Professional Documents
Culture Documents
R. M. Wilcox
Baker-Campbell-Hausdorff formulas
Journal of Mathematical Physics 15, 2090 (1974); https://doi.org/10.1063/1.1666587
Although everyone knows how to differentiate an differential equation method are used to obtain
ordinary exponential function, it is not widely realized closed-form expressions for Z = In (eAe B ) for certain
that a general closed formula exists for the derivative cases where A and Bare LCLE's. Generalizations of
of an exponential operator with respect to a parameter. formulas obtained by Sackl l and by Weiss and
[See Eq. (2.1) of Sec. 2.] We believe that this formula Maradudin 6 are obtained. A corollary to a theorem
is sufficiently simple and useful that many physicists of Bloch12 is also derived which is useful for obtaining
and applied mathematicians can profitably commit harmonic-oscillator thermal averages of general
it to memory. In Sec. 2, this formula is verified by the operator functions.
differential equation method and compared with In Sec. 8, we obtain solutions to the important
formulas of Feynman1 and Kubo. 4 A basic lemma is operator differential equation, dY(t)/dt = A(t) Yet).
derived which displays the formal correspondence In Sec. 8.1, we obtain the Magnus 7 and Fer13 solutions
between parameter differentiation and commutation. in a manner similar to that used to derive the BCH
The concept of differentiation with respect to an and Zassenhaus formulas, respectively. It should be
operator is also discussed. In Sec. 3, Eq. (2.1) is shown noted that all four of these formulas are derived by a
to be of use for problems in equilibrium quantum recursion procedure which manifestly displays the
statistics. In Sec. 4, Eq. (2.1) and the differential repeated-commutator nature of these expansions and
equation method are used to easily derive the Baker- which is quite suitable for computer programming.
Campbell-Hausdorff (BCH) formula. 5 This important An expansion recently obtained by Kumar14 and
formula determines Z such that eAe B = eZ is identi- another new expansion are shown to be derivable
cally satisfied. A good review of the history of this from the Fer and Magnus solutions, respectively, in
formula is given by Weiss and Maradudin. 6 A formula the same way. Both of these expansions may be more
of Zassenhaus, 7 said to be the dual of the BCH suitable than the Zassenhaus expansion for certain
formula, is also derived in Sec. 4 in a similar manner. purposes. In Sec. 8.2, Lie algebraic solutions in terms
This formula expresses eA +B as an infinite product of a single exponential are obtained and compared
of exponential operators. A number of examples with corresponding multiple-exponential solutions of
of the type of problem which exploits the BCH Wei and Norman. 15 In Sec. 9, solutions of the density
formula occur in connection with the Weyl prescrip- matrix equation, Eq. (9.1), are discussed. It is shown
tion for converting a classical function to a quantum that an initially canonical ensemble behaves as though
operator. This is described in Sec. 5. A more direct its temperature remains constant with a "canonical
derivation of a useful formula recently obtained by distribution" determined by a certain fictitious
Daughaday and Nigam8 is given there, while inter- Hamiltonian.
esting theorems of McCoy,9 Moyal, and Wigner10 are In Sec. 10, a collection of old and recent results
stated without proof. involving normal ordering of operators are efficiently
In Sec. 6, a definition of a Lie algebra is given and derived with the aid of the results and methods of
examples of various Lie algebras which occur in previous sections. Derivations, corollaries, or general-
quantum mechanics are given. Similarity transforma- izations of formulas obtained by Louisell,3 Heffner
tions involving operators which are linear combi- and Louisell,16 Schwinger,17 McCoy,18 Kermack and
nations of Lie elements (denoted LCLE) are readily McCrea,19 and Cohen20 are given. In Sec. 11, higher
obtained by the differential equation method. The derivatives of exponential operators are treated with a
results may be used to "change the representation" in formula due to Poincare. 21 [See Eq. (10.1).] Although
quantum-mechanical problems, and are also used
11 R. A. Sack, Phil. Mag. 3, 497 (1958).
in Secs. 7, 8, and 10. In Sec. 7, Eq. (2.1) and the 12 F. Bloch, Z. Physik 74, 295 (1932).
13 F. Fer, Bull. Classe Sci. Acad. Roy. Belg. 44, 818 (1958).
• Lectures in Theoretical Physics R. Kubo, W. E. Brittin, and L. 14 K. Kumar, J. Math. Phys. 6, 1928 (1965).
G. Dunham, Eds. (Interscience Publishers, Inc., New York, 1959), 1. J. Wei and E. Norman, J. Math. Phys. 4, 575 (1963).
Vol. I, p. 139, Eq. (2.17). 16 H. Heffner and W. H. Louisell, J. Math. Phys. 6, 474 (1965).
5 J. E. Campbell, Proc. London Math. Soc. 29, 14 (1898); H. F. 17 Quantum Theory of Angular Momentum, J. Schwinger, L. C.
Baker, ibid. 34, 347 (1902); 35, 333 (1903); 2, 293 (1904); 3, 24 (1904); Biedenharn, and H. van Dam, Eds. (Academic Press Inc., New
F. Hausdorff, Ber. Verhandl. Saechs. Akad. Wiss. Leipzig, Math.- York, 1965), pp. 274-276.
Naturw. KI. 58, 19 (1906); N. Jacobson, Lie Algebras (Interscience 18 N. H. McCoy, Proc. Edinburgh Math. Soc. 3, 118 (1932).
Publishers, Inc., New York, 1962), Chap. 5, p. 170. 19 w. O. Kermack and W. H. McCrea, Proc. Edinburgh Math.
6 G. H. Weiss and A. A. Maradudin, J. Math. Phys. 3, 771 (1962). Soc. 2, 220 (1931).
7 W. Magnus, Commun. Pure Appl. Math. 7, 649 (1954). 20 L. Cohen, J. Math. Phys. 7, 244 (1966).
8 H. Daughaday and B. P. Nigam, Phys. Rev. 139, BI436 (1965). .. H. Poincare, Trans. Cambridge Phil. Soc. 18, 220 (1899); see
9 N. H. McCoy, Proc. Math. Acad. Sci. U.S. 18, 674 (1932). also R. Bellman, Introduction to Matrix Analysis (McGraw-Hili
10 J. E. Moyal, Proc. Cambridge Phil. Soc. 45, 99 (1949); E. Book Company, Inc., New York, 1960), p. 103, Ex. 43. This formula
P. Wigner, Phys. Rev. 40, 749 (1932); C. L. Mehta, J. Math. Phys. is the basis of resolvent theory. See, e.g., Messiah, Ref. 48, Chap. 16,
5, 677 (1964). Sec. 3, p. 712.
964 R. M. WILCOX
even greater formal efficiency may be obtained A special case ofEq. (2.1) is an identity of F eynman, 27
with the Feynman ordering-operator calculus,1 we
have preferred to use the Poincare formula since it is
more firmly rooted in classical analysis and since it
[!!:..
de
ea+<p]
<=0
= (1eU-s)apesa ds,
Jo
(2.4)
readily permits one to differentiate general operator where ex and P are independent of e. Equation (2.4)
functions in a concise manner. It is shown to provide has been obtained by Feynman by means of his
a compact alternative for the derivation of formulas ordering-operator calculus, and this method may
of the type derived by Aizu. 2 We note that such for- also be used to derive Eq. (2.1).28 Also, Eq. (2.1) may
mulas may be used to derive all of the interesting sum be obtained from Eq. (4) by considering the Taylor
rules and hypervirial theorems treated in a recent paper expansion
by Morgan and Lafidsberg. 22
H(A + e) = H(A) + e(oHloA) + O(e2).
2. DERIVATIVE OF EXPONENTIAL However, Eq. (2.1) is a more convenient form to use
OPERATOR
than Eq. (2.4), particularly for differential equations.
If the operator H is a function of a parameter A, An identity of Kubo is a corollary of Eq. (2.1).4
H == H(A), then
.! e-PH = _ (Pe-(P-u)H oH e-uH duo (2.1)
[A, e-PJe ] = -J: due-(P-u)Je[A, Je]e- uJe . (2.5)
Another kind of differentiation which frequently to an unperturbed part Jeo , so that Je = Jeo + AV,
occurs in quantum physics is the differentiation of an then Eq. (3.2) becomes
operator by an operator. This may be defined by
means of parameter differentiation as follows. 31 Let o(V) = _ rPdu(euJeA Ve-uJeA V). (3.3)
HE H(Ql, Q2' ... ,QJ be a function of the
0), Jo
operators Ql' Q2' ... , Qn which need not commute Assuming, for notational convenience, discrete energy
with each other. Then the operator derivative with levels Jet or Je i , one easily obtains
respect to Qi is defined by
o(V) = -(Tr e-PJer1 ! I(il AV 1i>12
oH . oH
oQj E ~~ oA (Ql'···' Q; + A,···, Qn)·
-
oQ;
PH
oe-- = - i Pd ue-<P-u)H -
0
oRe
OQi
-uH
. (2.8)
which is seen to be nonpositive. 33 This result is the
quantum-statistical analog of the well-known result
that the second-order perturbation energy of the
ground state is always negative. The latter result may
The above basic lemma is also clearly valid with
be obtained from Eq. (3.4) as a special case by evalu-
A replaced by Qi. To obtain a familiar result which
ating the integral in Eq. (3.4) and letting (3 become
we make use of later, let A = q, H = p, A = p/in, so
arbitrarily large.
that [q,p] = ili'= inop/op. Then by the lemma,
Another application of Eq. (3.2) is to a system with
[q,f(p)] = inof(p)/op. (2.9) dipole moment M subjected to an applied field E, so
that it is described by the Hamiltonian
Lettingf(p) =r i I'P/II, Eq. (2.9) becomes
Je = Je o - M • E. (3.5)
(2.10) The field-dependent isothermal static susceptibility
From Eq. (2.10), it follows by a simple standard tensor X..p(E) for a sample of unit volume is defined by
argument that if Iq') is an eigenstate of q so that X..p(E) E o(M.. )/oEp , (3.6)
q Iq') = q' Iq'), then r i l'P/1i acts as a displacement
operator,32 where ~, (3 = x, y, or z. It follows from Eqs. (2) and
e-iJJ 'P/lIlq') = Iq' + #). (5) that
obtained from Eq. (2.1) by substituting Z == Z(A) for Equation (12) may be simplified by means of Eqs.
-H, 1 for p, and 1 - x for u. We seek to express Z (4.6), (4.9), and (4.11) to obtain
as a power series in A such that Z3i\[A, [A, B]] + i'-2[[A, B], B].
= (4.13)
(4.2) The BCH formula to third order is obtained by
substituting Eqs. (4.3), (4.9), (4.11), and (4.13) into
is identically satisfied. Thus,
Eq. (4.2) and setting A = 1:
00 00
Z Z'(A) =!nA n- 1z n ,
= !Anz n , (4.3) eAeB = exp {A + B + UA, B] + i\[A, A, B]
n=1 n=1
where the Zn are to be determined. Note that Z = 0
+ i'-2[[A,B],B] + ... }. (4.14)
The recursion scheme based upon Eq. (4.8) may, in
when A = 0 as required by Eq. (4.2). Differentiating
principle, be carried out to art)itrarily high order.
Eq. (4.2) with respect to A and multiplying both sides
Weiss and Maradudin have manually calculated Z
from the right by e-z = e-)"Be-)"A, one obtains
out to the fifth order, 6 while Richtmyer and Greenspan
have calculated Z out to the tenth order by computer. 38
fdxea:ZZ'(A)e-a:Z = A + e)..ABe-)"A. (4.4)
The expansion is not unique due to the existence of
the Jacobi identity
The quantity eAA Be- AA is easily expanded by Eq. (2.2)
in the form [[A, B], C] + [[C, A], B] + [[B, C], A] = 0,
00
eAABe-)"A = !Ai(j!rl{A 1, B}, (4.5) the identity
1=0
[[[A, B], C], D] + [[[B, C], D], A] + [[[C, D], A], B]
where the repeated commutator bracket is defined
inductively b y 37
+ [[[D, A], B], C] = [[A, C], [B, D]],
and others.
{AO, B} = B, {An+l, B} = [A, {An, B}]. (4.6) It frequently happens that the commutator of A
and B commutes with both A and B. In this case, Eq.
Similarly, since f~ Xi dx = l/(j + 1), Eq. (4.5) implies (4.14) reduces to
that
{Zk Z'} eAeB = eA+B+!c = eA+Be!c = etceA+B , (4.15)
!00
i
l
dxea:zZ' e-a:z = ' . (4.7)
o k=O (k + I)! where C == [A, B]. Equation (4.15) has been derived
in various ways and is frequently employed in physical
Substituting Eqs. (4.3), (4.5), and (4.7) into Eq. (4.4),
problems. sD Another form of Eq. (4.15), which has
we obtain
been used by Moyal and by Sudarshan, is10.40
!oo{ +
1 (00 )koo
! AmZ m' ! nAn- I
Z } (4.16)
k=O (k I)! m=l n=l n
00 Equation (4.16) follows from Eq. (4.15) if one inter-
= A +1=0
!Ai(j!rl{Ai, B}. (4.8) changes the order of A and Bin Eq. (4.15),
An
CJ) 00
CJ) Am+n where g(a, T) is the Fourier transform of the classical
o = L - {An, B} + 2A L L - {Am, B n, C2} function f(p, q). The generalization of the Weyl
n=l n! m=O n=O m! n!
00 00 00 Ak+m+2n prescription to the case of n independent pairs of
+ 3A 2 L L L {Ak, B m , C~, Cs } + .... canonical variables has been considered by Daughaday
k=O m=O n=O k! m! n! and Nigam (DN). 8 In effect, this amounts to inter-
(4.20) preting the quantities occurring in Eq. (2) as
The quantities {An, B} are defined by Eq. (6), while
the quantities {Am, Bn, C2 } are defined inductively as P = (P l ,P2,··· ,Pn),
{AO, Bn, C 2 } = {Bn, C2 }, (4.21a) a = (aI' a2'···' an),
da = da l da 2 ·• . dan,
{Am+!, Bn, C2} = [A, {Am, Bn, C2 }].
(4.21b)
Higher-order repeated-commutator brackets are simi-
aP = alPl + a2P 2 + ... + auP.. ,
larly defined. Clearly the coefficients of Ai must vanish etc.
in Eq. (20). Settingj = 1, one obtains
By a rather tortuous process, DN obtain the useful
C2 = -![A, B]. (4.22) result that the quantum function F(P, Q) corre-
Settingj = 2, one obtains sponding to the classical function
0= HA 2, B} + 2{A, BO, C 2} + 2{AO, B, C2} f(p, q) = p~pg ... P~P(ql' q2' ... , qn)
+ 3{AO, BO, C~, Cs} is given by8
or, upon using Eqs. (4.6), (4.21), and (4.22), 2-(cr+P+· .. +Y)[ . .. [[m P ] P] ... P ]
T' 1 +cr' 2 +P n +Y' (5.3)
Cs = ![B, [A, B]] + i[A, [A, B]]. (4.23)
where [p, Pl]+cr refers to the anticommutator bracket
We note that both the BCH formula and the repeated ot times. For example,
Zassenhaus formula could have been derived a little
more simply by repeatedly differentiating Eqs. (4.2) [p, Ptl+2 = [pP l + PIP, Pl ]+l
and (4.18), respectively, and setting A = 0 after each = ppi + 2Pl PP l + pip·
differentiation. Although Zn and C n may be obtained
in a recursive manner by this procedure, this method In DN, Eq. (3) has been used to obtain the quantum
does not manifestly display the general repeated- Hamiltonian for the case of two charged spinless
commutator nature of the expansion. This method particles interacting by retarded fields. We show here
has been used by Huang41 to obtain the first three that Eq. (3) may be derived much more directly with
factors in the Zassenhaus formula. Huang has used the aid of Eq. (4.16) written in the form
the result to treat quantum deviations from the ei(O"P+TQ) = ihO"PeiTQehaP. (5.4)
classical limit of the partition function for both a
Bose-Einstein and a Fermi-Dirac gas. Equation (5.4) is valid, since aP and TQ both commute
with their commutator. The Fourier transform of
5. WEYL'S PRESCRIPTION FOR QUANTIZING
A CLASSICAL FUNCTION f(p, q), g(a, T), is easily found to be given by
Let p, q denote a conjugate pair of canonical vari- i cr+P+·· ·+Y15(cr)(al )15(P)(a2)· . ·15(Y)(an)u(T), (5.5)
ables of classical mechanics, and let P, Q denote the
corresponding quantum operators,42 where, e.g., 15(cr)(a) denotes the otth derivative of the
"Dirac 15 function" and U(T) is the Fourier transform
[Q, P] = iii. (5.1)
of p(q). One easily obtains Eq. (5.3) by substituting
41 K. Huang, Statistical Mechanics (John Wiley & Sons, Inc.,
New York, 1963), p. 217, Eq. (10.60) . 43 H. Weyl, The Theory of Groups and Quantum Mechanics (E. P.
.. Only in Sec. 5 do small p and q denote classical variables. Dutton & Co., Inc., New York, 1931), p. 274.
968 R. M. WILCOX
Eqs. (5.4) and (5.5) into Eq. (5.2), and using the well- Theorem (WignerlO): Let G(P, Q) be any operator
known formula of distribution theory,44 defined in the Weyl manner, and let
We state without proof some other interesting (Q) may be calculated from the phase-space distri-
results involving the Weyl prescription which may be bution function I(p, q) by means of
dt!rived with the aid of the BCH formula and the
Fourier integral theorem.
(G) = II dp dqg(p, q)f(p, q), (5.10)
Theorem (MCCoy9): Let F(P, Q) be the quantum where
operator which corresponds to the classical function
I(p, q) according to the Weyl prescription, and let f(p, q) == -27T I
1 w*(q - tnT)e-lTPW(q + lnT) dT, (5.11) .
FQ(P, Q) be the function obtained from F(P, Q) by
ordering all Q factors to the left of all P factors with with w(q) == (q /w).
the aid of Eq. (5.1). Then FQ(P,Q) may be obtained by As shown by Moyal,lO this result may be easily
replacing p and q by P and Q with the Q's to the left proved with the aid of Eq s. (2.11), (4.16), and the
of the P's in all terms of the power-series expansion of Fourier integral theorem.
parameter x into the exponentials and determine The solution of Eqs. (6.13) and (6.14) is easily found
the differential equation which eXz Xie- xZ must to be given by
satisfy. It is seen that this procedure generally leads
to a coupled set of linear homogeneous first-order a(x) = a, b(x) = f3aloc + (b - f3afoc)ea.x. (6.15)
differential equations which, as is well known, may be From Eqs. (6.6)-(6.9) and (6.15), we then find that
solved as an eigenvalue problem.
The lowest-dimensional Lie algebra of physical e"X+PY(aX + bY)e-a.x-PY
interest is the non-Abelian two-dimensional algebra = aX + [f3aoc- + (b1
- f3aoc- l )e"]Y. (6.16)
with elements {X, Y} satisfying
Note that it could have been foreseen from Eqs. (2.2)
[X, Y] = Y. (6.5) and (6.5) that a(x) = a. In future calculations, we
This algebra has been treated by Sackl l in connection make such possible simplifications at the outset without
with his "Taylor's Theorem for Shift Operators" and comment. From Eqs. (6.4) and (6.16), it follows that
by Wei and Norman. IS The latter authors have shown e"X+PYp(X, Y)e-"X-PY
that this algebra occurs in two master-equation
problems: (a) a system of simple harmonic oscillators = P[X + f3oc- l (l - ea.)Y, ea.y]. (6.17)
with Landau-Teller transition probabilities; (b) the Special cases of Eq. (6.17) are 47
deuterium exchange reaction with only nearest-
neighbor transition probabilities. Other realizations ea.Xp(X, Y)e-a.x = P(X, ea.y), (6.18)
of this algebra are known in quantum mechanics. The
set {qlf.l, e-ip.P/h} satisfies Eq. (6.5), as may be seen from ePYp(X, Y)e- PY = P(X - f3Y, Y). (6.19)
Eq. (2.10). Some realizations which involve annihila-
tion and creation operators are {-ata, a}, {ata, at}, A frequently occurring three-dimensional Lie
{-yaq, a2}, {yatq, (a t )2}, {yaq, q2}, and {-yatq, q2},
algebra is spanned by the operators {P, Q, I} with
the commutation relations
where a = y(q + iw-Ip), at = y(q - iorlp), y =
(wI21i)!, and [a, at] = 1. This algebra also occurs as [P, Q] = cI, [P, I] = [Q, I] = O. (6.20)
a subalgebra of larger Lie algebras. For example, in
the algebra SUs there are 12 distinct pairs of elements It is exemplified by the coordinate-momentum or
which satisfy Eq. (6.5).46 annihilation-creation operator commutation rules.
To illustrate the parameterization method alluded Since all results for this algebra are well known, we
to above, let do not explicitly derive them here. However, all such
Z = aX + f3 Y, (6.6) results may be obtained as a special case of the four-
dimensional Lie algebra which we consider next.
G = aX + bY, (6.7) A Lie algebra of interest for harmonic oscillator
G(x) = exzGe- xz (6.8) problems is spanned by the operators {P, Q, W ==
p2 + Q2, el} with the commutation relations defined
= a(x)X + b(x)Y. (6.9) by Eq. (6.20) and bylS
Then G(x) satisfies the differential equation
[W, P] = -2eQ, [W, Q] = 2eP, [W, I] = O.
G'(x) = [Z, G(x)], (6.10) (6.21)
subject to the initial condition
A physical realization of this algebra is provided by
G(O) = G. (6.11) the set {p, wq,p2 + w 2q2, -iliwI}. An alternative and
Substituting Eqs. (6.6) and (6.9) into Eq. (6.10) and more convenient set to use is {W, X, Y, sI}, where
using Eq. (6.5), one finds that
X = Q - iP, Y = Q + iP,
a'(x)X + b'(x) Y = [ocb(x) - f3a(x)] Y. (6.12)
W = XY - lsI, s = 2ie. (6.22)
Since X and Yare independent operators, as is easily
proved from Eq. (6.5), we must have Equations (6.21) and (6.22) imply the commutation
relations
a'(x) = 0, b'(x) = ocb(x) - f3a(x) (6.13)
[W, X] = -sX, [W, Y] = sY, [X, Y] = sl. (6.23)
subject to the initial condition
47 Examples of Eqs. (6.18) and (6.19) occur in Louisell, Ref. 3,
a(O) = a, b(O) = b. (6.14) Theorems 9 and 6, respectively.
970 R. M. WILCOX
A physical realization of this algebra is provided by Hence b'(x) = a x b(x) subject to b(O) = b. The
the set {at a, a, at, I}. Let solution is easily found to be given by
z = yW + ~X + pY, (6.24) b(x) = (a· b)a(l - cos ax)
G = g W + dX + r Y, (6.25) + b cos ax + a x b sin ax, (6.33)
and let G(x) be defined by Eq. (6.S). Then Eqs. (6.10) where a == lal and a = a/a. Hence
and (6.11) are again satisfied. Letting e-ia•J F(J,., J'II ,Jz)e-ia.J = F(J(,., Ky, K z), (6.34)
G(x) = g W + d(x)X + rex) Y + u(x)I, where
one finds the differential equations K = a(a • J)(1 - cos a) + J cos a + J x a sin a.
(6.35)
d'(x) = -sy d(x) + s~g,
r'(x) = syr(x) - spg, Equations (6.34) and (6.35) specify how a general
operator function of J must transform under an
u'(x) = s~r(x) - sp d(x),
arbitrary rotation of the coordinate axes characterized
subject to d(O) = d, reO) = r, and u(O) = O. The by the vector a. An interesting special case is where
solution is easily found to be given by a,. = a and a'll = az = O. Then Eqs. (6.34) and (6.35)
reduce t048
d(x) = g~y-l + (d _ g~y-l)e-SY",
rex) = gpy-l + (r _
gpy-l)eSY", exp ( - iaJ,.)F(J,., J'II , Jz) exp (iaJ,.)
u(x) = ~y-l(r - gpy-l)(eSY " - 1) = F(J,., J'II cos a + Jz sin a, Jz cos a - J'II sin a).
+ py-l(d _ g~y-l)(e-SY" - 1). (6.26) (6.36)
Another three-dimensional Lie algebra of wide
From Eqs. (6.4), (6.S), (6.24), (6.25), and (6.26), one
interest is the "split three-dimensional simple algebra"
finds
characterized bylS
eZF(X, Y)e- Z = F(U, V), (6.27)
where [E, F] = H, [E, H] = 2E, [F, H] = -2F. (6.37)
U = e-syX + py-l(e- SY - 1)1, (6.2Sa)
A physical realization of the set {E, F, H} is given by
V = eSYY + ~y-l(eSY - 1)/. (6.2Sb) {iJ_, iJ+, 2Jz}, where J+ and J_ are the angular
Expressed in terms of Q and P by Eqs. (6.22), one momentum raising and lowering operators, respec-
finds, with Z = a.P + fJQ + yW, tively, J+ == J,. + iJ'II' J_ == J,. - iJ'II' These operators
occur, not only for ordinary spin, but also for isotopic
eZF(P, Q)e-z = F(R, S), (6.29) spin and for quasi-spin in many-fermion systems. 49
where Another guise in which these operators occur is where
R = P cos 2ey + Q sin 2ey
J_ = etb, J+ = bte, Jz = t(btb - ete), (6.3S)
- Vy-l[a.(l - cos 2ey) - fJ sin 2ey], (6.30a)
and b t, et, b, and e are the creation and annihilation
S = - P sin 2ey + Q cos 2ey operators of a two-dimensional harmonic oscillator,
- Vy-l[a. sin 2ey + fJ(l - cos 2ey)]. (6.30b)
[b, btl = [e, et ] = 1, [b, e] = [b, e t ] = O. (6.39)
Another well-known Lie algebra is that of the
Another physical realization of the set {E, F, H} iE
angular momentum operators (or rotation generators)
given by {P2/2e, Q2/2e, (QP + PQ)/2e}, where [P, Q] =
{J,., J'II' J.}, where el. 1S Let
[J,.,J'II] = iJ., [J'II,J.] = iJ,., [1.,J'II] = iJ,.. (6.31)
Z = a.E + fJF + yH, G = aE + bF + gH, (6.40)
(We use units in which Ii = 1.) Let
and let G(x) once more be defined by Eq. (S). Letting
b(x) • J ==b,.{x)J,. + by{x)J'II + b.(x)Jz
= e-i,.a·J(b. J)e i (08.J, (6.32)
G(x) = a(x)E + b(x)F + g(x)H, (6.41)
48 Special cases of Eq. (6.36) are as follows: A. Messiah, Quantum
where a and b are constant vectors. Then Mechanics (John Wiley & Sons, Inc., New York, 1962), Vol. II, p.
578, Ex. 8; C. P. Slichter, Principles of Magnetic Resonance (Harper
b'(x) • J = -i[a. J, b(x) • J] and Row, Inc., New York, 1963), p. 26, Eq. (13) .
•• See, e.g., Schwinger, Ref. 17; Heffner and Louisell, Ref. 16;
= [a x b(x)] • J. Messiah, Ref. 48; Lipkin, Ref. 46.
EXPONENTIAL OPERATORS AND PARAMETER DIFFERENTIATION 971
one finds the coupled equations Then d == d(x), e == e(x), and 1== I(x) satisfy
a'(x) = -2ya(x) + 2ocg(x), d' = 2eoce - eyd, e' = eye - 2efJd,
b'(x) = 2yb(x) - 2{Jg(x), (6.42) I' = efJe - eEd,
g'(x) = -fJa(x) + ocb(x). subject to d(O) = 1, e(O) =1(0) = O. The solution is
The solution of Eqs. (6.42) subject to a(O) = a, b(O) = given by
b, g(O) = g, is found to be given by d(x) = cosh AX - yd-1 sinh AX,
a(x) = (afJ + boc - 2gy)OCp-2 sinh2 px e(x) = -2{Jd-1 sinh AX,
+ a cosh 2px + (goc - ay)p-l sinh 2px, I(x) = -EcA-1sinh AX
b(x) = (a{J + boc - 2gy){Jp-2 sinh2 px + (e/A)2(EY - 2fJfJ)(cosh Ax - 1), (6.50)
+ b cosh 2px + (by - g{J)p-l sinh 2px, where
(6.51)
g(x) = (a{Jy + bocy - 2gOC{J)p-2 sinh 2 px
The expression for e"'z Qe-"'z may be obtained from
+ g + !(boc - afJ)p-l sinh 2px, Eqs. (6.49) and (6.50) by making the following
where substitutions:
P? Q, c~ -e, OC? fJ, fJ? E.
Thus, with Z given by Eq. (40), one finds
We conclude this section by considering an infinite-
eZf(E, F, H)e- Z = f(J, K, L), (6.43) dimensional Lie algebra. This set, which occurs in the
work of Kermack and McCrea,19 consists of P and
where J, K, and L are defined as
all functions of Q, where [P, Q] = el. Clearly this set
J == u(oc, fJ, y)E + {J2wF + v(oc, {J, y)H, satisfies the closure condition since [P, tp(Q)] = ctp'(Q)
K == oc2wE + u(oc, {J, -y)F - v({J, oc, -y)H, is in the set. Let tp(Q) and/(Q) be arbitrary functions,
L == -2v(fJ, oc, y)E + 2v(oc, {J, -y)F and let
+ (1 - 2ocfJw)H, (6.44) (exp {x[ocP + tp(Q))})[{JP + I(Q)]
w == p-2 sinh2 p, x (exp {-x[ocP + tp(Q)]}) == fJP + F(Q, x), (6.52)
u(oc, fJ, y) == cosh 2p + ocfJw - yp-l sinh 2p, where oc, fJ, and X are parameters. Then F(Q, x) must
v(oc, (J, y) == fJyw - !{Jp-l sinh 2p. (6.45) satisfy the differential equation
In case y = 0, w, u, and v simplify to of(Q, x)/ox = [ocP + tp(Q), (JP + F(Q, x)]
= occoF(Q, x)/oQ - (Jetp'(Q), (6.53)
w = (ocfJ)-l sin2 (oc{J)!,
u(oc, (J, 0) = cos 2 (oc{J)!, subject to the condition that F(Q, 0) = I(Q). The
solution is easily found to be given by
v(oc, (J, 0) = -!(fJ/oc)! sin 2(oc{J)!. (6.46)
Some other special cases,
F(Q, x) = I(Q + occx) - (Joc-1[tp(Q + ocex) - tp(Q)].
(6.54)
e'¥.Ef(E, F, H)e-~E = fee, F + ocH + oc 2E, H + 2ocE),
Three special cases of interest are obtained by letting
eftFf(E" F, H)e- fJF = feE - {JH + (J2F, F, H - 2{JF), I(Q), {J, and oc, respectively, go to zero:
eYHf(E, F, H)e-YH = f(e- 2rE, e27F, H), (6.47)
(exp {x[P + tp(Q)]})P(exp {-x[P + tp(Q)]})
may also be easily obtained directly from the com- = P - tp(Q + ex) + tp(Q), (6.55)
mutator expansion, Eq. (6.22).
Consider now the six-dimensional Lie algebra whose (exp {x[P + tp(Q)]})f(Q)(exp {-x[P + tp(Q)]})
elements are I, P, Q, p2, QP, and Q2, with [p, Q] =el. = f(Q + ex), (6.56)
The most general second-degree polynomial in P
eq>(Q1pe-q>(QI = P - etp'(Q). (6.57)
and Q is a linear combination of these elements. Let
Z = ocP 2 + (JQ2 + yQP + tJp + EQ, (6.48) 7. PRODUCT OF LIE EXPONENTIALS
and let Let A and B be LCLE's and let eAe B =
eZ. Then
~Pe-zz = d(x)P + e(x)Q + l(x)I. (6.49) from the BCH formula, Eq. (4.14), and the "closure
972 R. M. WILCOX
property," Eq. (6.1), it follows that Z is also a LCLE. Next, consider the four-dimensional Lie algebra
In this section, we determine the form of Z, given defined by Eqs. (6.23). Let A = ylW + blX + PlY'
A and B, for some of the Lie algebras introduced in B = Y2W + b2X + P2Y, and Z(A) = yW + b(A)X +
the previous section. We parametrize the problem peA) Y + a(A)I, where Y = Y2 + AYl, b(O) = b2, p(O) =
by setting P2, and a(O) = O. Then from Eqs. (7.2) and (6.26)
(7.1) one finds, after integrating over x and equating
coefficients of X, Y, and I, respectively,
subject to Z(O) = B. Then differentiating Eq. (7.1)
with respect to Aby means ofEq. (4.1) and multiplying ylby- l +W- ylby-l )( -ys)-\e-Y' - 1) = bl ,
from the right by e-z = rBr).A., one obtains (7. lOa)
il dxea:zZ'().)e-a:z = A. (7.2) YlPy- l + (p' - YlPy-l)(ysrl(e YS - 1) = PI' (7. lOb)
sional Lie algebra of Eq. (6.5), [X, y] = Y. Let A = + Py-lW - ylby- )[( _ys)-l(e- - 1) - 1] = o.
l YS
Then integrating over x in Eq. (7.~) and equating b(A) = AblCP( -AYlS, -Y2S) + b2CP(Y~, AYlS), (7.11)
coefficients of Y, one finds that f3().) satisfies peA) = ).PlCP(AYlS, Y2S) + P2( -Y2S, -AylS). (7.12)
otl f3ot- l + (f3' - ot l f3ot- l)ot-l (efZ - 1) = f3l' (7.3) From Eqs. (7. lOa), (7.lOb), and (7.1Oc), one finds that
Defining u == u().) = f3ot-I, Eq. (7.3) becomes
fZ fZ a'().) = y-l[o(pb)/o). - Plb - blP]. (7.13)
(f3l - otlu)-V = (e - 1)-1 = e- (1 - e-fZr\ (7.4)
Equation (7.13) is integrated with the aid ofEqs. (7.11)
whi<ih may also be written and (7.12) by putting the terms in a form similar to the
-1 0 [1n (otlU - f3l)]/O). = o[1n (1 - e-fZ)]/o).. (7.5) right side of Eq. (7.5). Setting
Integrating Eq. (7.5) and solving for f3(A), one obtains Tl = Pl/Yl' T2 = P2/Y2' #1 = bl/Yl' #2 = b2/Y2'
(7.14)
f3(A) = Af3lCP(Aotl' o(2) + f32CP(-ot 2, -AotJ, (7.6) one obtains finally
where
cp(x,y) == (1 + x-1y)(e'" - l)(e"*Y - 1)-1. (7.7)
a(1) = (Yl + Y2)-lp(1)b(1) - YlTl#l - Y2T2#2 + e,
(7.15)
Hence where
[exp (otlX + f3ly)][exp ( ot2X + f32Y)] e == 2S-l(Tl - T2)(f1l - #2) sinh (tSYl)
= exp [(otl + o(2)X + f3(1) y], (7.8)
X sinh (tsY2) csch [tS(Yl + Y2)]' (7.16)
where f3(I) is obtained by setting A = 1 in Eq. (7.6).
This result may be stated in terms of the X and Y
Some special cases of Eq. (7.8) have been derived
operators as follows: Let [X, Y] = sI. Then the
by Sack by means of his "Taylor's Theorem for Shift
equation
Operators" 11:
exp [Yl(X + TlI)(Y + #11)] exp [Y2(X + T2/)(Y +#21)]
exp [ot(X + ).Y)] = e"x exp [)'(1 - e-fZ)y]
= {exp [).(e" - l)Y]}e"x. (7.9)
= exp [(Yl + Y2)(X + TI)(Y + #1) + 01] (7.17)
Sack has applied Eqs. (7.9) to the last four realizations is identical~y satisfied provided is defined by Eq.e
of this algebra listed above Eq. (6.6) in order to obtain (7.16), and provided T and # are defined by
a formula for the matrix elements of a Gaussian T == Tl"P(Yl' Y2) + T2"P( -Y2, -Yl), (7.18)
potential. 50
# == #l"P(-Yl, -Y2) + #2"P(Y2, Yl), (7.19)
60 Sack, Ref. 11. Matrix elements for generalized Gaussian
potentials and other potentials which may be represented as Fourier where "P(x, y) is defined by
integrals are obtained in R. M. Wilcox, J. Chern. Phys. 45, 3312
(1966). "P(X, y) == [esa: - l][es(a:+111 - 1]-1. (7.20)
EXPONENTIAL OPERATORS AND P ARAMETER DIFFERENTIATION 973
The result may alternatively be stated in terms of to Bloch's theorem have occurred in connection with
the P and Q operators: Let [P, Q] = eI. Then the the scattering of x rays or neutrons by molecules or
equation harmonic lattices,52 as well as in treatments of the
exp {YI[(P + VII)2 + (Q + WI!)2]) coherent states of the radiation fie1d. 53
X exp {Y2[(P + v2I)2 + (Q + w2I)2]) We conclude this section by treating a special case
= exp {(YI + Y2)[(P + V!)2 + (Q + wI)!] + OI} of the infinite-dimensional Lie algebra whose elements
are P and all functions of Q, where [P, Q] = eI. We
(7.21)
is identically satisfied provided seek to find a function Z == Z(Q, A) which satisfies
the Magnus formula, we assume a solution of the result is said to be the continuous analog of the BCH
form formula.
Yit) = exp [!leA, t)] == eO., (8.2) We now derive the Fer formula,ls which may be
where said to be the continuous analog of the Zassenhaus
00
!leA, t) == 2 An~n(t)· (8.3) formula. Let the solutions of Eqs. (8.1;) be given by
n=l
Yit) = eAS1e AS2e AS3 ... ,
2 8
(8.10)
From Eqs. (8.1;), (8.2), and (4.1), it follows that
where Sl == Sl(t), etc. Substituting Eq. (8.10) into
fdxe",o.ne-"'o. = AA(t), (8.4) (8.1 A)' multiplying from the right by
e-;.8S3e-ABs2e-ASI = [Yit)rl,
n
where == o!l(A, t)/ot. Using the commutator expan-
and expanding in terms of repeated commutators as
sion and integrating over x, as in Eq. (4.7), and sub-
stituting in Eq. (8.3), one obtains in Eq. (4.20), one obtains
00 An+l 00 00 Am+2n+2
{k=Oi (k +1 I)! (iAn~n(t)\'
n=l }
i m=l
AmAm(t)} = AA(t). 2
n=o(n + I)!
{S~,5't} + 2 2 - - -
m=On=O m! (n + I)!
(8.5) 00 00 00 Am+2n+Sk+S
x {S;n,S~,S2} + 2 22----
A recursive procedure again results from equating m=O n=Ok=O m! n! (k + I)!
coefficients of Ai on the two sides of Eq. (8.5). For x {S;n, S~, S~, Ss} + ... = M(t), (8.11)
j = 1, one obtains Al(t) = A(t). Hence,
where the repeated commutators are again defined by
~l(t) = SotA(T) dT. (8.6) Eqs. (4.21). The recursion scheme based upon Eq.
(8.11) leads to
For j = 2 one obtains
(8.12)
A 2(t) + U~l(t), Al(t)] = O.
Hence,
(8.13)
(8.7)
be more suitable for some purposes. It is obta.ined For the case where A(t) is given by Eq. (8.21), Eqs.
from the Magnus solution to Eqs. (8.1)) and is given (8.19), (8.22), (8.25), (8.26), and (8.27) constitute the
by Eqs. (8.18), (8.2), (8.3), (8.6)-(8.9), ... , and (8.15). solution of Eq. (8.1). By means of Eq. (7.9), this
These expansions are more complicated than the result may be shown to be equivalent to the product
Zassenhaus expansion, Eq. (4.18), but appear to be form of Wei and Norman, 55
more useful when )'b is "small" compared with 0.
They also appear to be more useful for large values (8.28)
of t, provided U(t) is unitary. A fuller heuristic com- A solution may also be obtained for the four-
parison of the two infinite-product forms for the case dimensional harmonic-oscillator algebra ofEqs. (6.23).
where t-- 00 has been given by Kumar. 14 In Eqs. (8.1) and (8.19), let
8.2. Lie Algebraic Solutions A(t) == get) W + d(t)X + ret) Y + u(t)! (8.29a)
We now consider the case where A(t) in Eq. (8.1) == g(t)[X + w(t)!][ Y + vet)!) + J(t)!, (8.29b)
may be expressed as a LCLE,
O(t) == yet) W + b(t)X + pet) Y + fl(t)! (8.30a)
A(t) = 01(t)X1 + 02(t)X2 + ... + an(t)Xn.
Then if a solution of the form == y(t)[X + w(t)!)[Y + vet)/] + rp(t)l. (8.30b)
yet) = eQ(t) (8.19) Then proceeding in the same way as before, using
either Eqs. (6.26) or Eq. (6.27), one obtains the
exists, the repeated-commutator form for 0 and the
differential equations
"closure property" of the Lie algebra imply that
O(t) is also a linear combination of the X/so We will
y =g,
find some closed-form solutions for two Lie algebras gv + v(l - e-'Y)s-l = d = gv,
of physical interest by a method similar to those gw + w(e8Y - 1)S-1 = r = gw,
employed in previous sections. Our method is also if; = g(w - w)(v - v) + f(t),
basically the same as the method employed by Wei
and Norman to obtain solutions in the product formlS where y == yet), g == g(t), etc. The solution, subject
yet) = exp [g1(t)X1 ] ••• exp [gn(t)Xn] (8.20) to the conditions yeO) = b(O) = p(O) = fl(O) = 0, is
given by
if one ignores the technical group-theoretical con-
siderations of that treatment. In practice, however,
yet) = fg(-r) dT, (8.31)
the single-exponential solutions are more difficult to
obtain.
vet) = [eSy(t) - l]-lSIX(t), (8.32)
We again consider first the two-dimensional algebra
where [X, Y] = Y. In Eqs. (8.1), let wet) = [1 - e-Sy(t)]-ls{J(t), (8.33)
A(t) == a(t)X + bet) Y, (8.21)
where aCt) and bet) are arbitrary functions of t, and let rp(t) = LtdT{g(T) [W(T) - W(T)][V(T) - v(r)] +J(T)},
O(t) == lX(t)X + f3(t) Y (8.22) (8.34)
where
in Eq. (8.19). Substituting Eqs. (8.21) and (8.22) into
Eq. (8.4) with), = 1, using Eq. (6.16), integrating over lX(t) == fd(T)eSY(T) dT, (8.35)
x, and equating coefficients of X and Y, one obtains
!X = aCt), !Xu + u(ea - 1) = bet), (8.23) (J(t) == fr(T)e-SY(r) dT. (8.36)
where
u(t) == f3(t)/IX(t). (8.24)
For the case where A(t) is given by Eqs. (8.29), Eqs.
Integrating Eqs. (8.23), one obtains (8.19) and (8.30)-(8.36) constitute the solution of Eq.
(8.1). This may be compared with the product form
lX(t) = fa(r) dr, (8.25) of Wei and Norman,55
(J(t) = [1 - e-a(t)]-llX(t)y(t), (8.26) Yet) = ey (t)Wea(t)Xe/l(t)Y e1jl(t)I, (8.37)
where
55 Wei and Norman, Ref. 15, derive differential equations, but
(8.27) do not explicitly give their solution since they are mainly interested
in determining whether or not solutions exist.
976 R. M. WILCOX
where aCt), pet), and yet) are given by Eqs. (8.35), pet), it may appear that nothing has been gained.
(8.36), and (8.31), respectively, and where However, it may be easier to apply the boundary
conditions to H(O) than to p(O). In particular, H(O) ==
87 Tl
"P(t) == Id'T[u(-r) - slX('T)r(-r)e ( l· (S.38) Je(O) may be an LCLE although p(O) is not. Consid-
ering the various methods available for handling
One may use either of these solutions to solve the exponential operators, Eqs. (9.5) and (9.6) may also
problem of the driven harmonic oscillator, without be a useful starting point for dealing with practical
the necessity of first transforming to the interaction many-body systems for which exact solutions are
representation as is usually done. 66 impossible.
Since Tr pet) = I at all times, as is implied by Eq.
9. SOLUTIONS OF iltap(t)/at = [Je(t), p(t)] (9.4), Eq. (9.1), and the cyclic property of the trace,
Consider the equation for the time evolution of Eq. (9.5) may also be written in the form
the density matrix pet), for a system described by a pet) == e-PH(tl/Tr [e- PHW ]. (9.7)
Hamiltonian Je(t),57
Equation (9.7) shows that at all times the system
iliop(t)/ot = [Je(t), pet)]. (9.1) behaves as though its temperature remains constant
(Aside from a sign, the same equation of course with a "canonical distribution" determined by the
applies to any Heisenberg operator.) In case Je(t) is an instantaneous value of the fictitious Hamiltonian
LCLE and pet) is initially an LCLE, then Eqs. (9.1) H(t). It may thus be possible to treat nonequilibrium
and (6.1) imply that pet) will remain an LCLE for all situations by the methods of equilibrium statistical
time. A solution may then be readily obtained, as in mechanics.
Sec. 6. However, the condition that an arbitrary A simple illustrative example is the problem of an
density matrix be expressible as an LCLE is very arbitrary spin magnetic moment in an arbitrary time-
restrictive, so that not many cases occur in practice. varying magnetic field with a Hamiltonian
A case where an arbitrary density matrix may be Je(t) = yb(t) • J
expressed in terms of an LCLE occurs in the problem
== y[h.,(t)J", + h,it)J1I + hit)Jz]. (9.8)
of a spin -t magnetic moment in a time-varying
magnetic field. 68 Then H(t) is of the form
However, as is well known, even if p(O) is not an H(t) = yb(t) • J. (9.9)
LCLE, a solution of Eq. (9.1) is given by
It easily follows from Eqs. (9.6), (9.S), (9.9), and
pet) = U(t)p(O)ut(t), (9.2) (6.31) that the fictitious field bet) precesses about the
follows 6S : With [P, Q] = cf, let the function f(P, Q) Theorem: Let F(P, Q) and G(P, Q) be in normal
be defined by its formal power-series expansion, form: F(P, Q) == .N'[F(P, Q)]; G(P, Q) ;:= .N'[G(P, Q)].
f(P, Q) == af + bP + dQ + ePQ + gQP + .... Then the normal form of the product is given by
(10.1) .N'[F(P + c%Q, Q)G(P, Q)]. (10.6)
Then the linear "superoperator" .N' acting onf(P, Q) Proof' We prove the result first for the special case
moves all P's to the right of the Q's as though P and where F(P, Q) and G(P, Q) have the forms
Q commute; i.e.,
F(P, Q) = ewQe zP, G(P, Q) = ellQe zp .
.N'[f(P, Q)] == af + bP + dQ + (e + g)QP + .... Then
(10.2) F(P, Q)G(P, Q) = ewQezPe"Qe zP
= ellJllCewQe"QezPeZP
Although other operator expressions may be obtained
from f(P, Q) by application of the commutation = .N'[eZ"cewQe"QezPeZP ]
relation, we emphasize that their functional forms = .N'[ewQez(P+c"'e"QeZP]
will be different. Thus, even thoughf(P, Q) = g(p, Q) = .N'[ewQez(p+c%Q'eIlQeZP]
in the usual operator sense, it need not follow that
.N'[f(P, Q)] = .N'[g(P, Q)]. Some useful properties of
= .N'[F(P + e%Q, Q)G(P, Q)] .
.N' are a direct consequence of its definition. If The general case easily follows by the method of
f(P, Q) and g(P, Q) are any two operator functions, linear superposition. The theorem may be readily
then generalized to the case of more than one pair of
conjugate variables .
.N'[f(P, Q)g(P, Q)] = .N'[g(P, Q)f(P, Q)]. (10.3)
As an application of this theorem, consider the case
If IP') and IQ') are eigenstates of P and Q, PIP') = where F(P, Q) = pm, G(P, Q) = Qn. Then
P' IP') and Q IQ') = Q' IQ'), then
{.N'[f(P, Q)]} IP')= f(P', Q) IP'), (lO.4a)
pmQn = .N'[(P + e%Q)mQn]
=.N'I m! pm-lei o;Qn
(Q'I {.N'[f(P, Q)]} = (Q'lf(P, Q'*), (lO.4b)
i=Oj! (m - j)! OQi
(Q'I {.N'[f(P, Q)]} IP') = f(P', Q'*)(Q' IP'). (lO.4c) m m! n! e;Qn-;pm-J
=I (10.7)
Equations (10.4) usually occur in practice for the case J=oj! (m - j)! (n - j)! .
where P and Q correspond to the annihilation and To show how the theorem may be used for solving
creation operators a and at, respectively, and the differential equations, and to compare with the treat-
vacuum state 10) is involved, a 10) = 0 or (01 at = O. ment of Heffner and Louisell,l6 we consider their
If thefin Eq. (10.1) also depends upon a parameter example of a spin magnetic moment in a rotating
A, f == f(P, Q, A), then it is apparent that .N' com- magnetic field. The Hamiltonian is then of the form l6
mutes with O/OA since
:re = lliwo(btb - etc) + lliyHl(btee-irot + etbeirot),
.N'[oj(P, Q, A)/OA] (10.8)
= .N'[a'f + b'P + d'Q + e'PQ + g'QP + ... ] where the annihilation and creation operators b, bt ,
= a'I + b'P + d' Q + (e' + g')QP + ... e, and ct were defined in Eq. (6.39), and Ii, w o, y, H l ,
= o{.N'[f(P, Q, A)]}/OA, (10.5) and ware costants. We assume that the solution of
the equation
where a' == oa(A)/oA, etc. From the definition of the iliO == ilioU/ot = :reU (10.9)
derivative with respect to an operator, Eq. (2.0), it
is of the formes
readily follows that .N' also commutes with %P and
%Q. U = .N'{exp [(A - l)b t b + (B - l)ct c
An important problem of normal-ordering is to + Dbtc + Ectb]), (10.10)
express the product of two normal-ordered products
in normal form. S•le where .N' orders band c to the right of b t and et, and
It Our definition of the "nonnal-ordering operator" is not the U For convenience, we define U such that A, B, D, and E corre-
same as that given by Louisell, Ref. 3, and Heffner and Louisell, spond to the notation of Ref. 16, but this, of course, is not essential
Ref. 16, but we believe it is more convenient. for the solution of the problem.
978 R. M. WILCOX
the scalar functions A, B, D, and E are to be deter- substituting the results into Eq. (10.16), one obtains
mined. It clearly follows that ezPe"'QPellQ = X{exp [c-1(e"'" - l)QP
0= AbtUb + EctUc + DbtUc + EctUb. (10.11) + eC"'(zP + yQ + cyz1)]}. (10.19)
From the above theorem, 2JeU/1i is given by A special case of interest is obtained by setting z = 0,
X[(wob t + yHleiwtct)(b + %bt)U
P = a, Q = at, and c = 1 in Eq. (10.19):
+ (-woc t + yH1e-iwtb t)(c + %ct)U]. (10.12) e"'a t aellat = exp (ye"'at)X{exp [(e'" - 1)at a]). (10.20)
functions and associated Laguerre polynomials. A coefficients of p2, QP, Q2, P, Q, and I, respectively, on
corollary of one of these is the formula the two sides of Eq. (10.34) leads to the differential
equations
am(a t)m+r = m! (a ty .N'[L~( -a ta)], (10.26j
ex. = A'd2, y = 2A'de + G'd,
a result which one may also obtain directly from Eq.
(10.7) by using the definition of the associated P = A'e2 + B' + G'e, ~ = 2A'd! + D'd,
(10.35)
Laguerre polynomial. E = 2A'e! + G'f + D'e + E',
The next derivation shows how the exponential 0= A'f2 + eA'de + D'f + F',
of an arbitrary second-degree polynomial may be put
into normal form. This result generalizes a theorem of where d == d(x), e == e(x), and! == lex) are defined by
McCoy.Is Eqs. (6.50). Equations (10.35) simplify to
A'/ex. d- 2, G' = -e-I d- 2d',
= B'/P =
Theorem: Let
(10.27)
D' = p2('TG' + 2p,A'), E' = p2(pG' + 2'TB'),
where F' = -!d'd-I + p2(cp - yr5E) + p4'Tp,G'
Z == rxJ'2 + PQ2
+ yQP + ~p + EQ, (10.28) -ley + p4(4ex.pcp + y2cp - 8ex.Pyr5E)d-2.
(10.36)
W == AP2 + BQ2 + GQP + DP + EQ + Fl.
(10.29) Integrating Eqs. (10.36) subject to A(O) = ... =
F(O) = 0 and setting x = 1, one obtains Eqs. (10.30)
Then Eq. (10.27) is identically satisfied provided for A == A(1), B == B(1), etc. Q.E.D.
ex.-IA = P-1B = H == ..
(.~ l)-l sinh A, UsingEq. (2.8), one may verify that both sides ofEq.
G = e-1(J-l - 1), (10.27) satisfy the pair of partial differential equations
D = p2('TG + 2p,A), au/ap = 2AUP+ GQU + DU, (10.37a)
(10.30)
E = p2(pG + 2'TB),
aUjaQ = GUP + 2BQU + EU. (10.37b)
F= -t In J - ley + p2(cp - yr5E) + p4'Tp,G
Some special cases of Eqs. (10.27)-(10.31) are of
+ p4(4IXPcp + y2cp - 8ex.Pyr5E)H, If ~ = E = 0, then D = E = 0 and
interes~.
where
J == cosh A - py sinh A, p == A-Ie, exp (ex.p2 + PQ2 + yQP)
A = e[y2 - 4ex.P]~, cp == IXE2 + pr52 , (10.31) = [JecTt.N'[exp (Ap 2 + BQ2 + GQP)], (10.38)
'T == yr5 - 2ex.E, p, == yE - 2pr5. where A, B, G, and J are given by Eqs. (10.30) and
(10.31). This result may be shown to be equivalent to
Proof: Instead of Eq. (10.27), consider
McCoy's theorem. ls We note that McCoy gave an
(10.32) ingenious derivation based upon a pair of partial
differential equations like Eqs. (10.37) with D = E =
where W(O) = O. In Eq. (10.29), W, A, B, G, D, E, and
O. If we set y = 0 in Eq. (10.28) and define
F are now considered to be functions of x. Differ-
entiating Eq. (10.32) with respect to x, one obtains S == P + 01, R == Q + aI, y == 2e(ex.p)t, (10.39)
Ze xz = .N'[W'(x)eW(xl] then
= A'exZp 2 + B'Q2eXZ + G'Qexzp exp (IXS 2 + PR2)
+ D'exzp + E'Qe xz + F'e xz, (10.33) = (sec y)~.N'{exp [0'-1 tany)(ex.S2 + PR2)
where A' == aA/ax, etc. Multiplying Eq. (10.33) from + e-l(sec y - I)RS]}. (10.40)
the right by e-xz , one obtains If we set ex. = P = 0 in Eq. (10.28) and define Sand R
Z = A'e xZp2e-O:Z + (G'Q + D')eXZPe- xz by Eqs. (10.39), then
+ B'Q2 + E'Q + F'I. (10.34) e1RS = .N'{exp [e- 1(e C1 - l)RS]}. (10.41)
The quantities e"'zPe-xz and Equations (10.40) and (10.41) may also be derived
from Eq. (10.38) by using the fact that Sand R satisfy
exZp2e-XZ = [e XZPe- XZ ]2 the same commutation relation as P and Q.
are obtained from Eqs. (6.48)-(6.51). Equating An expansion of Kermack and McCrea, which
980 R. M. WILCOX
provided the starting point for Sach's derivation of. between states (q'l and lq"). Since g(F) has the same
his "Taylor's theorem for shift operators," 11 is as eigenstates as F, the left-hand side of Eq. (10.45) may
follows 18 : be written
00
The cases {J > 0 and (J < 0, respectively, are obtained The Poincare formula may also be used to efficiently
by choosing the contours of integration to be infinite obtain results of the type derived by Aizu for general
semicircles in the upper- and lower-half x plane. operator functions. 2 A formula from which a number
For the special case where H == H(;") is of the form of sum rules can be derived is obtained by considering
H = :Ie + ;" V, one finds that the nth derivative of H to be a function of two parameters, ;" and p, and
E({J) with respect to ;", onE({J)/o;"n == E(n)({J), is given taking matrix elements of o2j(H)/o;"op. Then from
by Eq. (11.1), one obtains
, ·n Jex)
~ dxei/l"'(x- iH)-l[V(X - ih)-l]n. (11.3) o":f(H)
(m/--/r) = (m/ AAI' + AI'A + BAI' /r), (11.7)
217i -ex) o;'op
Equation (11.3) follows directly from Eq. (11.2a) by where
using the formula for the derivative of the inverse, oH oH
o[(x - iH)-l]/O;" = i(x - iH)-lV(X - iH)-l.
(m/ AAI' /r) == ! (m/- /n)(n/- /r)Smnr (11.8)
n 0;" op
A recursive formula for E(n)({J) which shows its "time- and
ordered" form is given by 02H
(m/ BAI' /r) == (m/ o;'op /r)Tmr · (11.9)
E(n)({J) = -nf:dUE({J - u)VE(n-1)(u). (11.4) In Eqs. (11.8) and (11.9), Smnr and T mr are defined by
Tmr == _1
217i
i c
fez) dz
(z - Hm)(z - H r)
. (11.11)
T. - Il f'(H)
mr - mr m
+ Pmr f(HHm) -_ f(Hr)
H
(1112)
.
m r
and
Smnr = Illmnllnr/"(Hm) + Umnr + Urmn + Unrm ,
(11.5)
(11.13)
In the integrations over x and y, we have used Eqs. where'
(11.2) and (11.3), respectively. Equation (11.5) is the I,
Il mn == 1 - Pmn == ( (11.14)
same as Eq. (11.4), since E({J - u) = 0 if u > (J, and 0,
E(n-l)(u) = 0 if u < O. For the case where n = 1,
since E(O)(u) == E(u) = e-"H, Eq. (11.4) constitutes U = PmnPnrPrmf(Hm)
another derivation of Eq. (2.1). An equation like mnr - (Hm - Hn)(Hm - Hr)
Eq. (11.4) has been derived by Kumar in a different + Il [f(Hr) - f(Hm) _ f'(Hm)~.
manner, while the derivatives of the exponential mnPnr (Hr _ H m)2 Hr - H
evaluated at ;" = 0 occur in a well-known expansion (11.15)
of eJe+ AV• Of course, higher derivatives may also Equations (11.7}-(11.9) and (11.12H) 1.15) may be
be calculated from Eqs. (11.2) in the same manner shown to be equivalent to [25] of Aizu. 2
when H do~s not just depend linearly upon ;", but the It is straightforward to obtain matrix elements of
results will be more complicated. For example, the higher derivatives by the same method, but the results
second derivative, E"({J) = 02e- flH /0;"2, is then given obtained are complicated. If one specializes the
bye4
results so obtained to the case where I(H) =
E"({J) = - f: duE({J - u)H"E(u) 1(:Ie + ;"V) = exp (:Ie + ;'V) and compares with the
matrix elements of Eq. (11.4), one finds that the latter
results do not immediately have such a simple form.
+ 2 f: du !o"dVE({J - u)H'E(u - v)H'E(v), Comparison of the two forms reveals the existence of
(11.6) the curious identity
where H' == oHio;" and H" == 02H/o;"2. N N
" An equivalent formula given by Snider, Ref. 23, Eq. (B7), is not
! II (Xi - Xk)-l = 0, (11.16)
i-O k=l
as concise and symmetrical. k¢i
982 R. M. WILCOX