University of California Santa Barbara Spin-dynamics and relaxation in Posner Molecules A dissertation submitted in partial satisfaction of the requirements for the degree Bachelor of Science in Physics by Vincent Hou Committee in charge: Professor Matthew P.A. Fisher, Chair June 2020
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Consider a counterclockwise rotation of 2π/3 about the geometric symmetry axis z
(Fig. 2.1). The nuclear spins undergo a counterclockwise permutation after the rotation.
5
Basics of a Posner molecule Chapter 2
h- plane
h+ plane
(a)
ϕx
y
Cube center
Cube facial center
(b)
Figure 2.1: Geometry of Posner molecule. Each phosphate ion resides on the center ofa cube face. The threefold symmetry axis is chosen to be the z-axis, and highlightedas the black arrow in Fig. 2.1a. The cube diagonal in [−1, 0, 1] direction is pickedto be the x-axis of the internal frame and y-axis is then determined with right-handrule. Two equilateral triangular planes (h+ and h−) each formed by three phosphateions are perpendicular to the z-axis. The initial position of the triangles each hasone vertex with x-coordinate being zero and the corresponding opposite side parallelto the y-axis. Both triangle pairs are invariant under rotations by 2π/3 around thesymmetry axis, a consequence of the C3 symmetry. In Fig. 2.1b, the molecule isrotated by angle φ from its initial position. The long dashed line indicates a rotationof 2π/3 such that the triangle pair remains invariant.
We denote this permutation by operator C3, such that it transforms the Bcomp elements
as
C3 |m1m2m3m4m5m6〉 = |m3m1m2m6m4m5〉 . (2.1)
label the vertices from 1. . . 6 in Fig. 2.1. Since (C3)3 = 1, C3 has eigenstates with
eigenvalues ωτ where ω = ei2π/3 and τ = 0,±1. We define τ as the pseudospin quantum
number. The Posner dynamics conserve C3. We therefore seek for C3 eigenstates to be
the states that store quantum information [13]. Our convention of choosing eigenstates is
6
Basics of a Posner molecule Chapter 2
somewhat arbitrary, but convenient. We choose C3, Szlab1...6,2 and S 2
1...63 to be the complete
set of commuting observables that fully breaks the degeneracies [11]. The eigenspaces Hτ
have 24-, 20-, and 20-fold degeneracies for τ =0, 1 and -1 respectively. Hence we classify
the 64 eigenstates into 3 sectors depending on τ .
Here we present a detailed demonstration of this analysis within a simple, fictitious
molecule of three phosphorous nuclear spins residing on the vertices of a perfect triangle,
say the one on the h+ plane in Fig. 2.1a. The triple-spin plane corresponds to an 8-
dimensional Hilbert space C8. C3, Szlab123 , and S 2123 share a total of 8 eigenstates in Table
2.1.
State Decomposition S123 szlab123 τ
|1〉 |000〉 3/2 3/2 0|2〉 1√
3(|011〉+ |101〉+ |110〉) 3/2 1/2 0
|3〉 1√3(|100〉+ |010〉+ |001〉) 3/2 -1/2 0
|4〉 |111〉 3/2 -3/2 0
|5〉 1√3(|011〉+ ω2 |101〉+ ω |110〉) 1/2 1/2 1
|6〉 1√3(|100〉+ ω2 |010〉+ ω |001〉) 1/2 -1/2 1
|7〉 1√3(|011〉+ ω |101〉+ ω2 |110〉) 1/2 1/2 -1
|8〉 1√3(|100〉+ ω |010〉+ ω2 |001〉) 1/2 -1/2 -1
Table 2.1: The three nuclear spins correspond to an 8-dimensional Hilbert space. Aconvenient choice of eigenstates is the eigenbasis shared by S2
123, Szlab123 and C3. Allstates’ eigenvalues of each operator are listed in sequence in the table.
Coupling between phosphorus nuclear spins arises due to two factors: J-coupling,
which is an indirect interaction between two nuclear spins; and the magnetic dipole-
dipole interaction. The Hamiltonian of the J-coupling has an effective Heisenberg-like
2The internal frame rotates relative to the lab frame denoted by the subscript lab. Measurementsof spins is conducted under this reference frame. So we define operator Szlab
1...6 =∑6i=1 1
⊗(i−1) ⊗ Szlabi ⊗
1⊗(6−i), where 1 is the 2-by-2 identity matrix.
3The total-spin operator is defined as follows: S21...6 = (S1 +S2 + · · ·+S6)2 = (
∑6i=1 1
⊗(i−1)⊗S i⊗1⊗(6−i))2.
7
Basics of a Posner molecule Chapter 2
form of [11]
HJ =∑〈i,j〉
Jij · S i · S j. (2.2)
The pair-dependent interaction strength is denoted by Jij. HJ encompasses C3 symmetry,
SU(2) symmetry, and time-reversal symmetry. It is also time-independent, such that the
total-spin S1...6, z component of total-spin szlab1...6 and pseudospin τ are all conserved. Hence
we express the eigenstates using four quantum numbers,
|α〉 = |E, S1...6, szlab1...6, τ〉 , (2.3)
where E is the J-coupling energy (in Hz) and α = 1, 2, . . . , 64. The Hamiltonian of
dipole-dipole interaction is given by,
HD =∑〈i,j〉
JD,ij · [S i · S j − 3(S i · r̂ ij)(S j · r̂ ij)], (2.4)
with JD,ij = µ0γ2~4πr3ij
denoting the dipolar coupling strength. HD has time-reversal sym-
metry only, and conserves neither spin nor pseudospin. We therefore consider HD to be
responsible for both spin and pseudospin relaxations of the initial/eigenstates 4. HD is
also orientation-dependent, which is therefore also time-dependent when the molecule
undergoes rotational dynamics. The dynamics is therefore “stochastic”. For that reason,
we do not expect oscillations/revivals in observables (spin and pseudospin). Rather, we
expect exponential decay/relaxation of these observables to their values in equilibrium.
It is therefore possible to define a “relaxation time” for these observables.
4Here we neglected dipolar interactions contributed by surrounding molecules/ions mainly becausethey have larger inter-particle distances with the phosphorus nuclei. Coupling strength JD,ij is inverselyproportional to the inter-particle distances by a power of three. Therefore, dipolar interactions inducedby particles outside the Posner molecule is expected to be unimportant compared to intra-moleculardipolar interactions. The relaxation time can be as long as 21 days [11], much larger than our estimation(Sec. 4.3), indicating its insignificance in the relaxation process.
8
Chapter 3
Dynamical model of a single rotating
Posner molecule
3.1 Rotational motion of the molecule (time-dependence
of HD)
In a microscopic view, the motion of the molecule is composed of segments of ballistic
rotation, where the angular speed depends on the thermal energy of the Posner molecule;
and the change of directions between consecutive ballistic “duration”s, attributed to
collisions with the surrounding particles (e.g. water molecules). These can be summarized
as a discrete-time stochastic process made up by ballistic “hop”s (Fig. 3.1).
Let p be a phosphorus nucleus’ positional vector in the internal frame. At t = 0,
the internal frame is the same as the lab frame. Then the internal frame starts to rotate
randomly relative to the lab frame around its origin being a fixed point. It is convenient
to construct a random rotation matrix R that acts on p to determine its coordinate in
the lab frame. The random rotation matrix consists two parts: pick a random rotational
9
Dynamical model of a single rotating Posner molecule Chapter 3
axis, and perform the rotation by a random angle θ, generating a total of three degrees
of freedom: two from the selection of rotational axis and one from the random angle
rotated.
The selection of random rotational axis is done by performing a random SO(3) matrix
Q (App. A.2). This rotational axis is treated as the z-axis, such that an angle of ∆θ
is rotated around this axis in the clockwise direction. Let A denote the rotation matrix
around z-axis,
A =
cosθ sinθ 0
−sinθ cosθ 0
0 0 1
. (3.1)
Performing a linear transformation such that this rotation is performed around the ran-
dom z-axis picked by the random SO(3) matrix, we end up with
R = QTAQ. (3.2)
We further refine this model with several parameters describing the stochastic ballistic
angle, the angular velocity, and time duration for each discrete hop. Stochastic ballistic
angle ∆θ is a random variable with probability distribution of Uniform(−θb, θb), where
θb notates the largest possible ballistic angle swept by the molecule before it changes
rotational direction. The stochastic ballistic angle ∆θ is swept by a continuous rotation
(Fig. 3.1) at angular speed ωT determined by thermal energy, such that,
1
2Iω2
T =3
2kBT. (3.3)
The stochastic ballistic time ∆t taken by each hop is given by ∆θ/ωT . For clarity
we also introduce the ballistic time constant tb = |θb|/ωT . We conclude that ∆t ∼
10
Dynamical model of a single rotating Posner molecule Chapter 3
Uniform(0, tb).
(a) (b)
Figure 3.1: Here we present a simplified 2-dimensional demonstration of the rota-tional motion model. The arrowed vectors represent the orientations of the Pos-ner molecule at t0, t1, and t2, where ti+1 = ti + ∆ti+1. The tail/beginning of theorientation-vector is fixed at the center of a unit sphere, and the head/end of thevector lands on the surface of the sphere. Each stochastic hop sweeps an angle of∆θi ∼ Uniform(−θb, θb). Each hop is a continuous rotational motion with angularspeed ωT , as demonstrated in Fig. 3.1b. Stochastic ballistic time ∆t has probabilitydistribution |Uniform(−θb, θb)|/ωT ≡ Uniform(0, tb).
When viewed in the long-time limit , there are three limiting cases of θb:
• When θb → 0, the model turns into a Brownian motion/rotational diffusion.
• When θb → π, the model turns into a discrete jump where the landing points of the
orientation-vector’s head are distributed uniformly on the surface of a unit sphere.
• When θb →∞, the model turns into a ballistic motion.
However, we notice that the limits of this motion are not directly related to pseu-
dospin/spin dynamics (Sec. 3.2).
11
Dynamical model of a single rotating Posner molecule Chapter 3
3.2 Pseudospin/spin dynamics
In this thesis, all Hamiltonians are written in units of Hz, such that ~ does not appear
in time evolution operators. The strengths of the two interactions described in Sec. 3.1
depends on constants J and JD respectively. The total Hamiltonian H(t) that acts on
the intial state is
H(t) = HJ +HD(t), (3.4)
1 such that the time-evolution operator has the form
U(t, t0) = T e−i∫ tt0dt′H(t′)
. (3.5)
The time-ordering operator in Equation 3.5 is essential because [HD(ti), HD(tj)] 6= 0.
There are no simple ways in which to handle the time-ordered exponentials given the
complexity of HD’s dependence on time. So we go for the form of direct numerical
integration which evaluates the propagation of a wavefunction over a time period t by
discretizing time into n small steps of width δt = t/n, assuming the change of the
system is small in each step. The time-evolution operator is then simplified with the
1For the Posner molecule, the dipolar coupling strength JD (≈ 988.8 Hz, shown in Sec. 4.3) is severalorders of magnitude greater than the J-coupling strength J (computation by Swift shows J1 = 0.178 Hz,J1 = 0.145 Hz and J3 = −0.003 Hz, each representing J-coupling strength of nearest-neighbor, second-nearest-neighbor and third-nearest-neighbor) [11]. We therefore neglect the influence from J-couplingson the initial states in the numerical simulations.
12
Dynamical model of a single rotating Posner molecule Chapter 3
Such that the time operator over time t can be written as a product of n operators over
the small intervals,
U(t, t0) = limδt→0
[Un · · ·U2U1] = limn→∞
n∏k=1
Uk, (3.7)
where time propagation over the kth step/interval is
Uk = e−iδtH(tk−1); tk = kδt. (3.8)
The final state after an evolution of time t is
|α, t0; t〉 = U(t, t0) |α, t0〉 = limn→∞
n∏k=1
e−iH(tk−1)δt |α, t0〉 . (3.9)
For a single ballistic motion that sweeps an angle of ∆θ (Fig. 3.1), the time taken is
∆t = ∆θ/ωT . Hence in numerical integration form, the total steps/intervals n = ∆t/δt,
such that
|α, t0; t0 + ∆t〉 = limδt→0
∆t/δt∏k=1
e−iH(tk−1)δt |α, t0〉 . (3.10)
Suppose the stochastic process runs for a total ofm steps, and denote ∆θ1,∆θ2, . . .∆θmi.i.d.∼
Uniform(−θb, θb) as the sizes of the hops, while ∆tj = ∆θj/ωT . Then we arrive at a full
expression of the time propagation with the defined model (Sec. 3.1),
∣∣∣∣∣α, t0; t0 +m∑j=1
∆tj
⟩= lim
δt→0
1δt
∑mj=1 ∆tj∏
k= 1δt
∑m−1j=1 ∆tj
e−iH(tk−1)δt
· · ·∆t1/δt∏
k=1
e−iH(tk−1)δt
|α, t0〉 ,(3.11)
It is unrealistic to push δt to the infinitely-near-zero regime in numerical simulations due
to both obstacles coming from the highest precision a numerical quantity may have and
13
Dynamical model of a single rotating Posner molecule Chapter 3
the total computation time. Therefore, δt is taken as a small enough time comparing to
the ∆t (App. A.1).
3.3 Heuristic picture of spin/pseudospin dynamics
Consider a spin on a Bloch sphere, where the north and south poles are chosen to be
the spin-up |0〉 and spin-down |1〉 states. First consider placing this spin in an external
field that continues to be turned on and off frequently, each time pointing towards a
random direction in 3-dimensional space. Then consider a single phosphorus spin in a
Posner molecule, which continues to rotate randomly, also, in 3-dimensional space. Since
both cases have stochastic Hamiltonian changing with time - one is an external field, on is
the dipolar coupling, we can draw an analogy between both cases. We provide the single
spin analysis of the simple model (random field) and reach a scaling relation which we
hope to also be valid for the original model (random rotational dynamics of the Posner
molecule). The spin state then will precess an angle of ϕ on the Bloch sphere, towards
around the direction of the field, by an angle of
ϕ ∼ JDτc, (3.12)
where τc is the time that the field stays on in this direction. The Bloch vector of the spin
performs a random walk on the surface of a unit sphere as the field is turned on and off
in random directions. The spin state can be recognized as completely relaxed when its
spin direction is flipped (a total precession of π). Following the random walk argument,
it takes an average of (π/ϕ)2 steps for the spin state to flip entirely (e.g. from |0〉 to
|1〉), where each step consumes a time of τc. Therefore the average time needed for a
complete relaxation is ( πJDτc
)2τc, suggesting that the spin and pseudospin relaxation time
14
Dynamical model of a single rotating Posner molecule Chapter 3
has scaling form
T〈Sz〉 ∼ J−2D τ−1
c ; Tτ ∼ J−2D τ−1
c , (3.13)
where this result is consistent with the Solomon-Bloembergen equation [14].
The definition for τc can be arbitrary. The only significance is its dependence on θb
and tb, coming from our rotational dynamics model. We define τc to be the rotational
correlation time of the Posner molecule, the average time it takes for it to rotate φ radians
2. We establish a rough relationship between τc and {θb, tb}. Begin with describing the
stochastic process described in Sec. 3.1. Let θn denote the mean-square angular deviation
of the rotational diffusion,
〈θ2n〉 = n〈∆θ2〉, (3.14)
where n denotes the number of stochastic steps (Sec. 3.1). The probability distribution
of ∆θ is Uniform(−θb, θb), such that
〈∆θ2〉 = V ar(∆θ) + 〈∆θ〉2 =[θb − (−θb)]2
12+ 0 =
θ2b
3. (3.15)
Each hop takes time ∆t to finish. Since ∆t ∼ Uniform(0, tb), the average duration of n
stochastic steps is
n〈∆t〉 = ntb/2, (3.16)
Now suppose
〈θ2n〉 = φ2, (3.17)
where φ is a constant number. Then the time to reach a mean-square angular deviation
of φ2 is
n〈∆t〉 =3φ2
θ2b
· tb2
=3φ2
2
tbθ2b
. (3.18)
2φ can be defined arbitrarily, but usually taken as 1
15
Dynamical model of a single rotating Posner molecule Chapter 3
For φ being a small number (the mean-square angular deviation is small), the orientation
of the molecule is approximately not changed, which is equivalent to the field being on
in one direction for time τc. Hence we conclude that
τc =Ktbθ2b
=K
ωT θb, (3.19)
with K being a tunable order-one constant that depends on the exact choice of φ’s
magnitude. The scaling form is
τc ∼ ω−1T θ−1
b . (3.20)
Define scaling parameters C〈Sz〉 and Cτ . Combining with Equation 3.13, we arrive at a
universal scaling form of
T〈Sz〉 =C〈Sz〉θbωT
J2D
; Tτ =CτθbωTJ2D
. (3.21)
For this Bloch sphere picture to be valid, we require
∆ϕ ∼ JDθbωT
<< 1, (3.22)
which corresponds to certain regimes of θb, ωT and JD in numerical simulations. The
constraint equation is
θb >> JD/ωT . (3.23)
This condition (Equation 3.22) is supposed to be satisfied by the real Posner molecule.
We expect common values of θb to fall in the range of π/10 ∼ π. Based on that, with
parameters JD ≈ 988.8 Hz and ωT = 3.2 × 1011 rad s−1 (details in Sec. 4.3), ∆ϕ has a
range of 10−9 to 10−8 that is much less than 1.
Our approach in this paper is to confirm the universal scaling form above (with
16
Dynamical model of a single rotating Posner molecule Chapter 3
numerically accessible sets of parameters constrained by Equation 3.23), and extrapolate
to parameters for a real Posner molecule (these parameters are dramatic/impossible to
simulate numerically), thereby getting a semi-quantitative estimation of T〈Sz〉 and Tτ .
T〈Sz〉 can, in principle, be directly measured in NMR experiments, whereas one can infer
the other (Tτ ).
17
Chapter 4
Results of numerical simulations
4.1 Results on spin relaxation
We choose the initial spin states in similarly as the initial pseudospin states (Sec.
2.2), except that spin states are not necessarily the eigenstates of C3, therefore
|sα〉 = |E, S1...6, Szlab1...6〉 . (4.1)
In principle, each initial state would have a decaying rate (Fig. 4.1a). As mentioned in
Sec. 2.2, we expect observable 〈Szlab1...6〉 (denoted as 〈Sz〉 in this chapter for simplicity) to
perform exponential relaxation, therefore,
〈Sz〉(t) = e−t/T〈Sz〉〈Sz〉(t = 0). (4.2)
This statement is supported by numerical simulations as shown in Fig. 4.1b. Despite
some noisy peaks around the equilibrium value, which is 〈Sz〉 = 0, the simulated data is
linear on the log-linear scale.
There are 64 initial spin states in total. Listing the decaying rates for all initial states
18
Results of numerical simulations Chapter 4
t [sec]
0.00 0.01 0.02 0.03 0.04
JD= 200 Hz
θb= π
-3
-2
-1
0
1
2
3
⟨Sz⟩
3.0-3.02.0-2.01.0-1.00.0
⟨Sz⟩(t=0)
(a)
t [sec]
0.00 0.01 0.02 0.03 0.04
JD= 200 Hz
θb= π
e-10
e-5
e0
e5
|⟨Sz⟩|
3.0-3.02.0-2.01.0-1.0
⟨Sz⟩(t=0)
(b)
Figure 4.1: 〈Sz〉(t) of different initial spin states, presented in linear-linear scale andlog-linear scale. The numerical inputs are ωT = 1000π rad s−1, JD = 200 Hz, andθb = π rad. In Fig. 4.1b, absolute values are taken in order to present the negative Szdata. Spin states with initial 〈Sz〉(t = 0) = 0 performs no relaxation, so are excludedin the log-linear plot (Fig. 4.1b). Results of numerical simulations provide a strongsupport on our assumption on the relaxation being exponential. The equilibrium valueis shown to be 〈Sz〉 = 0. When the relaxation approach the equilibrium value, oursimulation becomes less accurate. It is not possible for the computer to output valuesthat are infinitely close to 0. Hence peaks of noise/fluctuation occur at the tail in Fig.4.1b.
is not very helpful. However, we can define a “total”/“effective” rate, when starting
from a mixed state of all 64 eigenstates. For concreteness, we choose a state as prepared
by applying a magnetic field at finite temperature, and the canonical-ensemble partition
function is
Z =64∑α=1
e−βhszα . (4.3)
19
Results of numerical simulations Chapter 4
The probability of the mixed state is in state |sα〉 is
pα =e−βhs
zα
Z. (4.4)
The net magnetization in z-direction of the mixed state at time t is
Mz(t) =∑α
pαe−t/tαszα, (4.5)
where tα is the spin lifetime of the state |sα〉. By Taylor expansion in the small parameter
βh� 1, e−βhszα ≈ 1− βhszα, we have
Mz(t) ≈ −βh
64
∑α
(szα)2e−t/tα . (4.6)
For t� tα, before the state is fully relaxed, we can also Taylor expand the e−t/tα term,
Mz(t) = −βh64
∑α
[(szα)2
(1− t
tα+ · · ·
)]; (4.7)
in particular, Mz(0) = −βh64
∑α (szα)2. We also expect that
Mz(t) = Mz(0)e−t/T〈Sz〉 = −βh64
(∑α
(szα)2
)·[1− t
T〈Sz〉+ · · ·
]. (4.8)
By Equation 4.7 and 4.8, the total/effective rate is given by
1
T〈Sz〉=
∑64α [(szα)2/tα]∑64
α (szα)2. (4.9)
We acquired data by time propagating all 64 spin states respectively with fixed parameter
ωT = 1000π rad s−1, inputing different sets of JD and θb within the regime indicated in
20
Results of numerical simulations Chapter 4
Sec. 3.3 (Equation 3.22). We collect data of individual relaxation times tα from numerical
simulations and then compute T〈Sz〉 with Equation 4.9.
θbωT/JD
2 [sec]
0 1 2 3 4
C1=0.12
C2=0.11
C3=0.1
C4=0.1
C5=0.1
C6=0.09
C7=0.08
C8=0.08
C9=0.07
C10=0.07
Slope rate
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
0.0
0.1
0.2
0.3
0.4
0.5
T⟨Sz⟩ [sec]
(a)
θbωT/JD
2 [sec]
10-3
10-2
10-1
100
101
C=0.09
Confdence Interval=[0.12, 0.07]
Total/effective rate
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
10-4
10-3
10-2
10-1
100
101
T⟨Sz⟩ [sec]
(b)
Figure 4.2: Data-collapse of relaxation time T〈Sz〉. Angular velocity ωT is fixed at1000π rad s−1. Sets of JD and θb values are marked distinctively in shape-keys andcolor-keys. Individual relaxation times tα are estimated by the best fit slopes of the inthe log-linear scale (presented in Fig. 4.1b). Collections of tα’s are used to computethe total/effective relaxation time T〈Sz〉(JD, θb) by Equation 4.9, marked by coloredshapes in both figures. Scaling parameter C is weakly dependent on θb (Fig. 4.2a).The parameter C falls in the range [0.07, 0.12]. The representative scaling parameteris C = 0.09.
We pick 6 values for JD ranging from 50 Hz to 500 Hz, and 10 values for θb from a range
of π/10 to π. In Fig. 4.2a, we plot T〈Sz〉 against θbωT/J2D and fit each group of data with
the same θb. All the fitted lines are straight lines passing through the origin, declaring
the linear relationship as expected. The slopes are weakly dependent on θb1. Within the
range we choose for θb, the slope rates do not differ from each other much. Therefore we
can still structure a data-collapse that uses all data for the linear regression, presented
1Equation 3.19, which we use to obtain the universal scaling form (Equation 3.21) is based on theassumption that the rotational dynamical is approximately diffusive, meaning that θb → 0. When θbgets larger in magnitude, Equation 3.19 may not necessarily hold. Hence it is reasonable to see suchdependency.
21
Results of numerical simulations Chapter 4
in Fig. 4.2b, on a log-log scale. The slope in Fig. 4.2b is 1, indicating the linearity, while
the offset demonstrates the size of scaling parameter C. The rough confidence interval
of C is [0.07, 0.12]. Representative rate C = 0.09 gives an approximation of the scaling
parameter C〈Sz〉 in Equation 3.23.
4.2 Results on pseudospin relaxation
Pseudospin τ is not a directly measurable observable. However, pseudospin relaxation
of an initial state can be measured by the probability amplitude of it being in its initial
τ -sector. Suppose the ensemble is in a mixed state such that each of the pure states α
occurs with probability pα. Then the corresponding density operator has form
ρ =∑α
pα |α〉 〈α| . (4.10)
Denote the projection operator in τ -sector as P̂τ and |m; τ〉 as the eigenstate in the sector,
P̂τ =∑m
|m; τ〉 〈m; τ | . (4.11)
The expectation value of this measurement can be calculated by
pτ = Tr[P̂τρ]. (4.12)
The sectors have 24, 20, and 20 states for τ = 0, 1and − 1 respectively (Sec. 2.2).
Therefore, a completely relaxed state (defined by pα = 1/64 for all α) has pτ=0 = 24/64
and pτ=±1 = 20/64.
We choose initial states to be maximally mixed within the sector for numerical sim-
ulation - that is, pα = 1/24 for α in the τ = 0 sector, and pα = 1/20 for α in the τ = ±1
22
Results of numerical simulations Chapter 4
t [sec]
0.00 0.01 0.02 0.03 0.04 0.05
JD= 200 Hz
θb= π
0.0
0.2
0.4
0.6
0.8
1.0
Pτ
-110
τ(t=0)
(a)
t [sec]
0.00 0.01 0.02 0.03 0.04 0.05
JD= 200 Hz
θb= π
e-6
e-5
e-4
e-3
e-2
e-1
e0
Pτ -
Pτ(t→∞)
-110
τ(t=0)
(b)
Figure 4.3: Pτ (t) of different initial pseudospin states, presented in linear-linear scaleand log-linear scale. The numerical inputs are ωT = 1000π rad s−1, JD = 200 Hz,and θb = π rad. Initial states are maximally mixed within each τ -sector. The prob-ability functions are shifted exponential decays. The eigenspaces have 24-, 20-, and20-fold degeneracies for τ =0, 1 and -1 respectively (Sec. 2.2). Therefore, theoreticalequilibrium values of Pτ are P0(t → ∞) = 24/64 and P±1(t → ∞) = 20/64. In Fig.4.3b, the theoretical equilibrium values are subtracted in order to reveal the linearity.Results of numerical simulations provide a strong support on our assumption on therelaxation being exponential. Similar to 〈Sz〉 data (Fig. 4.1b), when the relaxationapproaches the equilibrium value, our simulation becomes less accurate. Hence largernoise/fluctuations occur at the tail in Fig. 4.3b.
23
Results of numerical simulations Chapter 4
sector. Time propagating density matrix of τ -sector initial state is
ρτ (t) = U(t, t0)
[∑α
pα |α; τ〉 〈α; τ |
]U(t, t0)†. (4.13)
Hence we numerically evaluate
pτ (t) = Tr[P̂τρτ (t)]. (4.14)
We expect pτ to be a shifted exponential decay (Sec. 2.2),
pτ (t) = pτ (t→∞) + (1− pτ (t→∞))e−t/Tτ , (4.15)
where pτ (t→∞) is the equilibrium probability amplitude. This statement is supported
by numerical simulations as shown in Fig. 4.4. Despite some noisy peaks around the
equilibrium value, the modified simulated data is linear on the log-linear scale (Fig. 4.4b).
A data-collapse of 0-sector state is presented in Fig. 4.4. Similar to 〈Sz〉 data, we pick
6 values for JD ranging from 50 Hz to 500 Hz, and 10 values for θb from a range of π/10 to
π 2. The linearity presented by the numerical results significantly support out universal
scaling form (Equation 3.23). In Fig. 4.4a, we plot Tτ against θbωT/J2D and fit each group
of data with the same θb. Like T〈Sz〉 data, all the fitted lines are straight lines passing
through the origin, declaring the linear relationship as expected. The slopes are weakly
dependent on θb. The rough confidence interval of C is [0.04, 0.08]. Representative rate
C = 0.06 gives an approximation of the scaling parameter Cτ in Equation 3.23. Notably,
Cτ scales similarly as C〈Sz〉. This conclusion implies the possibility of using experimentally
measurable spin relaxation time to predict unmeasurable pseudospin relaxation time.
2See Appendix A.3 for ±1-sectors.
24
Results of numerical simulations Chapter 4
θbωT/JD
2 [sec]
0 1 2 3 4
C1=0.08
C2=0.07
C3=0.07
C4=0.06
C5=0.06
C6=0.06
C7=0.05
C8=0.05
C9=0.04
C10=0.04
Slope rate (τ=0)
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
0.00
0.05
0.10
0.15
0.20
0.25
Tτ [sec]
(a)
θbωT/JD
2 [sec]
10-3
10-2
10-1
100
101
C=0.06
Confdence Interval=[0.08, 0.04]
Sector τ=0
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
10-4
10-3
10-2
10-1
100
101
Tτ [sec]
(b)
Figure 4.4: Data-collapse of relaxation time Tτ=0. Angular velocity ωT is fixed at1000π rad s−1. Sets of JD and θb values are marked distinctively in shape-keys andcolor-keys. Each data point Tτ (JD, θb) is computed by Equation 4.9. Individualrelaxation time (data point) is estimated by the best fit slope in the log-linear scale(presented in Fig. 4.3b). Scaling parameter C is weakly dependent on θb (Fig. 4.4a).The parameter C falls in the range [0.04, 0.08]. The representative scaling parameteris C = 0.06.
4.3 Extrapolation towards the reality
We define dipolar coupling strength JD (Sec. 2.2) as
JD,ij =µ0γ
2~4πr3
ij
. (4.16)
25
Results of numerical simulations Chapter 4
Exact values of parameters and constants are as follows:
µ0 = 4π · 10−7[H m−1] [15] (4.17)
~ = 1.05457 · 10−34[J s] [15] (4.18)
rij ≈ 4.75 · 10−10 ∼ 5.65 · 10−10[m] [16] (4.19)
γP = 108.291 · 106[rad s−1 T−1] [17] (4.20)
Hence we compute JD ≈ 988.8 Hz. By Equation 3.3, with a moment of inertia 1.22×10−43
kg m2 [11], ωT = 3.2×1011 rad s−1 at 300 K. θb is a parameter remains unknown, which we
believe should fall in the regimes of π/10 to π. By Equation 3.21, substituting the scaling
parameters estimated from the numerical simulations, we arrive at spin relaxation time
T〈Sz〉 ≈ 2.5 ∼ 25 hrs and pseudospin relaxation time Tτ ≈ 1.7 ∼ 17 hrs. The extrapolated
numbers show that the spin/pseudospin relaxation times of a Posner molecule is in the
scale of hours, which is a fairly long time comparing to the long believed assumption that
quantum coherence is short-lived. The long-lived spin/pseudospin states in the Posner
molecule surely can play a key role in the “quantum brain” idea [6].
26
Appendix A
Numerical simulation method
A.1 Numerical integration
We choose time interval δt = tb/10 for the numerical integration. A great advantage
of doing this is that the average computation time of one integration over a hop is the
same for all tb, as we do not want to have the total running time for θb = π to be 10
times as long as the θb = π/10 data.
Computer works only with an integer number of intervals in the numerical integration.
Since ∆t is a random variable with distribution Uniform(0, tb), we are not guaranteed
that ∆t/δt always ends up as an integer. So we generate random integers n, which is
the total number of δt’s that a hop lasts. Since n = ∆t/δt, it should have a theoretical
distribution of Uniform(0, tb/δt) ≡ Uniform(0, 10). Hence we randomly draw a number
from 0 to 10 at each time for the value of n. We first pick a random rotational axis and
then propagate n steps with angular step-size of θb/10 and time step-size tb/10. The
iteration function is
|α, ti〉 = e−iH(ti−1)δt |α, ti−1〉 , (A.1)
27
Numerical simulation method Chapter A
for i = 1, 2, . . . . When i reaches n, we pick again a random rotational axis and propagate
with a new n steps, drawn from 0 to 10, and so on. Each step we measure the spin and
pseudospin expectation 〈Sz〉 and Pτ . The iteration functions are
where |m〉 represents all states in that specific τ sector.
A.2 Uniform random rotation matrix
In order to find random orientations for the Posner, we need to construct a uniform
random rotation matrix R such that rotates the initial z-axis of the lab frame into a
random direction, such that it can serve as the new rotational axis for the molecule.
There are many ways to generate such a matrix, here we will introduce one approach.
Suppose we have a distribution X such that
X = [X1, X2, · · · , Xn] i.i.d. ∼ N(0, 1), (A.4)
then,
X ∼ N(0, In), (A.5)
which is a multivariate normal distribution with mean being 0 and identity covariance
matrix. Then for any rotation matrix Q which is by definition orthogonal,
‖QX‖2 = XTQTQX = ‖X‖2 , (A.6)
28
Numerical simulation method Chapter A
such that the magnitude of the mean is conserved. The covariance of QX is just QInQT =
In, which is also conserved. Hence we conclude that,
QX ∼ N(0, In). (A.7)
Hence the distribution of X is invariant under rotations. Now let Y = X/ ‖X‖2,
QY = QX/ ‖QX‖2 = QX/ ‖X‖2 , (A.8)
for any rotation matrix Q. From that we can conclude that Y is also invariant to rotations
and ‖Y ‖2 = 1. The only one probability distribution for Y that it satisfies both conditions
at the same time is the uniform distribution on a unit sphere. With this in mind, we
can construct a uniform random rotation matrix by picking out 3 uniformly distributed
vectors at random and let them be v1, v2 and v3. Then apply the Gram–Schmidt process
to construct the orthonormal basis,
u1 = v1, e1 = u1/ ‖u1‖ , (A.9)
u2 = v2 −v2 · u1
u1 · u1
, e2 = u2/ ‖u2‖ , (A.10)
u3 = v3 −v3 · u1
u1 · u1
− v3 · u2
u2 · u2
, e3 = u3/ ‖u3‖ . (A.11)
Finally, let each component of the orthonormal basis e1, e2 and e3 be a column of the
matrix R,
R = [e1, e2, e3], (A.12)
and we have obtained a uniform random rotation matrix. The very final step is to
check the determinant of R, which will be either 1 or -1. If the determinant is -1 then
interchange two of its columns. For an SO(3) matrix, we have a simple standard criterion
29
Numerical simulation method Chapter A
for uniformity, namely that the distribution be unchanged when composed with any
arbitrary rotation. Let O be an arbitrary orthogonal matrix, then matrix OR has columns
of Oe1, Oe2 and Oe3. As shown earlier, we know that v1, v2 and v3 are all invariant to
rotations. Thus,
Oe1 =Ov1
‖Ov1‖=Ov1
‖v1‖, (A.13)
Oe2 =Ou2
‖Ou2‖=
Ov2 − Ov2·Ov1
Ov1·Ov1∥∥∥Ov2 − Ov2·Ov1
Ov1·Ov1
∥∥∥ =Ov2 − v2·v1
v1·v1∥∥∥v2 − v2·v1
v1·v1
∥∥∥ , (A.14)
and same reasoning for e3. Therefore we conclude that e1, e2 and e3 are invariant to
rotations and hence also matrix R, meaning that OR has the same distribution as R.
Hence shown R to be a uniform random rotation matrix.
30
Numerical simulation method Chapter A
A.3 Simulation results on pseudopin τ = ±1 sectors
θbωT/JD
2 [sec]
0 1 2 3 4
C1=0.06
C2=0.06
C3=0.06
C4=0.05
C5=0.05
C6=0.05
C7=0.05
C8=0.04
C9=0.04
C10=0.04
Slope rate (τ=1)
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
0.00
0.05
0.10
0.15
0.20
0.25
Tτ [sec]
(a)
θbωT/JD
2 [sec]
10-3
10-2
10-1
100
101
C=0.05
Confdence Interval=[0.06, 0.04]
Sector τ=1
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
10-4
10-3
10-2
10-1
100
101
Tτ [sec]
(b)
Figure A.1: Data-collapse of relaxation time Tτ=1. Angular velocity ωT is fixed at1000π rad s−1. Sets of JD and θb values are marked distinctively in shape-keys andcolor-keys. Individual relaxation time (data point) is estimated by the slope of thebest fit line in the log-linear scale (presented in Fig. 4.3b). Scaling parameter C isweakly dependent on θb (Fig. A.1a). The parameter C falls in the range [0.04, 0.06].The representative scaling parameter is C = 0.05, which differed from the scalingparameter of Tτ=0 by a small amount.
31
Numerical simulation method Chapter A
θbωT/JD
2 [sec]
0 1 2 3 4
C1=0.07
C2=0.06
C3=0.06
C4=0.05
C5=0.05
C6=0.05
C7=0.05
C8=0.04
C9=0.04
C10=0.03
Slope rate (τ=-1)
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
0.00
0.05
0.10
0.15
0.20
0.25
Tτ [sec]
(a)
θbωT/JD
2 [sec]
10-3
10-2
10-1
100
101
C=0.05
Confdence Interval=[0.07, 0.03]
Sector τ=-1
500
400
300
200
100
50
JD
(Hz)
1
2
3
4
5
6
7
8
9
10
θb (π/10)
10-4
10-3
10-2
10-1
100
101
Tτ [sec]
(b)
Figure A.2: Data-collapse of relaxation time Tτ=−1. Angular velocity ωT is fixed at1000π rad s−1. Sets of JD and θb values are marked distinctively in shape-keys andcolor-keys. Individual relaxation time (data point) is estimated by the slope of thebest fit line in the log-linear scale (presented in Fig. 4.3b). Scaling parameter C isweakly dependent on θb (Fig. A.2a). The parameter C falls in the range [0.03, 0.07].The representative scaling parameter is C = 0.05, which is similar to the scalingparameter of Tτ=1.
32
Appendix B
Special features of triangular 3-spin
system
B.1 Non-decaying states under planar rotation around
C3 symmetry axis
Suppose the molecule rotates around the internal z-axis (defined in Sec. 2.2) in the
counterclockwise direction, with a constant angular speed ω (see the demonstration in
Fig. 2.1b). The time-dependent position vectors are
r1(t) = (r, ωt), (B.1)
r2(t) = (r,2π
3+ ωt), (B.2)
r3(t) = (r,4π
3+ ωt)). (B.3)
Let φi be the initial polar angles of the nucleus with label i, e.g. φ2 = 2π3
. We rewrite
the Hamiltonian of dipole-dipole interaction (see Equation 2.4) using Sz, S+ and S−
33
Special features of triangular 3-spin system Chapter B
operators,
HD = JD∑〈i,j〉
[Szi Szj −
1
4(S+
i S−j + S−i S
+j ) +
3
4S+i S
+j e−i(φi+φj+2ωt) +
3
4S−i S
−j e
i(φi+φj+2ωt)].
(B.4)
Now define modified ladder operators,
S̃±i = S±i e−iωt, (B.5)
we can then write down a unitary operator Ui, such that for any states |a〉 and |b〉,
〈a|U†iS±i Ui |b〉 = 〈a| S̃±i |b〉 , (B.6)
〈a|U†iSzi Ui |b〉 = 〈a|Szi |b〉 . (B.7)
By the fact that
S±i e−iωt = e−iωtS
zi S±i e
iωtSzi , (B.8)
and
e−iωtSzi Szi e
iωtSzi = e−iωtSzi eiωtS
zi Szi = Szi , (B.9)
we find
Ui = eiωtSzi /~. (B.10)
Define U =∏
iUi. In the 3-spin case,
U(t) = U1U2U3 = eiωt(Sz1+Sz2+Sz3 ) = eiωtS
z123 (B.11)
34
Special features of triangular 3-spin system Chapter B
Hence we have found a unitary transformation that describes the planar rotation around
z-axis. We are able to define time independent Hamiltonian H ′D
H ′D = JD∑〈i,j〉
[Szi Szj −
1
4(S+
i S−j + S−i S
+j ) +
3
4S+i S
+j e−i(φi+φj) +
3
4S−i S
−j e
i(φi+φj)]. (B.12)
such that after the unitary transformation,
HD(t) = U †(t)H ′DU(t), (B.13)
while time independent J-coupling Hamiltonian is invariant under the transformation,
HJ = U †(t)HJU(t). (B.14)
The total Hamiltonian of the system is
H(t) = HJ +HD(t) = U †(t)(HJ +H ′D)U(t). (B.15)
Its discretized form is
H(δt) = HJ +HD(δt) = U †(δt)(HJ +H ′D)U(δt). (B.16)
In Sec. 3.2, we have defined the time-evolution operator such that the final state after
time t has form (Equation 3.9),
|α, t0; t〉 = limn→∞
n∏k=1
e−iH(tk−1)δt |α, t0〉 . (B.17)
35
Special features of triangular 3-spin system Chapter B
Let Uk denote U(kδt). Substituting Equation B.16,
|α, t0; t〉 = limn→∞
n∏k=1
e−iH(tk−1)δt |α, t0〉 (B.18)
= limn→∞
[U †ne
−i(HJ+H′D)δtUnU†n−1e
−i(HJ+H′D)δt · · · e−i(HJ+H′D)δtU0
]|α, t0〉 . (B.19)
For any k in {0, 1, · · ·n},
UkU†k−1 = eiωkδtS
z123e−iω(k−1)δtSz123 = eiωδtS
z123 , (B.20)
Therefore,1
|α, t0; t〉 = limn→∞
[Une
−i(HJ+H′D)δteiωδtSz123 · · · e−i(HJ+H′D)δt
]|α, t0〉 (B.21)
= limn→∞
[Une
−iδt(HJ+H′D−ωSz123) · · · e−iδt(HJ+H′D−ωS
z123)]|α, t0〉 (B.22)
= Une−i
∫(HJ+H′D−ωS
z123)dt |α, t0〉 . (B.23)
Since HJ , H]D, and Sz123 are time-independent operators, while
1U0 = eiω·0·δt = 1, hence is not marked out in Equation B.21.
36
Define time independent effective hamiltonian H = HJ +H ′D − 2ωSz123,
Pτ (t) =∑m
| 〈m; τ | e−iHt |α, ti−1〉 |2. (B.26)
Initial states of the triple-spin plane are listed in Table 2.1. Notably, states |2〉 , |3〉 , |5〉 , |7〉
happen to be the eigenstates of H, therefore Pτ (t) = 1 for them. These states experience
no decay under planar rotation around the C3 symmetry axis (Fig. B.1).
t [sec]
0 5 10
0.0
0.5
1.0
1.5
Pτ
4,62,3,5,71,8
Initial state
Figure B.1: Pτ (t) of initial pseudospin states listed in Table 2.1 that undergoes pla-nar rotation around the internal z-axis in counterclockwise direction. The numericalinputs are ω = 1 rad s−1, JD = 1 Hz, and J = 0.1 Hz. Notably, states |2〉 , |3〉 , |5〉 , |7〉are the four non-decaying states in this case. States {|1〉 , |8〉} and {|4〉 , |6〉} sharesimilar oscillatory behavior within each pair.
37
Bibliography
[1] M. Tegmark, Importance of quantum decoherence in brain processes, Phys. Rev. E61 (Apr, 2000) 4194–4206.
[2] C. Seife, Cold numbers unmake the quantum mind, Science 287 (2000), no. 5454791–791, [https://science.sciencemag.org/content/287/5454/791].
[3] F. W. Wehrli, Temperature-dependent spin-lattice relaxation of 6li in aqueouslithium chloride, Journal of Magnetic Resonance (1969) 23 (1976), no. 3 527 – 532.
[4] P. Hore, Nuclear Magnetic Resonance. Oxford chemistry primers. OxfordUniversity Press, 2015.
[5] H. Hu and M. Wu, Spin-mediated consciousness theory: possible roles of neuralmembrane nuclear spin ensembles and paramagnetic oxygen, Medical Hypotheses63 (2004), no. 4 633 – 646.
[6] M. P. A. Fisher, Quantum cognition: The possibility of processing with nuclearspins in the brain, 2015.
[7] K. Onuma and A. Ito, Cluster growth model for hydroxyapatite, Chemistry ofMaterials 10 (1998), no. 11 3346–3351, [https://doi.org/10.1021/cm980062c].
[8] X. Yin and M. J. Stott, Biological calcium phosphates and posner’s cluster, TheJournal of Chemical Physics 118 (2003), no. 8 3717–3723,[https://doi.org/10.1063/1.1539093].
[9] L.-W. Du, S. Bian, B.-D. Gou, Y. Jiang, J. Huang, Y.-X. Gao, Y.-D. Zhao,W. Wen, T.-L. Zhang, and K. Wang, Structure of clusters and formation ofamorphous calcium phosphate and hydroxyapatite: From the perspective ofcoordination chemistry, Crystal Growth & Design 13 (2013), no. 7 3103–3109,[https://doi.org/10.1021/cg400498j].
[10] L. Wang, S. Li, E. Ruiz-Agudo, C. V. Putnis, and A. Putnis, Posner’s clusterrevisited: direct imaging of nucleation and growth of nanoscale calcium phosphateclusters at the calcite-water interface, CrystEngComm 14 (2012) 6252–6256.
[11] M. W. Swift, C. G. Van de Walle, and M. P. A. Fisher, Posner molecules: fromatomic structure to nuclear spins, Physical Chemistry Chemical Physics 20 (2018),no. 18 12373–12380.
[12] A. S. Posner and F. Betts, Synthetic amorphous calcium phosphate and its relationto bone mineral structure, Accounts of Chemical Research 8 (1975), no. 8 273–281,[https://doi.org/10.1021/ar50092a003].
[13] M. P. A. Fisher and L. Radzihovsky, Quantum indistinguishability in chemicalreactions, Proceedings of the National Academy of Sciences 115 (Apr, 2018)E4551–E4558.
[14] R. van Eldik and I. Bertini, Advances in Inorganic Chemistry: Relaxometry ofWater-Metal Ion Interactions. ISSN. Elsevier Science, 2005.
[15] P. J. Mohr, B. N. Taylor, and D. B. Newell, Codata recommended values of thefundamental physical constants: 2010, Rev. Mod. Phys. 84 (Nov, 2012) 1527–1605.
[16] G. Mancardi, C. E. Hernandez Tamargo, D. Di Tommaso, and N. H. de Leeuw,Detection of posner’s clusters during calcium phosphate nucleation: a moleculardynamics study, J. Mater. Chem. B 5 (2017) 7274–7284.
[17] M. Levitt, Spin Dynamics: Basics of Nuclear Magnetic Resonance. Wiley, 2008.