Systematic Errors in the ACME Electron EDM Experiment

Daniel Ang
Harvard University
(Dated: December 17, 2016)
The search for the permanent electric dipole moment of the electron (eEDM) is a powerful probe
for high energy physics beyond the Standard Model. In the last few decades, atoms and now
molecules have provided the means to dramatically improve statistical sensitivity in measuring
de . However, increases in statistical sensitivity must be balanced with corresponding increases in
characterization and elimination of sources of systematic errors. In this paper, we focus on the
measurement methods and systematic errors present in the ACME EDM experiment [3], currently
the most sensitive experiment searching for the electron EDM. We discuss both heuristic models
and physical mechanisms which can result in systematic errors in measuring de , and methods taken
to reduce known systematic errors. We also discuss attempts to solve the several systematic-error-
related puzzles which still remain in the experiment.

I. INTRODUCTION AND THEORETICAL decades. Such an EDM would indicate the presence of a
MOTIVATIONS FOR ELECTRON EDMS T-violation (time-reversal violation) in nature. This can
be seen that if we perform the transformation t t to
In physics, electric multipole moments arise when one a system where an electron EDM d~e is initially aligned
tries to study electric potentials by expressing them in with the electrons spin ~ , then d~e d~e but ~ ~ .
terms of the electric potentials of ideal basic combina- Thus if |d~e | > 0, the system is T-odd. By the CPT the-
tions of elementary charges, such as monopoles, dipoles, orem, where the product of CPT (charge, parity, and
quadrupoles, etc. This is called the multipole expansion. time symmetry) must always be preserved, this implies
The dipole term is the term proportional to 1/r2 , i.e. [5] a CP-violation.

d~ r
Vdip = , (1)
r2 II. EDMS OF FUNDAMENTAL PARTICLES AS
A PROBE FOR NEW PHYSICS
where for a general charge distribution (~r), the electric
dipole moment (EDM) d~ is defined to be The Standard Model of particle physics does allow for
Z some CP-violation that results in a non-zero EDM for
d~ = (~r)~rd3 r. (2) fundamental particles. In the case of the electron EDM
(eEDM), this arises from coupling of the electron to vir-
Thus in this classical picture, an EDM can be roughly tual quarks via the W boson [6]. However, the contri-
understood as the degree to which an electric potential bution is only non-zero at the level of 4-loop Feynman
resembles that of the charge configuration consisting of a diagrams, which gives a very small SM prediction for d~e :
pair of positive and negative unit charges separated by a
distance s, where we take the limit s 0. In order to fix |d~e | 1038 e cm, (4)
the value of d~ in Eq. 2 independent of coordinate system,
much smaller than the sensitivity of the state-of-the-art
the origin is commonly taken to be at the center of mass.
eEDM measurement, which is only at 1028 e cm.
Thus d |d~e | can be understood as the separation be- However, many new models of physics beyond the
tween the centers of charge and mass of the particle (or Standard Model (BSM) predict a much larger value for
system).
In atomic and particle physics, de is commonly given d~e . These models contain more CP violation than the
the units e cm. The (nonrelativistic limit) of the energy SM, which is appealing because it can solve other major
contribution from a dipole moment when placed in an problems in physics. One of the most notable puzzles is
~ is that the CP-violation in the SM is too small to explain
electric field E
electroweak baryogenesis in the early universe resulting
H = d~ E.
~ (3) in the present asymmetry between baryonic matter and
antimatter. Some BSM models, such as the Minimal Su-
In modern atomic, molecular, and optical physics, one persymmetric Model (MSSM), predict |d~e | 1029 ecm
can talk about the EDMs of atoms and molecules sim- (with certain natural assumptions) [7]. Thus, AMO
ilar to EDMs in a classical system. It is common for eEDM experiments are a useful tool to test these the-
composite molecules to have a permanent EDM, such ories. Compared to efforts at large particle accelera-
as water. However, the search for a permanent non- tors, eEDM experiments are often several orders of mag-
zero EDM in fundamental particles such as the neutron, nitude cheaper while probing comparable (TeV) energy
proton and electron have been ongoing for the last few scales [4].
 2

III. ELECTRON EDMS IN ATOMIC AND angular momentum onto the internuclear axis. The H-
MOLECULAR SYSTEMS manifold of ThO has = 1 and contains a pair of so-
called -doublet states. These states have opposite par-
To make an experiment measuring an eEDM, one must ity, so when the molecule is subjected to an electric field
subject an electron to an electric field in order to create ~ Stark shifts separate the energy levels by 2dH |E|
E, ~ (as-
a measurable energy shift as shown in Eq. 3. However, suming M = 1). We denote the two states by the
it is difficult to perform experiments on free electrons. It quantum number N = n E = M sign (E z) = 1,
is much more advantageous to perform an experiment by which also denotes whether the effective electric field is
examining an unpaired valence electron in orbit around aligned with the laboratory electric field (Fig. 1) [9].
the nucleus of an atom or molecule. When one takes into A diagram of the whole experiment can be seen in
account relativistic effects,1 we will have Fig. 2, although for the purposes of this paper we will fo-
cus only on aspects that are relevant to discussing the sys-
d~A = d~e , (5) tematic errors arising in the experiment. The molecules
are first produced via laser ablation of a thorium tar-
where d~A is the atomic/molecular EDM and is a con- get in a buffer gas-cooled cell. Initially, they are all
stant that depends on the atom or molecule involved. in the ground X state, with a mixture of different J
Generally, Z 3 , where Z is the atomic number of the and M values. Optical pumping using a 943 nm laser
atom [6].2 Hence the effective electric field felt by the transfers molecules in the |X 1 + 0 , J = 1, M = 1i state
electron is amplified, amplifying the energy shift being to the short-lived |A3 0 , J = 0, M = 0i excited state,
measured: which spontaneously decays to the metastable states
|H, J = 1, M = 1, N i and |H, J = 1, M = 0 i.
H 0 = d~e E
~ eff , (6) Next, optical pumping is performed with a 1090
nm laser between the states |H, J = 1, M = 1, N i
where E ~ eff = E.
~ This allows one to apply electric fields 1
and |C 1 , J = 1, M = 0, P = +1i. The 1090 nm laser
far beyond what is feasible in a laboratory. For example, (which is called the prep laser ) is such that it can select
in the case of ThO, 109 , so by applying a E ~ of tens which N state to pump out, but leaves the different M
of kilovolts, one can get an effective electric field of tens states unresolved. Hence, it pumps out one of the two
of gigavolts. Similarly large values of also exist in the possible superpositions of the two possible superpositions
case of YbF and Hf+ . of |M = 1, N i (called the bright state) out from the H-
manifold into the C-manifold, upon which they decay
into the ground state. The other superposition (the dark
IV. SPIN PRECESSION MEASUREMENT FOR state) remains and becomes the experimental state:
ELECTRON EDMS
i = 1 (|M = +1, N
|( ), N i |M = 1, N
i). (7)
In state-of-the-art eEDM experiments such as the ThO 2
(Harvard-Yale) [4] and YbF (Imperial College) [8] exper- At this point, the molecules enter the interaction re-
iments, the measurement of the eEDM in a molecule is gion, where the spin precession measurement begins.
performed via a spin precession measurement. In the rest Note that the forward (longitudinal) velocity of the
of this paper, we shall discuss the method of measuring molecules is taken to be along the x direction. Constant
de and the systematic errors that can arise from it, us- ~ = E z) and magnetic (B ~ = B z) fields are ap-
electric (E
ing the case of the ACME EDM Generation I experiment
plied uniformly in the z direction, causing an energy shift
(henceforth referred to as ACME I) as the main case
given by
study.3 This experiment published the best current limit
on de in 2014 [3]. dH |E|
H = N ~ M H |B|
~ B
In the ACME I experiment, spin precession is
performed using thorium monoxide molecules in the MN B|
E||
~ B|
~ ~ ~
B M N E(de Eeff ), (8)
|H 3 1 , J = 1, M, i metastable state. is the quan-
tum number denoting the projection of total electronic where H is the magnetic moment of the H-state, and

is a parameter characterizing the E-dependent magnetic
moment difference between different N levels. We have
also used the notation B, E = 1 to denote whether the
1 ~ ~
E, B are aligned or anti-aligned with the fixed laboratory
In the non-relativistic limit, Schiff s theorem tells us that any
electron EDM will be canceled out by the other charges in the axis z. The energy shift causes the initial state in Eq. 7
atom or molecule. Sandars [1] was one of the first to realize that to acquire a phase
the theorem is evaded in the relativistic limit.
2 Or the larger atom in a molecule. 1 L
Z
3 The following explanation for the spin precession measure- = (H(M = +1, N ) H(M = 1, N ) dx ,
ment method is adapted from various published materials about 2 0 v
ACME I, especially [4], [6] and [9]. (9)
 3

and M . Figure adapted from [4].

Figure 1. Structure of H-state of ThO, showing the various shifts dependent on N

Figure 2. Schematic of the whole ACME experiment, including the beam source, stem, and interaction region. Figure adapted
from [4].

where v is the forward velocity of the molecules. Compo- read prep is the relative polarization between the
nents with different M values acquire opposing phases, preparation and readout lasers, and C is a unitless param-
resulting in the state eter between 0 and 1 characterizing other imperfections
in the system that reduce the number of photons, such
i = 1 (ei |M = +1, N
|( ), N i + ei |M = 1, N
i). as imperfect optical pumping, decay from C back to H,
2 etc. The relative polarization angles are chosen such that
(10) | | /2.5
We then project this spin-precessed state into a known During the experiment, read is rapidly switched at 200
state by optical pumping on the H C transition4 using kHz between alignment with the x and y lab axes, result-
linearly polarized light (which we call the readout laser). ing in FX = F (0) and FY = F (/2). We then compute
After some time the molecules will decay back to the an asymmetry
ground state X, resulting in fluorescence given by
FX FY
  A= = |C| cos (2( )), (12)
1 + |C| FX + FY
F () = f N0 cos2 ( ), (11)
2
from which we extract the phase . From Eq. 8, we can
where N0 is the number of molecules in the relevant N
state, f is the fraction of detected fluorescence photons,
5 We measure the contrast for a given N , E,
B by dithering the

value of = nr +, where nr is the non-reversing component
4 In ACME II, this is the H I transition. Throughout this of and = 1. The contrast is given by C = 12 A
. It is
paper we will mostly refer to the ACME I experiment. commonly measured to be 0.95.
 4

see that by performing ion measurement multiple times6

and switching the relevant values of N , E,
B = 1, we
will obtain a parity sum of all the components of :

, E,
(N B) N + E
=N E + B
B
E
NE + N
B
N B + +EB
EB
+N
+N N EB
EB (13)

from which we can extract whichever term we want by

adding multiple measurements of the appropriate com-
bination of parities. In practice, all 16 possible combi-
nations of the 4 binary switches N , E, are performed
B,
7
(called a block) . Then the electron EDM de is ex-

tracted from the component of the phase N E which does

not reverse under switching of N E via the equation

N E Figure 3. Data acquisition structure of the ACME experiment
N E = , (14) at the block and superblock level. Figure adapted from [4].


where N E is the precession frequency and the time
of precession. The spin precession time is determined
of the experiment. As long as the contribution from the
from the Zeeman precession phase: extracting B from error source reverses under N E,
it will be washed out
~ , using known measured
the equation B = B gH |B| as more data are averaged, remaining only as harmless
~
values of H and |B|. statistical noise. As the measurements are performed at a
The fundamental statistical uncertainty on de comes repetition rate frep (where commonly, frep = 50 Hz), only
from the quantum shot noise, which can be expressed as noise occurring at extremely close to nfrep (where n N)
will take impractically long to average out.8 Thus, any
1 systematic error that shows up in the ACME experiment
de = p , (15)
2|C| |Eeff | N T
~ must be those that correlate with N E.
9
An additional safeguard against systematic errors in
where is the spin precession time, N is the number ACME is the presence of other binary switches in ad-
of molecules, and T is the averaging time (commonly dition to the 4 main block switches. For example, in
taken to be about a week). For ACME I, this resulted in Generation I there were 4 superblock switches, such as
de 5 1029 e cm. This is an important number to which swapped the physical leads on the power supply
L,
remember as we go through discussions about systematic
to the field plates generating the electric field,10 and P,
errors, as a systematic uncertainty smaller than statisti- which switched frequency the readout laser beam to al-
cal uncertainty means is effectively negligible. ternate between the two -doublet states in the C-state.
Fig. 3 shows a glimpse of the data acquisition structure at
the block and superblock level. For each possible value of
V. PROCEDURE FOR SEARCHING FOR the superblock switches (24 = 16 possible states), all the
SYSTEMATIC ERRORS

The ACME experiments method of extracting de al-

8
lows it to be resilient against potential sources of system- Even noise at exactly nfrep Hz is in principle still not a source
of systematic error, as long as it is not non-reversing under N E.

atic errors which may enter the system in various parts
That being said, it is still undesirable for the system to have
statistical noise coming from any source which causes a statistical
error larger than that which arises naturally from the quantum
shot noise, so much effort is still devoted to eliminating sources
6 As the ThO beam source is produces molecules in batches (or of noise when one can.
9 Much of the discussion on systematic errors in this and following
shots) which repeat at frep = 50 Hz (due to the method of
generating ThO via laser ablation), it is natural to denote spin sections was adapted from the thorough exposition in [10]. It
precession on one shot of molecules as one measurement. The is also worthwhile to note that at the time of writing, ACME
measurement would be repeated 25 times and then averaged, II is already underway, introducing various improvements that
creating one trace, before switching to a different combination will result in its own possible plethora of systematic issues. This
of binary switches. reference also discusses these potential Generation II issues in
7 A block actually consists of 25 4 16 = 1600 shots, as each greater detail.
switch combination (giving a trace) is performed 4 times and 10 The E switched is performed by switching the set voltage on the
averaged. power supply.
 5

possible values of the block switches also had to be per- where ,i , 2 ,i , dr ,i are constants of appropriate di-
formed. There were also even higher-level uberblock mension. To isolate the component that mimics an EDM
~ and B~ and E
switches, such as changing the magnitude of E and thus a systematic error, we need to take i N i
reversing the propagation direction of the readout laser.
and r,i N E
r,i in the equation above, looking for only
With the exception of the last one (which we shall cover E.

the contributions to the phase that correlate with N
in a later section), none of the superblock or uberblock
From this empirical model, we now proceed to the actual
switches caused any change in the measured de , so all the
physical mechanisms that can cause these terms to arise.
block data was simply averaged over.
In addition to the safeguards that are a part of the
fundamental data acquisition structure, over 40 other
A. Non-Reversing Electric Fields Resulting in
parameters were intentionally varied in search for possi- Changes to i

ble systematic correlations with the N E EDM channel.
These were divided (in Ref. [4] and [10]) into parame-
First, we shall focus on the i factor in the first term
ters with an ideal value and those without. The latter
of Eq. 16. It turns out that this factor can arise from im-
category includes parameters used in the data analysis
perfections in the electric fields applied to the molecule,
routine and others which are outside of the scope of this
namely non-reversing fields or E nr .11 This could be the
paper. (In addition, none of the parameter variations
result of an electronic offset in the power supply to the
found a statistically significant systematic error.) We
field plates, for example. Such an imperfect E could be
shall instead focus on several entries in the former cat-
parameterized as
egory, especially those which relate more closely to the
atomic and molecular physics of the experiment. Exam- Eztot = Ez E + Eznr + ... (17)
ples of this category are non-reversing electric and mag-
netic fields, transverse magnetic fields (Bx,y ), prepara- where E is the intended applied field. We have also taken
tion/readout laser detunings from the ideal frequency, only the z-component of E (which is the intended direc-
etc. tion), assuming that the other components are zero or
For each of these parameters the deviation from the cancel out throughout the interaction region. The pres-

ideal was amplified up to 10 times and the resulting N E ence of Eznr causes a Stark shift in the experimental H-
noted. These were called Intentional Parameter Varia- state that is odd under N E,
as can be seen in Fig. 4.
tions (IPV). It was assumed that the relation between The Stark shift changes the frequency for the H C

N E and each parameter could be well-characterized lin-
early, preventing the necessity of taking a full, two-week
dataset on each possible IPV. Higher-order polynomial
fits were also tried when applicable, but no significant
increase in systematic uncertainty was observed. It is of
course always possible that nonlinearity is present in be-
tween the data points that we took in the IPV, but this
possibility was lessened by taking data over as wide of
range of parameters as possible.

VI. SYSTEMATIC ERRORS CAUSED BY AC

STARK SHIFTS

We now proceed to discuss the systematic errors found

in ACME I, starting with the two major ones. One
of the main mechanisms by which additional terms can
contribute to and potentially result in a systematic-

inducing N E is through AC Stark shifts coming from the
Figure 4. Structure of Stark shifts in the H-state as a result
prep and readout lasers. These shifts were first detected of Eznr . Notice that the sign of the shift changes depending
in ACME I as it was observed that the measured phase on the N -state. Figure adapted from [4].
had a dependence on the laser detuning frequencies i
and Rabi frequencies i , where i {X, Y, prep}. This
dependence can be expressed as the general equation
X dr,i
(, r ) = [,i i + 2 ,i 2i + dr ,i + ...], 11 For the rest of this paper, nr stands for non-reversing - i.e.,
r,i
i X nr indicates the component of X which does not reverse under
(16) any binary switch in the experiment.
 6

transition, which we can parameterize as where dH is the transition dipole moment between H
C and I is the laser intensity. Thus it seems that any
1 N
E N E
nr
HC,i = HC Pi C,J=1 + N
dH Eztot , (18) r,i must be the result of some W (where W is the
2
laser power) or dN E
H . However, W
NE
was measured using
E
where the Pi term exists because we switch the parity photodiodes and found to be too small to explain N r,i .
of the C-state addressed by the readout lasers (which is It was then thought that the effect could be explained
E
either X or Y ; see Section V). Assuming that the laser by dNH , arising from Stark interference between E1 and
frequency is L,i such that i = HC,i L,i , inserting M1 dipole transition amplitudes, the fuller derivations
the result of Eq. 17 into Eq. 18, we immediately see that which are present in [10] and [11]. However, despite the
an N E-dependent
term arises in i : lengthy calculation in [10], it is conceded that the Stark
interference effect vanishes for interference of multiple
nr
i = HC dH |Ez E + Eznr | + ...
+N E
plane waves. Thus, while in the end N r,i was thoroughly
= nr N E + ... measured and constrained to be at the 1031 e cm level,
i + N Ei (19)
its origin as well as its puzzling dependence on (k z)
remains an open puzzle to be solved in future generations
where N i
E
= dH Eznr (other terms, including those de-
pendent on E, N , or P have been subsumed into the of the ACME experiment.
...). Inserting this back into Eq. 16, this result in a Altogether, the resulting systematic error arising from
term that masquerades as the EDM in . the laser detuning and Rabi frequency is then given by
E
X
N
Eznr = ,i dH Eznr , (24)
i
B. The case of r,i E
X
N
NE = dr,i (N E nr
r,i /r,i ). (25)
r
i
Analogously, one can derive a similar result for the
dependence of the H C transition Rabi frequency,
assuming the existence of some mechanism to cause a C. Physical Mechanism for ,i , 2 ,i , and dr ,i
N E
non-reversing r,i :
Now we shall proceed to explaining a model for how the
r,i = nr N E
r,i + N Er,i + ... (20) AC Stark shifts are caused - in other words, how the con-
stant factors ,i , 2 ,i , dr ,i arise in Eq. 16. According
Unlike the case of N E
arising from Eznr which we have to this model, when pumping on the H C transition
i
E (either with the prep or readout lasers) some mixing be-
just discussed, in ACME I a statistically significant Nr,i
tween dark and bright states occurs due to imperfections
was detected, but the physical mechanism is still not fully
in the laser polarizations. This causes a precession phase
understood. This effect was first detected in the form of
E-dependent
shift.12
an N fluorescence signal that reversed with
We first start with the three states involved in the opti-
the laser propagation direction k z,
cal pumping between H C, as described in Section IV:
the dark and bright states of H and the excited C-state:
F NE
(2.4 103 )(k z) (21) |B( , Pi
), N , |D( , Pi
), N , |C, Pi
, (26)
F nr
where we have included the fact that the bright and dark
as well as a non-zero N E B which was similarly dependent
states depend on the polarization  of the laser. The
on Bz and k z. From this it is believed that the N E B above basis is valid valid for both the prep and read-
arises from the relation out stages of the experiment. We transform this into a
new 3-state basis in which it is easier to understand the
N E
N E B = d
B
Bz r
nr
, (22) phenomenon [10]:
r
r
+ |B(
|B i |C, Pi , Pi
), N ,

which is a B-odd AC Stark shift. Eq. 22 was empirically , Pi

|Di = |D(
), N (27)
verified by inserting an artificial N E into the experiment
and observing that
EB
N
changed accordingly. for some coefficients , which can be easily found, along
When looking for possible physical mechanisms that with the eigenvalues of the new basis vectors. We as-
sume that some physical mechanism causes the laser po-
could give rise to N E , one can first observe that the
larization to be imperfect, possibly changing its ellipticity
Rabi frequency r,i is determined by [11]

dH p
r,i = Ii 0 c, (23) 12 Full details of the following calculations can be found in Ref. [10].
~
 7

and/or linear polarization angle. Allowing the new ba- which imprinted some ellipticity on the light, given by the
sis state vectors of Eq. 27 to vary in time, this means expression [4]
that the laser polarization imperfection is also evolving
in time. One can thus define a complex polarization ro- S3 /I = (x) sin (2( )), (35)
tation rate which takes into account both types of polar-
ization imperfection: where S3 /I is the ellipticity Stokes parameter, (x) is the
material-dependent retardance on incident laser light on
= i( + ), (28) the field plates, is the linear polarization angle of the
laser and the angle of the fast axis of the birefringent
where 21 S3 /I (S3 /I being the laser ellipticity Stokes field plates. The form of Eq. 35 indicates that it is pos-
parameter) and is a detuning that accounts for the sible to minimize the imparted ellipticity by adjusting .
Zeeman shift due to the presence of the magnetic field B ~ Indeed, various values of were tried and the resulting
in the experiment. From here, the Hamiltonian can be dependence of on measured (see Fig. 5). The which
expressed in the new basis, as well as the time-evolving gave the flattest slope (in the case of the figure, = 0 )
state of the molecule during the spin precession: was picked as the experimental configuration, minimizing
the effect of Eznr . In addition, Eznr was further reduced
|(t)i = cD (t) |Di + cB+ |B+ i + cB |B i . (29)

Applying time-dependent perturbation theory, for the

case of the prep laser, one finally obtains a state that
consists of a mix of dark and bright components:
, Pi
0prep , N
|D( = |D( , Pi
prep , N (30)
+ d(, t) |B( , Pi
), N , (31)

where 0prep = prep + d(, t)i z prep and (, t)

is a parameter that characterizes the dynamics of the
evolving laser polarization during the spin precession. In
the end, one will obtain

Im()
,prep dprep , (32) Figure 5. Dependence of measured phase on prep laser
prep
2 detuning for different values of . Figure adapted from [4].
Im()
2 ,prep (dprep t), (33)
prep
Re() by using a chopper wheel rotating at frep to reduce the
dr ,prep (dprep t), (34) time-averaged laser power incident on the field plates by
r
a factor of 2. All of these methods succeeded to reduce
explaining the origin of the constants in Eq. 16 for the Eznr systematic was to the 5 1030 e cm level.
case of the prep beam. A similar derivation can be per- The second possible source for polarization imperfec-
formed for the readout lasers as well. Thus we see how tions is the Gouy phase shift, which arises from the
z
the combination of an Enr and some residual bright state fact that Gaussian laser beams must result in polariza-
admixture in the experimental state can cause a shift in tion gradients at some level, which is a consequence of
the measured phase. Maxwells equations. Such an effect would be dependent
on the beam size. Assuming a laser beam propagating
in the z-direction, the resulting gradient would be in the
D. Sources of Polarization Imperfections xy-plane [10]:

The arguments of this section has so far shown that 2 y

, (36)
any deviation in the polarizations of the lasers can poten- x 2 wx wy
tially cause a phase shift due to its coupling to the N E-

nr where is the laser wavelength and wx , wy are the dimen-
correlated Ez . Thus, it is always preferable to minimize
any sources of laser polarization gradients. In ACME I, sions of the beam. For ACME I, this effect turned out to
a particularly interesting mechanism was thermal stress- be too small to contribute significantly to the systematic
induced birefrigence of the electric field plates. This oc- error budget.13
curred as a result of the prep and probe lasers being shone
through the ITO-coated field plates responsible for pro-
ducing the electric field. The field plates absorbed the
13 However, in ACME II, instead of optical pumping, STIRAP is
laser light, causing it to heat up and inducing birefrigence
 8

VII.
E-CORRELATED SYSTEMATIC ERRORS The alignment between the readout laser beam and
the laboratory electric field was varied.
Due to the existence of the pair of opposite parity N -
The magnitude of Ez was varied. In each case,
states and our ability to switch between them, the ACME the molecule was fully polarized, so no significant
experiment is resilient against systematic effects resulting change was expected.

in a contribution to E , such as leakage currents, geomet-
ric phases and motional magnetic fields. These effects are We did not include in our discussion several other
mostly washed away when we reverse N . In addition, a theoretical models of systematic errors which were
great advantage of ThO is that it is fully polarized by considered but estimated to come in at a negligi-
only applying an electric field of 10 V/cm. Thus the ble level, for example interference from levels other
experiment is always operating in the regime where the than the three-level system we discussed in Sec-
M -states are fully mixed, such that the contribution of tion VI C. The most complete exposition of all of
de to the Hamiltonian is always maximized. these models can be found in [10].

However, there is still a small contribution to N E in In all of the above cases (as well as others we did not have
states have slightly
the presence of E because the two N
space to include), there was no significant effect on N E .
different magnetic moments, parameterized by , as was
Still, in the final systematic error budget published in
mentioned briefly when discussing Eq. 8. The contribu-
[3], some uncertainties from the above effects were added
tion is given by [4]
anyway as a precaution, based on experience of other
experiments (notably PbO [13] and YbF [8]). Even this
|Ez |B E
N E
E
= , (37) extremely conservative total systematic uncertainty was
gH still below the statistical uncertainty: the final measured

value on |de | published in [3] was
which was empirically verified by applying BzE = 1.4mG,
The
i.e., a magnetic field that reversed direction with E. de = (2.1 3.7stat 2.5syst ) 1029 e cm, (38)
N E
resulting effect on was found to be three orders of
a 12-fold improvement over the previous result of [2].
magnitude smaller than the effect on E . Thus this effect
did not contribute significantly to the systematic error
budget. IX. SUMMARY AND DISCUSSION

From our discussion of systematic effects, we see that

VIII. OTHER POSSIBLE SOURCES OF while there are several traits of ThO which make it possi-
SYSTEMATIC ERRORS
ble for ACME to reach extraordinary statistical precision
(most notably its large internal electric field and high
We have now finished discussing most of the theoreti- electric polarizability), the true power of the ThO ex-
cal explanations for the major systematic effects detected periment lies in the magic of the H-state, peculiarly the
in the ACME experiment for its 1st generation result [3]. -doublet structure. As we have seen, and the method
As outlined in Section V, it is worthwhile to note that of extracting de by measuring an asymmetry and then
besides these, almost all possible knobs on the experi- switching N , E, states makes it incredibly resilient
B,
ment were varied and amplified to see if any effect on against most systematic effects. Still, there remain two
N E would result. Examples include: major contributors to systematic error: Eznr , which is rel-
E
atively well-understood and the mysterious N r . How-
Arbitrary large magnetic gradients (including in ever, the fact that statistical improvements to ACME II
the x, y directions). Such gradients were respon- have been accomplished smoothly gives us good reason
sible for the ultimate systematic limitations on the to hope that in the next generation of the ACME exper-
earlier electron EDM experiment using PbO [13]. iment more thorough investigations on these systemat-
However, this effect was anticipated to be small for ics and any new ones can be performed rapidly.14 One
ACME as the H-state has a very small value of gH . can appreciate that the ACME experiment has resulted
in great gains in understanding systematics arising from
Laser detunings other than those correlated with EDM searches which can potentially be of great benefit to
Eznr which we have already discussed previously concurrent and future electron permanent EDM searches

(i.e., N P
i , i , etc.) were exaggerated. [8][14].

used for state preparation, which involves highly focused laser

prepares the initial state at the start of the precession measure-
beams which would have resulted in a large Gouy polarization
gradient if not for the presence of a refinement beam which re-
 9

[1] P.G.H. Sandars. The electric dipole moment of an atom. [8] D M Kara, I J Smallman, J J Hudson, B E Sauer, M R
Physics Letters, 14(3):194 196, 1965. Tarbutt, and E A Hinds. Measurement of the electrons
[2] J. J. Hudson, D. M. Kara, I. J. Smallman, B. E. Sauer, electric dipole moment using ybf molecules: methods and
M. R. Tarbutt, and E. A. Hinds. Improved measurement data analysis. New Journal of Physics, 14(10):103051,
of the shape of the electron. Nature, 473:493496, 2011. 2012.
[3] J. Baron, W. C. Campbell, D. DeMille, J. M. Doyle, [9] B. Spaun. A Ten-Fold Improvement to the Limit of the
G. Gabrielse, Y. V. Gurevich, P. W. Hess, N. R. Hutzler, Electron Electric Dipole Moment. PhD thesis, Harvard
E. Kirilov, I. Kozyryev, B. R. OLeary, C. D. Panda, University, 2014.
M. F. Parsons, E. S. Petrik, B. Spaun, A. C. Vutha, [10] B. OLeary. In search of the electrons electric dipole
and A. D. West. Order of magnitude smaller limit on moment in thorium monoxide: an improved upper limit,
the electric dipole moment of the electron. Science, systematic error models, and apparatus upgrades. PhD
343(6168):269272, 2014. thesis, Yale University, 2016.
[4] J. Baron, W. C. Campbell, D. DeMille, J. M. Doyle, [11] P. Hess. Improving the Limit on the Electron EDM: Data
G. Gabrielse, Y. V. Gurevich, P. W. Hess, N. R. Hut- Acquisition and Systematics Studies in the ACME Exper-
zler, E. Kirilov, I. Kozyryev, B. R. OLeary, C. D. Panda, iment. PhD thesis, Harvard University, 2014.
M. F. Parsons, E. S. Petrik, B. Spaun, A. C. Vutha, and [12] C. D. Panda, B. R. OLeary, A. D. West, J. Baron, P. W.
A. D. West. Methods, analysis, and the treatment of Hess, C. Hoffman, E. Kirilov, C. B. Overstreet, E. P.
systematic errors for the electron electric dipole moment West, D. DeMille, J. M. Doyle, and G. Gabrielse. Stimu-
search in thorium monoxide. To be submitted, 2016. lated raman adiabatic passage preparation of a coherent
[5] D.J. Griffiths. Introduction to Electrodynamics. Always superposition of tho H 3 1 states for an improved elec-
learning. Pearson, 2013. tron electric-dipole-moment measurement. Phys. Rev. A,
[6] David Commins, Eugene D.DeMille. The electric dipole 93:052110, May 2016.
moment of the electron. Lepton Dipole Moments, pages [13] S. Eckel, P. Hamilton, E. Kirilov, H. W. Smith, and
519581, 2009. D. DeMille. Search for the electron electric dipole mo-
[7] Yingchuan Li, Stefano Profumo, and Michael Ramsey- ment using -doublet levels in pbo. Phys. Rev. A,
Musolf. Bino-driven electroweak baryogenesis with 87:052130, May 2013.
highly suppressed electric dipole moments. Physics Let- [14] H. Loh, K. C. Cossel, M. C. Grau, K.-K. Ni, E. R. Meyer,
ters B, 673(1):95 100, 2009. J. L. Bohn, J. Ye, and E. A. Cornell. Precision spec-
troscopy of polarized molecules in an ion trap. Science,
342(6163):12201222, 2013.

ment. More details about STIRAP can be found in [12].

14 Various improvements to the apparatus have recently resulted in
over a factor of 600 gain in signal for ACME II.