+−α in (∇21)L. If a+, c+ and α+ are held fixed, then the equation (∇21)R=(∇21)L must be true for arbitrary a, c, and α. Their coefficients must thus vanish.
This yields the integral relations (r1−r2cosθ)r1r2=1c+(−(a++1)r21+α+r1) from the coef. of a, and (r1−r2cosθ)(a+−α+r1)r1r2=r2r1r2(ˆr2⋅ˆ∇+1)−(c++1)r2 from the coef. of c. The coef. of α gives an equation equivalant to (I).
Furthermore, if can be show that (see problem) ∑Λ∫dτrrcr2CΛ(ˆr2⋅ˆ∇y1)PΛ(cosθ)=∑Λ∫dτrrccr1r2CΛ(1)PΛ(cosθ)×(l′1(l′1+1)−l1(l1+1)−Λ(Λ+1)2)
and similarly for (ˆr2⋅ˆ∇y′1) with l1 and l′1 interchangable, then it follows that rc+r2(ˆr2⋅ˆ∇+1)=−rc+c+r1r2Λ(Λ+1)
and rc+r2(ˆr2⋅ˆ∇−1)=rc+c+r1r2[l′1(l′1+1)−l1(l1+1)] where equality applies after integration and summation over Λ.
Thus (II) becomes (r1−r2cosθ)(a+−α+r1)r1r2=−Λ(Λ+1)c+r21−(c++1)r2.
Problem
Prove the integral relation ∑Λ∫dτrf(r1,r2)1r(ddrq(r))CΛ(ˆr2⋅ˆ∇y1)PΛ(cosθ)=∑Λ∫dτrf(r1,r2)r1,r2q(r)CΛ(1)PΛ(cosθ)×(l′1(l′1+1)−l1(l1+1)−Λ(Λ+1)2).
where CΛ(1) are the angular coefficients from the overlap integral ∫dΩYM∗l′1l′2L(ˆr1,ˆr2)=∑ΛCΛ(1)PΛ(cosθ12).
Hint: Use the fact that l21 is Hermitean so that ∫dτ(l21Y′)∗q(r)Y=∫dτY′∗l21(q(r)Y) with →l1=1i→r1×∇1.
It is also useful to use (cos2θ−1)PL(cosθ)=L(L−1)(2L−1)(2L+1)PL−2(cosθ)=2(L2+L−1)(2L−1)(2L+3)PL(cosθ)+(L+1)(L+2)(2L+1)(2L+3)PL+2(cosθ)
together with the integral recursion relation IL+1(1rddrq(r))=(2L+1)IL(1r1r2q(r))+IL−1(1rddrq(r))
Of course l21YMl1l2L(ˆr1,ˆr2)=l1(l1+1)YMl1l2L(ˆr1,ˆr2).
General Hermitean Property
Each combination of terms of the form <f>R=a2f1af2+abf3+b2f4+bf5+af6∇y1+bf7∇y2+f8(y) can be rewritten <f>L=(a+−a)2f1+(a+−a)f2+(a+−a)(b+−b)f3+(b+−b)2f4+(b+−b)f5+(a+−a)f6∇y′1+(b+−b)f7∇y′2+f8)y′
Since these must be equal for arbitrary a and b, a2+f1+a+f2++b+f3+b+f4+b+f5+a+f6∇y′1+b+f7∇y′2+f8(y′)−f8(y)=0
Adding the corresponding expresssion with y and $y^\prime$ interchanged yields a2+f1+a+f2+a+b+f3+b2+f4+b+f5+12a+f6∇+1+12b+f7∇+2=0
Subtracting gives f8(y)−f8(y′)=−12[a+f6∇−1+b+f7∇−2] a[−2a+f1−2f2−b+f3−f6=nabla+1]=0 b[−2b+f4−2f5−a+f3−f7∇+2]=0
Adding the two forms gives
<f>R+<f>L=12(a2++a2−)f1+a+f2+12(a+b++a−b−)f3 +12(b2++b2−)f4+b+f5+12f6(a+∇+1+a−∇−1) +12(b+∇+2+b−∇−2)+f8(y)+f8(y′)
Subtracting x2×(I) gives
<f>R+<f>L=12[(1−x)a2++a2−]f1+(1−x2)a+f2+12[(1−x)a+b++a−b−]f3+12[(1−x)b2++b2−]f4+(1−x2)f5b++12f6[(1−x2)a+∇+1+a−∇−1]+12[(1−x2)b+∇+2+b−∇−2]+f8(y)+f8(y′)
If x=1, <f>R+<f>L=12[a2−f1+a+f2+a−b−f3+b2−f4+b+f5+(12a+∇+1+a−∇−1)f6+(12b+∇+2+b−∇−2)f7]+f8(y)+f8(y′)
The General Hermitean Property for arbitrary x gives ∇21=14{1r21[(1−x)a2++a2−+2(1−x2)a+−2[l1(l1+1)+l′1(l′1+1)]−2r1[(1−x)α+a++α−a−+2(1−x2)α+]+(1−x)α2++α2−]+2(r1−r2cosθ)r1r2[(1−x)(a+−α+r1)c++(a−−α−r2)c−]−2r2r1r2[(1−x2)c+ˆr2⋅ˆ∇+1+c−ˆr2ˆ∇−1]+(1−x)c2++c2−+2(1−x2)c+r2}.
Use \begin{eqnarray}\ \ \frac{-2r_2}{r_1r^2}[(1-\frac{x}{2})c_+\hat{r}_2\hat{\nabla}_1^++c_-\hat{r}_2\hat{\nabla}_1^-]=\frac{2}{r_1^2}\left[(1-\frac{x}{2})\Lambda(\Lambda+1)-\frac{c_-}{c_+}[l_1^\prime(l_1^\prime+1)+l_1(l_1+1)]\right]\end{eqnarray},
\begin{eqnarray} \frac{2(r_1-r_2\cos\theta)}{r_1r^2}(a_--\alpha_-r_1)c_- = 2\frac{c_-}{c_+}\left[\frac{-1}{r_1^2}[a_-(a_++1)]+\frac{1}{r_1}[a_-\alpha_++\alpha_-(a_++2)]-\alpha_-\alpha_+\right]\nonumber \end{eqnarray},
and \begin{eqnarray}\ \ \frac{2(r_1-r_2\cos\theta)}{r_1r^2}(a_+-\alpha_+r_1)c_+ = 2\left[-\frac{1}{r_1^2}[a_+(a_++1)]+\frac{1}{r_1}[a_+\alpha_++\alpha_+(a_++2)]-\alpha_+^2\right] \end{eqnarray}.
Substituting into \nabla_1^2 gives
\begin{eqnarray} \nabla^2_1 = \frac{1}{4}\{\frac{1}{r_1^2}[-(1-x)a_+^2 + a_-^2 +xa_+ + 2(1-\frac{x}{2})\Lambda(\Lambda+1)\nonumber\\ -2l_1(l_1+1)(1-\frac{c_-}{c_+}) - 2l_1^\prime(l_1^\prime+1)(1+\frac{c_-}{c_+})-\frac{2c_-a_-}{c_+}(a_++1)]\nonumber\\ - \frac{2}{r_1}\left[-(1-x)\alpha_+(a_++2)+\alpha_-a_-+2(1-\frac{x}{2})\alpha_+ -\frac{c_-}{c_+}[a_-\alpha_++\alpha_-(a_++2)]\right]\nonumber\\ -(1-x)\alpha_+^2+\alpha_-^2-2\frac{c_-}{c_+}\alpha_-\alpha_+ +\frac{1}{r^2}\left[(1-x)c_+^2+c_-^2+2(1-\frac{x}{2})c_+\right]\}\nonumber \end{eqnarray}
This has the form \nabla_1^2 = \frac{1}{4}\left[\frac{A_1}{r_1^2}+\frac{B_1}{r_1}+\frac{C_1}{r^2}+D_1\right] with
\begin{array}{ccc} A_1=-(1-x)a_+^2+a_-^2 +xa_++2(1-\frac{x}{2})\Lambda(\Lambda+1)-2l_1(l_1+1)(1-\frac{c_-}{c_+})\ -2l_1^\prime(l_1^\prime+1)(1+\frac{c_-}{c_+})-\frac{2c_-a_-}{c_+}(a_++1) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \end{array}
B_1 = 2\left[(1-x)\alpha_+(a_++2)-\alpha_-a_--2(1-\frac{x}{2})\alpha_++\frac{c_-}{c_+}[a_-\alpha_++\alpha_(a_++2)]\right]
C_1 = (1-x)c_+^2+ c_-^2 +2(1-\frac{x}{2})c_+
D_1 = -1(1-x)\alpha_+^2 +\alpha_-^2 - 2\frac{c_-}{c_+}\alpha_-\alpha_+
The complete Hamiltonian is then (in Z-scaled a.u.)
\begin{eqnarray} H =-\frac{1}{2}\nabla^2_1 -\frac{1}{2}\nabla^2_2 -\frac{1}{r_1} -\frac{1}{r_2} +\frac{Z^{-1}}{r}\nonumber\\ = -\frac{1}{2}\nabla^2_1 -\frac{1}{2}\nabla^2_2 -\frac{1}{r_1} -\frac{Z-1}{Zr_2}+Z^{-1}\left(\frac{1}{r}-\frac{1}{r_2}\right)\nonumber\\ \ =-\frac{1}{8}\left[\frac{A_1}{r_1^2}+\frac{B_1+8}{r_1}+\frac{C_1}{r^2}+D_1+D_2 +\frac{A_2}{r_2^2} + \frac{B_2 + 8(Z-1)/Z}{r_2}+\frac{C_2}{r^2}\right]\nonumber + Z^{-1}\left(\frac{1}{r}-\frac{1}{r_2}\right)\nonumber \end{eqnarray}
The screened hydrogenic energy is \begin{eqnarray} E_{SH}=-\frac{1}{2}\left[\frac{1}{n_1^2}+ \left(\frac{Z-1}{Z}\right)^2\frac{1}{n_2^2}\right] \end{eqnarray} so that
\begin{eqnarray} H-E_{SH} &=& -\frac{1}{8}[ \frac{A_1}{r_1^2}+\frac{B_1+8}{r_1}+\frac{C_1}{r^2} + D_1 + D_2 -\frac{4}{n_1^2} -\left(\frac{Z-1}{Z}\right)^2\frac{4}{n_2^2}\nonumber +\frac{A_2}{r_2}+\frac{B_2+8(Z-1)/Z}{r_2}+\frac{C_2}{r^2}] + Z^{-1}\left(\frac{1}{r} -\frac{1}{r_2}\right)\nonumber \end{eqnarray}
Optimization of Non-linear Parameters
- The traditional method of performing Hylleraas calculations is to write the basis set in the form
\Psi = \sum_{i,j,k}c_{i,j,k}\varphi_{i,j,k}(\alpha,\beta)\pm exchange
with
\varphi_{i,j,k}(\alpha,\beta) = r_1^ir_2^jr_{12}^ke^{-\alpha r_1 - \beta r_2}\mathcal{Y}^M_{l_1l_2L}.
The usual procedure is to set \alpha = Z so that it represents the inner 1s electron, and then to vary \beta so as to minimize the energy.
Since \beta appears in \Psi as a non-linear parameter, the enitre calculation must be repeated for each value of \beta. However, the minimum becomes progressively smaller as the basis set is enlarged.
- Difficulties
1. If the basis set is constructed so that i +j +j \leq N, the the number of terms is (N+1)(N+2)(N+3)/6 N=14 already gives 680 terms and an accuracy of about 1 part in 10^10 for low-lying states. A substantial improvement in accuracy would require much larger basis sets, together with multiple precision arithmetic to avoid loss of significiant figures when high powers are incluuded.
2. The accuracy rapidly deteriorates as one goes to more highly excited states - about 1 significant figure is lost each time the principle quantum number is increased.
- Cure
We have found that writing basis sets in the form \begin{eqnarray} \Psi &=& \sum_{i,j,k}\left(c^{(1)}_{i,j,k}\varphi_{ijk}(\alpha_1\beta_1) + c_{ijk}^{(2)}\varphi_{ijk}(\alpha_2\beta_2)\right) \pm exchange \nonumber\\ &=& \psi({\bf r}_1,{\bf r}_2) \pm \psi({\bf r}_2,{\bf r}_1)\nonumber \end{eqnarray} so that each combination of powers is includd twice with different non-linear parameters gives a dramatic improvement in accuracy ffor basis sets of about the same total size.
However, the optimization of the non-linear parameters is now much more difficult, and an automated procedure is needed.
If E = \frac{<\Psi|H|\Psi>}{<\Psi|\Psi>},
then in <\Psi|\Psi> = 1, \frac{\partial E}{\partial\alpha_t} = -2 <\Psi|H-E|r_1\psi({\bf r}_1,{\bf r}_2;\alpha_t)\pm r_21\psi({\bf r}_2,{\bf r}_1;\alpha_t)> where \psi({\bf r}_1,{\bf r}_2;\alpha_t) = \sum_{i,j,k}c_{ijk}^{(t)}\varphi_{ijk}(\alpha_t\beta_t)
- Problem
1. Prove the above.
2. Prove that there is no contribution from the implicit dependence of c_{ijk}^{(t)} on \alpha_t if the linear parameters have been optimized.
The Screened Hydrogenic Term
If the Hamiltonian is written in the form (in Z-scaled a.u.) \begin{eqnarray}\ H = H_0({\bf r}_1,Z) + H_0({\bf r}_2,Z-1) + Z^{-1}\left(\frac{1}{r}-\frac{1}{r_2}\right) \end{eqnarray} with \ H_0({\bf r}_1,Z) = -\frac{1}{2}\nabla_1^2 - \frac{1}{r_1}\nonumber and H_0({\bf r}_2,Z-1) = -\frac{1}{2}\nabla_2^2 - \frac{Z-1}Z{r_2}\nonumber , then the eigenvectors of H_0({\bf r},Z) +H_0({\bf r}_2,Z-1) are products of hydrogenic orbitals \Psi_0=\psi_0(1s,Z)\psi_0(nl,Z-1) and the eigenvalue is \begin{eqnarray} E_{SH} = \left[-\frac{1}{2} - \left(\frac{Z-1}{Z}\right)^2\frac{1}{2n^2}\right]Z^2 a.u. \end{eqnarray} called the screened hydrogenic eigenvalue. From highly excited states, E_{SH} and \Psi_0are already excellent approximations.
For example, for the 1s8d states, the energies are
\begin{eqnarray}\\ E(1s8d^1D)= -2.00781651256381 a.u.\nonumber\\ E(1s8d^3D)= - 2.00781793471171 a.u.\nonumber\\ E_{SH} = -2.0078125\nonumber \end{eqnarray}
It is therefore advantageousto include the screened hydrogenic terms in the basis set so that the complete trial function becomes \Psi = c_0\Psi_0 + \sum_{ijk}\left[c_{ijk}^{(1)}\varphi_{ijk}(\alpha_1,\beta_1)+c_{ijk}^{(2)}\varphi(\alpha_2,\beta_2)\right]\pm exchange
Also, the variational principal can be re-expressed in the form \begin{eqnarray} E= E_{SH}+\frac{<\Psi|H-E_{SH}|\Psi>}{<\Psi|\Psi>} \end{eqnarray} so that the E_{SH} term can be cancelled analytically from the matrix elements.
For example <\Psi_0|H-E_{SH}|\Psi_0> = 0
and <\varphi_{ijk}|H-E_{SH}|\Psi_0> = Z^{-1}<\varphi_{ijk}|\frac{1}{r}-\frac{1}{r_2}|\Psi_0>
Recall that \begin{eqnarray}\ \ I_0(a,b,c) = \frac{2}{c+2}\sum^{[c+1]/2}_{i=0}\left(\begin{array}{c} c+2\\ 2i+1 \end{array}\right) \left[f(p,q;\beta) + f(p^\prime,q^\prime;\alpha)\right] \end{eqnarray} where
\begin{eqnarray} f(p,q;x) = \frac{q!}{x^{q+1}(\alpha+\beta)^{p+1}}\sum^p_{j=0}\frac{(p+j)!}{j!}\left(\frac{x}{\alpha+\beta}\right)^j \end{eqnarray}
\begin{array}{cc} p=a+2i+2 & p^\prime = b+2i+2\\ q= b+c-2i+2 & q^\prime =a+c-2i+2 \end{array}
If \beta\ll\alpha, the f(p,q;\beta)\gg f(p^\prime,q^\prime;\alpha) and the i=0 term is the dominant contribution to f(p,q;\beta). However, since this term depends only on the sum of powers b+c for r_2, it cancels exactly from the matrix element of \begin{eqnarray}\frac{1}{r_12} - \frac{1}{r_2}\end{eqnarray} and can therefore be omitted in the calculations of integrands, there by saving many significant figures. This is especially valuable when b+c is large and a is small.
Small Corrections
- Mass Polarization
If the nuclear mass is not taken to be infinite, then the Hamiltonian is \begin{eqnarray}\ \ H = \frac{P_N^2}{2M} + \sum_{i=1}\frac{p_i^2}{2m} + V \end{eqnarray}
where M =M_A-nm is the nuclear mass and m is the electron mass.
Change variables to the C of M \begin{eqnarray} {\bf R} = \frac{1}{M+nm}\left[M{\bf r}_N + m({\bf r}_1 + {\bf r}_2 + \ldots)\right] \end{eqnarray}
and relative variables
{\bf s}_i = {\bf r}_i - {\bf r}_N.
Then
\begin{eqnarray}\ \ \ \ \ H = \frac{1}{2(M+nm)}{\bf p}_R^2+\frac{1}{2m}\sum_{i=1}^n{\bf p}_{s_i}^2 +\frac{1}{2M}\sum_{i,k}{\bf p}_{s_i}\cdot {\bf p}_{s_k} + V({\bf s}_1, {\bf s}_2, \ldots)\nonumber = \frac{1}{M+nm}{\bf p}_R^2+\frac{1}{2\mu}\sum_{i=1}^n{\bf p}_{s_i}^2+\frac{1}{2M}\sum_{i\neq k}{\bf p}_{s_i}\cdot {\bf p}_{s_k}+V({\bf s}_1, {\bf s}_2, \ldots)\nonumber \end{eqnarray}
where \begin{eqnarray} \mu = \left(\frac{1}{M} + \frac{1}{m}\right)^{-1} \end{eqnarray}
If n=2, the Schroedinger equation is \begin{eqnarray}\ \ \ \left[ -\frac{\hbar^2}{2\mu}(\nabla^2_{s_1}+ \nabla^2_{s_2}) - \frac{\hbar^2}{M}\nabla_{s_1}\nabla_{s_2}-\frac{Ze^2}{s_1}-\frac{Ze^2}{s_2}+\frac{e^2}{s_12}\right]\psi = E\psi \end{eqnarray}.
Define the reduced mass Bohr radius \begin{eqnarray} a_\mu = \frac{hbar^2}{\mu e^2} = \frac{m}{\mu}a_0 \end{eqnarray} and
\begin{eqnarray} \rho_i = \frac{Zs_i}{a_\mu}\nonumber\\ \varepsilon = \frac{E}{Z^2(e^2/a_\mu)}\nonumber \end{eqnarray}
The Schroedinger equation then becomes \begin{eqnarray}\ \ \ \left[ -\frac{1}{2}(\nabla^2_{\rho_1}+\nabla^2_{\rho_2})-\frac{\mu}{M}\vec{\nabla}_{\rho_1}\cdot\vec{\nabla}_{\rho_2} - \frac{1}{\rho_1}-\frac{1}{\rho_2}+\frac{Z^{-1}}{\rho_{12}}\right]\psi = \varepsilon \psi \end{eqnarray}
The effects of finite meass are thus
1. A reduced mass correction to all energy levels of (e^2/a\mu)/(e^2/a_0) = \mu/m, ie. \begin{eqnarray} E_M=\frac{\mu}{m}E_\infty \end{eqnarray}
2. A specific mass shift given in first order perturbation theory by \begin{eqnarray} \Delta\varepsilon = -\frac{\mu}{M}<\psi|\nabla_{\rho_1}\cdot\nabla_{\rho_2}|\psi> \end{eqnarray} or
\begin{eqnarray} \Delta\varepsilon = -\frac{\mu}{M}<\psi|\nabla_{\rho_1}\cdot\nabla_{\rho_2}|\psi>\frac{e^2}{a_\mu}\nonumber\\ = -\frac{\mu^2}{mM}<\psi|\nabla_{\rho_1}\cdot\nabla_{\rho_2}|\psi>\frac{e^2}{a_0}\nonumber \end{eqnarray}
Since \nabla_{\rho_1}\cdot\nabla_{\rho_2} has the same angular properties as {\bf\rho}_1 \cdot {\bf\rho}_2, the operator is like the product of two dipole operators. For product type wave functions of the form \psi = \psi(1s)\psi(nl) \pm exchange the matrix element vanishes for all but p-states.