Appendix B. Time Ordering and the Magnus Expansion
We consider the evolution of a system under the influence of its Hamiltonian htf. The system is described by its density matrix p(t), which evolves ac- cording to the von Neumann equation
ß(t) = -il*(»,P(t)l. (B-1) If #? is constant with respect to time, Eq. (B-1) may easily be integrated
formally:
p(t)= UWPMU-H», (B-2)
where
U(t) = exp[--ijr/]. (B-3) If 2tf is not constant in the time interval of interest (say, from t = 0 to t — tc),
but is constant in smaller intervals xt,..., xk,..., τη with
*i + ·" + **+ ·*· + τη = ' o then by repeated integration we obtain
U(tc) = exp [- « ; τJ · · · exp [- Ufk τJ · · · exp [- ijft τχ], (Β-4) where we denote by J^k the Hamiltonian operative in the &th time interval.
U(tc) as given by Eq. (B-4) equals β χ ρ [ - / ( ^ τΜ+ · · · + ^Tjk+---+Jf1T1)]
only if all J^k commute with each other. Finally, if 2tf varies continuously from t = 0 to t = tc, U(tc) may be expressed formally by
U(tc) = Texp - 1 Γ Jf(t) dt . (B-5)
Equation (B-5) is to be considered the limiting case of Eq. (B-4). T is the 173
Dyson time-ordering operator, which orders operators of greater time arguments in the expanded exponential to the left, e.g.,
(jr(t')Jf(t") if t' > t\
T{^{t')J^{t")} =
\je(t")$e(t') if t" > t\
but also T{Jf2 J f J = Τ{^^2} = J^2^i if JtT2 "acts" at later times on the system than does J^x.
Magnus Expansion
The goal of the Magnus expansion is to find an expression in the form of exp[— iFtc~\ for U(tc) as given by Eq. (B-4) or (B-5). In order to find F we start by expanding Eq. (B-4) in a power series:
U(tc) = [··■(/!)···] x - x [ - ( * ) · · - ] x -
x
r
i + (_ ^
i T i ) +(-^^
2 +^--;
V-^
+, „ | , (B-6)
3! + J' where (k) means that the indices of Jt and τ are k.
We consider ^kxk as a small quantity of first order. In case it is not, we simply subdivide the time interval 0, ...,tc further until J^krk is small. By rearranging Eq. (B-6) we get
U(tc) = 1
+ ( - / ) { ^1τ1+ ^2τ2 + · · · + ^ τ „ }
(-i)3
+ - j p « \ 3 + - + 3 K /12T2T12 + /22/1T22T1 + - ] + 6[\?&3 J*i2 J^i t^l^l ~f~ *"j)
+ ···. (B-7a) Note that in all products the operators are ordered in ascending order to the
left! Equation (B-7a) can be rewritten in a compact form as follows:
(-i)2 (-i)m
U(tc) = 1 + (-/){·.·} + ^ T { . . · }2 + ... + L - j L T{.-.}- + ..., 2! ml
(B-7b) where {···} = { ^1τ1+ ^2τ2 + · · · + ^ τ/ Ι} .
For example, T{···}2 = • • · Τ ( ^1^2 + ^2^Τ1)τ2τ1 = —23Ρ2#ιτ1τ1 as it should according to Eq. (B-7a).
Next we make an ansatz for F:
F = JF + <#(1) + Jf(2) + ···. (B-8)
In this ansatz we assume tcJF,...,tcJF(k\... to be small quantities of first, ...,Ath,... order. We shall confirm this assumption later. Now, as above, we expand exp [— iFtJ:
exp[-/Ffc] = e x p [ - / ic( ^ + J*(1) + i f( 2 )+ · · · ) ]
= 1
[φ <2> (D <φ
+ ^ ^ | ( ^ )2 + ^ ^( 1 ) + ^( 1 )^ + (^(1))2
2· ( © (D φ <3)
+ J ^( 2 ) + jf( 2 )jf + J>f(2)iF(1) + ···
® @ <D
(~*03 i(^)3 + (^)2^( 1 ) +JP( 1 )(^)2 3!
+ j f ^( 1^ + («#(1))3 + ··
( , / c ) 4{(jpr+·
4!
+ ·.·. (B-9) The encircled numbers indicate of which order the corresponding terms are
small.
We rearrange terms in descending order of magnitude:
exp [ - « * „ ] = 1 ©
-iteJP Φ
_// c^(t) + t|£>!(iP)2 φ
etc.
(B-10)
By equating terms of equal order in Eqs. (B-10) and (B-7a) [or (B-7b)] we find
* = 7 W = 7 { ^ ιτι + ^ 2 τ2 + - + ^ τπ} , (B-ll)
^( 1 ) = ^ [ T { - }2- ( i , ^ )2] = ^ | T { - }2- { · · · }2] (B-12a,b)
= -^-([■*2,Ji'l3T2T1 + [Ji'3,jri]T3T, + [ j r 3 , j r2] T 3 T2+ - ) , (B-12C)
jf( 2 ) = - — ( T i - l ^ i / e ^ - S / ^ C i P ^ ^ + ^ ^ i P ] ) (B-13a)
6/c
= ^ ( π · · · } 3 + 2 { . . . }3- ^ [ { . . . } Γ { . . . }2 + Τ{...}Μ...}]^ (B-13b)
+ 5 { [ ^ 2 , [ ^ 2 , ^ ι ] ] τ22τ ^ (B-13c)
etc. From Eqs. (B-12) and (B-13) it is clear that icif( 1 ) and tc^(2) are small quantities of second and third order, if tc i f is a small quantity of first order—
as we actually assumed—and so our ansatz is consistent.
If 3tf(t) jumps only once in the interval 0,..., tc, and ^τ1 = A, J^2τ2 — B, we find
tcJF = A + B, /cJf(1) = -!rlB,Al, and
It follows that
e~iBe-iA = cxp(-i(A + B) -\{.B,A~\ + ^ { [ A [ J M ] ] + [ [ * , 4 M ] } + -Λ (B-14) Equation (B-14) is the Baker-Campbell-Hausdorff formula.
To arrive at the Magnus formula we must subdivide the interval 0, ...,tc
more and more finely. The limiting case of Eq. (B-l 1) for all xk -► 0 is
- I f
ctc Jo JT(t)dt. (B-15)
Thus «# ("^f-bar") is the average of J^(t) in the interval 0, ...,tc and is therefore called the average Hamiltonian.
Similarly, the limiting case of Eq. (B-12) is
^ (1) = ΪΓ f dh f dh Lje(t2) ' * ( , ι ) ] · (B " 16)
The domain of integration (see Fig. B-l) ensures that everywhere the operator with the greater time argument is on the left—where it should be.
Z(t2) dt2
FIG. B-l. Domain of integration for ^( 1 ).
The limiting case of Eq. (B-l3) is
1 Γ*° f* Γ<2
&w = ~7- dt3 dt2 dh
orc Jo Jo Jo
x {i>r a 3 ), w 2 ) , ^(hm+\_w(h\ ^(t 2 )i *?{t,)-]}.
(B-17) The domain of integration (see Fig. B-2) ensures again that everywhere the operators with greater time arguments are ordered to the left. It also takes care of the factor \ in front of the second line of Eq. (B-l3c).
For the reader who is not satisfied with our claim that Eqs. (B-15)-(B-17) are the limiting cases of Eqs. (B-11)-(B-13) we recommend that he try to recover the latter from the former in a case where 34?(t) jumps n times (n = 5, say) in the interval considered.
m)
X(t3)
FIG. B-2. Domain of integration for ^( 2 ). (Rear, upper left corner of cube).
'' dt, za>
t
r, *i£/^/' Γ / 1
/ y\
/ y
'*>^L
We proceed to indicate proofs for the special cases mentioned in Chapter IV, Section D. By comparing Eqs. (B-10) and (B-7b) for jf( k ) = 0, k < «, the validity of
tc<&(n) = Γ - Τ ^Τ{ · · · Γ+ 1 pf(k) = 0 for k<n) (B-18) becomes immediately evident.
As
A ^ T { . . . r + i ^ ^ ( _ / T Λ l Λ A Λ
x jr(/
n+1) j r ( 0 - JT(r
2) jr(ii), (B-i9)
we establish Eq. (4-44), Note there is no denominator (n+\)\ in the right-hand expression of Eq. (B-19). However, the volume of integration is "only"
fe»+7(1+1)!.
Before we turn to the special properties of symmetric cycles we prove a lemma. Consider βχρ[/^τ„] ···exp[/^f2τ2] εκρ[}^ιτϊ\- We call this expres- sion U(— tc) but avoid assigning any physical meaning to it. For U{— tc) we make the ansatz exp[//ic], and for/we make the ansatz
/ = R + RU>+ #*> + .... (B-20) The lemma we want to prove is
Lemma 1:
h™ = (-l)*JF<k>.
For the sake of compactness of notation we define if(0) = ^f. Furthermore let us define the symbol Jifk+1 as the sum of all products of the type
>p<*i)jp<*2)...jp<kn) in Eq. (β-9) that contain n factors and that are small of order m = k+1. Examples are
> 3 (if( 0 ))3, ^r23 = jf( 0 )^( 1 ) + iF( 1 )iF( 0 ), « ^3 = e^(2).
^ 3
Correspondingly we define the symbol hkn+1; it refers, of course, to an equation analogous to Eq. (B-9) in which the signs of all the /'s are reversed.
As
/! + £ * , = (*! + l) + (Jfc2 + 1) + ··· + (kH+l) = m = * + 1, (B-21)
i
it follows from Lemma 1—if it is correct—that
or
- ( - l )w^f c + 1 = (-1)*Λ*+1. (B-22)
To prove Lemma 1 consider the construction laws of iF( k ) and h(k) [cf. Eqs.
(B-9) and (B-7b), and the corresponding equations with the signs of all I'S reversed]:
(— i)k 1 (— it Ϋ
3! 3lt+1 v " " / *pk+l 2! "Jf|2 + 1, (B-23)
" "(*+!)!//* ' (£+1)Λ
+1-(- ^ Α * 3+ 1- | * Ι+ 1. (B-24) By taking into account Eq. (B-22) we recognize that «#(fc) and Bk) differ, term
by term, by ( - 1)\ Therefore, we recover Lemma 1 for k, having made use of it only for k' < k. Recall that the lowest value of« in Eqs. (B-23) and (B-24) is two, and that the largest value of kt occurring in J f |+ 1 and hk2+ x is kmax = k— 1 [see Eq. (B-21)]. As Lemma 1 holds obviously for k = 0 (iF = h) we conclude, by induction, that it holds for all k.
Now consider symmetric cycles,146 i.e., cycles for which Jif(t) = Jif(tc — t).
An obvious consequence of the symmetry of the cycles is U'Htc) = e x p ^ r j . - e x p ^ v l
= exp [iFs ic] [consequence of Eq. (B-4)]
= exp \i3tfn τ J · · · exp [/«#! τ{] (because the cycle is symmetric)
= U(-tc) (by definition)
= exp [ifs /c] (by definition).
The indices s on F% and/s stand for "symmetric."
We conclude Fs=fs, or
fc = 0
and with the aid of Lemma 1
Σ jrik) = Σ R(k)>
= 0 k = 0
j p w = (-1)* j?<*>, which requires
jf(fc) = 0 for A: odd.
From now on we follow closely Wang and Ramshaw.1
For the sake of completeness we also consider briefly antisymmetric cycles, i.e., cycles for which 3P(t) = -3f(tc-t).
U(tc) = exp[-«feTj exp[-üfB_ t τ„_ J ··· exp[- Uf2τ2] exp[-ί*± τ{\
= expCüfiTi] exp[^2τ2] •••exp[/Jf„_1Tn_1] expftJi^J
(consequence of antisymmetry)
or
exp[-iF./e] = exp[ + iF./J (a stands for antisymmetric). It follows that
Fa = -Fa = 0,
which means that for antisymmetric cycles M vanishes together with all its correction terms. Whereas symmetric cycles are of great practical importance in high-resolution NMR in solids, antisymmetric cycles are not, for obvious reasons.