
2
FIG. 2. Tweezer Rearrangement and Internal State Initial-
ization. (a) Probability of creating defect-free molecular ar-
rays via rearrangement. A fit to pn, where nis the array size,
gives a single particle rearrangement fidelity of p= 0.974(1).
(b) Exemplary images of defect-free arrays. (c) Optical pump-
ing (orange arrow) prepares molecules in |Di. Microwave
sweeps (dashed green arrows) transfer |Dimolecules to |↑i.
PREPARING PAIRS OF LASER-COOLED
MOLECULES AND INITIALIZING THEIR
INTERNAL STATES
Our work starts with single laser-cooled CaF molecules
trapped in a dynamically reconfigurable array of optical
tweezer traps [19, 25]. Through a series of steps involv-
ing laser-cooling, optical trapping and transport, single
molecules are transferred from a magneto-optical trap
into a 1D array of 37 identical optical tweezer traps with
a uniform spacing of 4.20(6) µm. Since our laser-cooling
scheme relies on a closed optical cycle present only for the
X2Σ(v= 0, N = 1) in CaF [38], the molecules initially
loaded into the tweezers occupy a single rovibrational
manifold.
To remove the randomness in tweezer occupation, we
use a rearrangement approach pioneered in neutral atom
experiments [26, 27]. We non-destructively detect the
tweezer occupations using a variant of Λ-imaging [39].
The empty tweezers are identified, switched off, and the
remaining occupied tweezers are then rearranged into the
desired 1D pattern. We characterize the rearrangement
procedure by measuring the probability of successfully
creating uniform arrays, and find a single particle rear-
rangement fidelity of 97.4(1)%. As shown in Fig. 2a, we
are able to create uniform arrays up to a size of 16 with a
probability >0.6. The rearrangement fidelity is limited
by the non-destructive detection fidelity, with minimal
loss (0.2(10)%) due to movement of the tweezer traps.
After rearrangement, we initialize the internal state
of the molecules, which are distributed among the 12
hyperfine states in the X2Σ(v= 0, N = 1) rovibra-
tional manifold. To prepare molecules into a single hy-
perfine state, we optically pump molecules into |Di=
X2Σ(v= 0, N = 1, J = 3/2, F = 2, mF= 2). Subse-
quent microwave sweeps along with an optical clean-out
pulse transfer the molecules into the target final state
|↑i =X2Σ(v= 0, N = 1, J = 1/2, F = 0, mF= 0)
(Fig. 2c). The overall fidelity of preparing molecules in
|↑i is 82.4(11)%. Our preparation sequence ensures that
the dominant preparation error is in the form of unoccu-
pied tweezers, with a small contribution (≈1%) coming
from molecules prepared in the incorrect internal state
|+i=X2Σ(v= 0, N = 0, J = 1/2, F = 1, mF= 1). The
state initialization errors come from imperfect microwave
transfer, polarization impurity of the optical pumping
light, and loss due to heating in the tweezer traps.
PROBING SINGLE-MOLECULE COHERENCE
In order to produce entanglement via the dipolar in-
teractions between molecules, we require long coher-
ence times compared to the typical interaction timescales
of h/J ∼10 ms at our tweezer separations. Achiev-
ing long coherence times for optically trapped molecules
has been an ongoing experimental challenge with steady
advances being made. For molecules, different inter-
nal states can experience different trapping potentials,
which, in combination with motion due to finite temper-
ature, can lead to decoherence. For 1Σ bialkali molecules,
long coherence times of different nuclear spin states
have been reported [40, 41], and work using “magic”
trapping conditions have demonstrated extended coher-
ence times between rotational states in both 1Σ and 2Σ
molecules [20, 42–45].
Since the effective spin-exchange interactions couple
different rotational states, we desire long rotational co-
herence times between the two interacting states |↑i and
|↓i =X2Σ(v= 0, N = 0, J = 1/2, F = 1, mF= 0).
Building upon previous work in CaF [45], we identify a
pseudo-magic trapping condition where both spin states
experience approximately identical trapping potentials.
Our pseudo-magic condition takes into account vector
and tensor shifts and is achieved by applying a magnetic
field orthogonal to the tweezer light polarization at a re-
duced tweezer depth compared to that used for initial
loading and imaging.
To measure the resulting coherence time, we prepare
tweezer pairs with only one molecule initialized in |↑i.
We next apply a Ramsey pulse sequence consisting of
two π/2 microwave pulses (first pulse along ˆx, second
pulse along ˆn= cos θˆx+ sin θˆy) separated by a variable
free evolution time (Fig. 3a). The remaining fraction of
|↑i molecules, P↑, oscillates as a function of θ, with the
oscillation amplitude directly measuring the coherence.
Fitting to an exponential decay curve yields a bare coher-
ence time T∗
2of 2.5(3) ms. Adding a spin-echo improves
the coherence time to T2= 29(2) ms. Following previous
work exploring dipolar interactions of KRb molecules in