CAPÍTULO III.MATERIALES Y MÉTODOS
2.3 P ROCEDIMIENTO
2.3.3.2.10 Especificar cómo se construirá la VPN
We onsider here the problem of population transfer within the rotational
framework as an optimization problem, subjet to the numerial modeling
for diatomi moleulespresentedearlier. The objetive to be metis dened
as the probability to populate a spei target rotational level, given the
initial groundstate:
wherepossibly
Jtarget
∈ {0,2,4,6,8, . . . , Nrot}
. Inouralulations,theyield subjet to maximization is simplyα
(g)
Jtarget(T)
2
, in terms of the notation
introdued earlier. Also,bydenition,
M
= 0
.Weonsider
Nrot= 20
,where this expansionwasonrmed to giveon- verged resultsinthepresent alulations. Themoleule underinvestigationhasa rotational onstant of
Brot
=Bg
=Be= 5cm
−1
.
Solving the dening dierential equations for the population transfer
problem(Eq.8.13)isobviouslyomputationallyexpensive. Inpratie,given
aneletrield,asingleevaluationoftheresultingwavepakethasthedura-
tionof approximately
5s
onasingle P4-HT2.6GHz
proessor. We arethus interested in optimization proedures withas minimal funtion evaluationsaspossible.
8.2.1 Experimental Proedure
There are several dening parameters in the present alulations. Some of
them are ritial, as they pose diret onstraints on the quantum system
at hand, and pratially determine its ontrollability. In our model, suh
parameters are the peak Rabi frequeny, whih plays the equivalent role
of the laser intensity, as well as the pulse duration. Setting these two pa-
rameters denes the simulatedphysial system. Given thetarget rotational
level, itis thenpossible to aim at steering thesystem toward it. Thus, we
hooseto onsiderthe population transfer asa funtion of these two den-
ing parameters, where the fous will be on spei values that reet best
state-of-the-art laboratoryexperiments.
Fromthealgorithmiperspetive,wehoosetorestritouralulationsto
theDR2andtheCMAalgorithms,whihperformedbestontheTwo-Photon
Proess problems. Theyboth employsmall populations,and onsider rst-
order andseond-order information, respetively.
PreliminaryRuns Preliminaryalulationsrevealedalearpiture,whih
ouldhavebeenpreditedbyintuition 1
. Thesepreliminaryalulationswere
onsisted of
10
runsperalgorithm onJtarget
={0,2,4,6,8}
withthefollow- ingpeakRabifrequenies:Ωge
={40,60,80, . . . ,160,180} ×1012s−1.
Given aRabi frequenyof
Ωge
= 160×10
12s−1
,thequantum systemould
easily be steered into perfet ontrol for low
J
values(J
={0,2,4})
. This task beame infeasible for higherJ
values with the given Rabi frequeny. However,whenthelatterwasinreased,e.g.,Ωge= 180×10
12s−1
,itbeame
1
AsmuhasintuitionexistsforQuantumMehanis;"Mybattingaverageonintuition
feasible. Hene, there is a trend of ontrollability as a funtion of thelaser
intensity,espeiallyforthehigherrotationallevels. Asfarasthealgorithmi
performanewasonerned, the DR2and theCMAperformedequally well
on the given systems. Most importantly, there was never a situation where
the DR2obtained ontrollability on a given system on whih theCMA did
not, nor vieversa.
We onsider the ase of a target rotational level of
J
= 4
as an inter- esting ase-study. This is due to the fat that it allows perfet ontrol atΩge
= 160×1012s−1
, but yet it is a hallenging task for the optimization routines. Also,theeetofdereasing the peakRabifrequenywhilelosingontrollability anbe observed relativelyeasily.
8.2.2 Numerial Observation:
J
= 0−→J
= 4
WeappliedtheDR2algorithmtotheoptimizationofthepopulationtransfer
problemfrom
J
= 0
toJ
= 4
. Theseoptimizationswereperformedforthree values ofthe peakRabifrequeny:Ωge
=
80×1012s−1,
120×1012s−1,
160×1012s−1
.
All alulations were arried out with
80
runs, limited to10,000
fun- tion evaluations per run. These alulations obtained qualitatively dier-ent results for the three intensities onsidered. For
Ωge
= 80×10
12s−1
the optimizations were unable to aomplish the transfer from
J
= 0
toJ
= 4
with unit eieny. The best eieny obtained was≈
32%
. ForΩge
= 120×1012s−1
and forΩge
= 160×10
12s−1
the transfer eieny
approahed
100%
inmost of thealulations.Aimingatomparingtheresultsofindividualoptimization runs,wede-
neaorrelationoeientthatomparespulse-shapesattainedintworuns
i
andj
,bymeans oftheir eld intensities:ci,j
=
max∆t{PtIi(t)Ij(t+ ∆t)}
hq
P
tIi2(t)
q
P
tIj2(t)
i
(8.16)where
Ii(t)
andIj(t)
are theeld intensities of the pulsesobtained inrunsi
andj
, respetively. Taking the maximum as a funtion of∆t
is due to the fatthatpulse-shapesattained bytheoptimization maybeshiftedwithrespettoeahother. Thesumsareoverthedisretetimesteps,asonduted
in the numerial alulation. Eq. 8.16 thus yields
ci,i
= 1
, andci,j
= 0
if pulsesi
andj
do not overlap at all.Case 1:
Ωge= 80×10
12s−1
FigureA.4presentstheorrelationoeient
for the
80
optimization runsof theΩge
= 80×10
12s−1
test-ase. The runs
population are highly orrelated. Upon examination of the atual alula-
tions,itisobservedthatallofthese solutionsareverylose toa singleFTL
pulse. Deviations from the FTL pulse do not only lead to a drop in the
orrelation oeient, but alsoin thepopulationtransferyield.
Case 2:
Ωge
= 120×10
12s−1
In Figure A.5 the orrelation oeient
is plotted for the
80
optimization runs that were performed for theΩge
=
120×1012s−1
test-ase. Here,thelaserpulseenergywassuienttotransfer
populationfrom
J
= 0
toJ
= 4
withnear-uniteieny. Thebestsolutions, whihhaveapopulationtransfereienyof99.982%
and99.98%
,wereonly weakly orrelatedto eah other,and wereonlyweaklyorrelated tomost oftheothersolutions. Speially,therewereonly
9
solutionsamongthesetof80
thatshareaorrelationoeientlarger than0.95
withthebestsolution (indexedas1
). Manyoftheremainingsolutionsarestronglyorrelated with the3
rd
-best solution, whih hasa populationtransfer yieldof
99.975%
: As many as41
solutions shared a orrelation oeient larger than0.95
with thatsolution (indexed as3
). Whilethethree good solutions1
,2
,and3
are ratherdierentfromeahother, theyontainmost ofthedominantfeaturesof the identied optimized solutions.
Solutions1-3arepresentedinFigure8.1. Despitetheirdierenehara-
teristis,all threesolutions inFigure8.1aredominated bya seriesofpeaks
with a separation of
4.79×10
−13s
. This orresponds to thebeating period
of a oherent superposition of
J
= 2
andJ
= 4
(∆E
= 14B
). Additional good solutions likely exist, possibly ontinuously onneted on a ommonlevel set, and further speialnumerial methods are needed to explore this
possibility,suh asthe D-MORPHalgorithm (Setion6.1.3).
Case3:
Ωge
= 160×10
12s−1
FigureA.6presentstheorrelationoeient
for
80
optimization runs of theΩge
= 160×10
12s−1
test-ase. While the
degreeofpopulationtransferisveryhighinalmostalltherunsatthisinten-
sity,the orrelationbetween thevarioussolutions is verylimited. Clearly,a
largenumberofsolutionsthattransferthepopulationwithuniteienyo-
exist,withvery little ommonality between them. Indeed, inspetion of the
atualpulseshapesobtainedintheserunsrevealshighlyompliatedpulses,
withfewregularfeatures,andanabseneofthepeakarisingfromoherene
between
J
= 2
andJ
= 4
inthe Fourier transform powerspetrum. 8.2.3 Intermediate DisussionUponinreasingtheintensityfrom
Ωge
= 80×10
12s−1
to
Ωge
= 160×10
12s−1
we nd that population transfer is aomplished with an ever inreasing
numberof distinguishablesolutions.
Figure 8.1: Comparison of the
3
best-performing pulse shapes that were obtained in80
runsof the DR2for the population transferproblem ofJ
=
0−→J
= 4
atΩge
= 120×10
12s−1
. All solutionsonsist oftrainsof pulses
with a spaing of
4.79×10
−13s
, whih orresponds to the beating period
that ontrollable quantum systems with no onstraints plaed on the on-
trolsonly have extrema thatorrespond to perfet ontrol, or to no ontrol
at all; Additional analysis revealed the fundamental nature of ontrol level
sets(seeCorollary6.1.3)atthe absoluteextremaandatsub-optimalontrol
yields.
Astriking aspetof theresultsisthe evidenethatthe number
of independent solutions produed by an optimization seems to
ritially depend on the diulty of the problem. In the urrent
populationtransfer alulations we observed that at lowintensity,
wherereahing the target isahardproblem with less thanperfet
yield, the trials invariably onverge onto one and the same solu-
tion, whereas at higher intensity, where this represents an easier
problem, a wide variety of solutions are enountered.