Post on 28-Apr-2022
Magnetic resonance experiments:
Standard experimental settings and theoretical concepts
Almost all modern approaches to dynamics and control of spins
and qubits are based on magnetic resonance-type settings.
π΅π
ππΏ
Single spin or ensemble of spins in a strong magnetic field π΅π
βQuantizing fieldβ β determines
primary quantization axis(why βprimaryβ? Will see shortly)
Induces Larmor precession with the
frequency π0
π» = ππΏ ππ§ +π»β²
Informally we say that π»β² βͺ ππΏ , i.e. typical energies of π»β² are much
smaller than ππΏ. But it is π»β² that is usually of most interest.
V. Dobrovitski, L. 2
How do we excite (βdriveβ) spin dynamics?
xy
Rotating frame
S
Spin Oscillating field
co-rotating
(resonant)
counter-rotating
(negligible)
Pointer states | Ϋ§β and | Ϋ§β . How do we make spins move?
Oscillating field h along X, frequency ππΏ β performs rotations.
Drives spins out of equilibrium state. How?
βπ₯ π‘ = 2β cosππΏπ‘
π΅π
ππΏ
Key concept: Rotating frame
ππΏ ππΏ ππΏ
V. Dobrovitski, L. 2
Rabi oscillations and rotating frame
π» = ππΏ ππ§ + βπ₯ π‘ ππ₯ = ππΏ ππ§ + 2β ππ₯ cos(ππ‘ + πΌ)
We are interested only in dynamics at frequencies close to ππΏ
Driving at the frequency π close to ππΏ : π β ππΏ βͺ ππΏ , β βͺ ππΏ
Rotating frame transformation: π = exp(βπππ‘ ππ§)Rotation with frequency π (not ππΏ!) around the Z-axis
π π‘ = π β ππ (π‘) , ππ π‘ β ev.op. in the rotating frame
π αΆπ = π» π π‘ β π αΆππ = πβ π»π βπ ππ§ ππ =
= ππΏ β π ππ§ + 2β cos(ππ‘ + πΌ) eπππ‘ππ§ ππ₯ eβπππ‘ππ§ ππ =
= ππΏ β π ππ§ + β ππ₯ cos πΌ + ππ¦ sin πΌ + {nonsec. terms} ππ
Secular Hamiltonian : π»π = ππΏ β π ππ§ + β ππ₯ cos πΌ + ππ¦ sin πΌ
V. Dobrovitski, L. 2
Rabi oscillations and rotating frame
Isidor Rabi [Rah-bee], Nobel prize 1944
Weak β βͺ ππΏ time-dependent driving became strong β~|ππΏ βπ|and time-independent
ΟL
βπ₯(π‘)
ππΏ β π
h
Driven spins rotate around π , ππ§ oscillates: Rabi oscillations.
Labframe:
Rotatingframe:
πlabπ
Rotating frame: we transform π» β π»π , density matrix π β ππ , β¦
But not the observables!
Different from interaction representation, rotating-wave approx., etc.
1) Inconvenient: time dependence of ππ₯,π¦ , mutual dependence,β¦
2) Rotating-frame observables are what is actually detected in standard
resonance experiments (I/Q channels)
See e.g. C. P. Slichter, βPrinciples of Magnetic Resonanceβ
V. Dobrovitski, L. 2
Spin driving: qualitative picture
Ideal:π» = π0
αππ§
| Ϋ§β
| Ϋ§βπ
π0
π π = πΏ(π β π0)Single
sharp line
Noise:π» = π αππ§
| Ϋ§β
| Ϋ§βπ
π0
π π
π
β
Spins resonant
with driving
Broadened line,
determined by
π π½
Strong
driving:
π
β
Weak
driving:
Spins resonant
with driving
V. Dobrovitski, L. 2
Rabi oscillations: strong driving
ππΏ β π
h
Rotating frame:
π π» = π΅ ππ§ + β ππ₯ with π΅ = ππΏ β π
How do we take noise into account ?
Weak non-resonant fields (πβ², π΅β²) : |π0 βπβ²| β« π΅β²
If directed along X- and Y-axes: non-secular, neglect
V. Dobrovitski, L. 2
Therefore, only static noise along Z-axis matters: π΅ = π΅0 + π½
ππΏ = π0 + π½ , e.g. π π½ =1
2ππ2exp β
π½2
2π2
Consider quasi-static noise: inhomogeneous broadening
This is precisely what we studied in the previous lecture
Rabi oscillations: strong driving
Strong off-resonant driving: β, π΅0 β« π
Initial state decays towards equilibrium (pointer) states
Pointer states β already analyzed earlier in detail
Decay of all components has Gaussian form, decay time π2β:
π2β = πβ1 for π΅0 β« β
π2β = πβ1 β β/π΅0 for π΅0 βͺ β
V. Dobrovitski, L. 2
ππΏ β π
h
Rotating frame: π
π» = π΅ ππ§ + β ππ₯ ,
π΅ = ππΏ β π
π π½
Any questions? Hint: π2β = πβ1 β/π΅0 for π΅0 βͺ β. What about π΅0 = 0?
Rabi oscillations, strong driving
For π΅0 = 0 :
First order in π½ is gone, but there is second:
Ξ© = Ξ©0 + π΅0 β Ξ€π½ Ξ©0 + Ξ€β2 2Ξ©0 β Ξ€π½ Ξ©02 +β―
Ξ© = π΅2 + β2 β‘ π½2 + β2 β β + Ξ€π½2 2β
π β βπ‘ + Ξ€π‘ π½2 2β
ΰΆ±π π½ ππ½ eππ‘π½2/(2β) = 1
2ππ2ΰΆ±eπ
π‘π½2
2β eβ
π½2
2π2 ππ½ = 1 β π π2π‘2β
β1/2=
= π(π‘) β eππ(π‘) , with π = 1 + π2π‘
2β
2 β1/4
and π = 1
2tanβ1
π2π‘
2β
Strong resonant driving: π΅0 = 0 , β β« π
Resonant: π = π0 , i.e. exactly at resonance with the line center
Spectroscopic language: driving saturates the line
The integral which determines the form of decay:
V. Dobrovitski, L. 2
Initial decay at π‘ βͺ β/π2 :
quadratic, π π‘ β 1 β π΄π‘2 , with linearly changing phase π(π‘), i.e.
Rabi frequency is renormalized by Ξ€π2 (2β).Looks like regular Gaussian, butβ¦
Long-time decay at π‘ β« β/π2 :
extremely slow power-law decay, π(π‘) β π‘β1/2 with π β π/4
where
π = βπ‘ + π(π‘)
αππ₯ π‘ = αππ₯ (constant in time),
αππ§ π‘ = π π‘ αππ§ cos π + αππ¦ sin π
αππ¦ π‘ = π π‘ αππ¦ cos π β αππ§ sin π
All operators are understood as averaged over the noise, the symbol β¦ π½ omitted
Rabi oscillations, strong resonant driving
Leading order in π/β βͺ 1 : decaying rotation around X-axis
Explicit
results:
V. Dobrovitski, L. 2
Rabi oscillations, strong resonant driving
Leading order in π/β βͺ 1 : quantization axis π along X-axis
Note 1: this is in the rotating frame! In the lab frame the quantization
axis is precessing around the Z-axis, and the pointer states have
explicit time dependence.
Recall the comments from the previous lecture: pointer states can
depend on time.
Note 2: we have quasi-equilibriuim (pointer) states along the
rotating-frame X axis: 12(| Ϋ§β Β± | Ϋ§β )
Same idea is used in quantum optics: the concept of βdressed statesβ.
Atomic states change under strong optical driving, become an equal-
weight superposition of | Ϋ§π and | Ϋ§π .
But the theory is more complex: quantum photons instead of classical
driving field, account of spontaneous decay, etc.
V. Dobrovitski, L. 2
Transition from resonant to non-resonant regime occurs when first
and second orders are similar: π΅0 β Ξ€π½ Ξ©0 ~ Ξ€β2 2Ξ©0 β Ξ€π½ Ξ©02
In this case π΅0 is small, and the transition occurs when π΅0~π
Strong driving changes the oscillation frequency of ππ₯,π¦ :
from π΅0 to Ξ©0 = π΅02 + β2 β π΅0 + β2/(2π΅0)
Similar phenomenon in quantum optics: ac Stark shift of atomic
levels under strong optical driving
The power-law decay π‘β1/2 has no well-defined lifetime(formally, Rabi rotations in this regime have infinitely long lifetime)
Jumping ahead: cw spectroscopy would show two lines, split by βThese lines are narrow: Fourier transform of π‘β1/2 is πβ1/2
(formally, infinitely narrow)
Similar phenomenon in quantum optics: Autler-Townes splitting
Rabi oscillations, strong driving: comments
V. Dobrovitski, L. 2
Rabi oscillations: weak driving
Strong driving, weak noise (β β« π) : driving saturates the line
Weak driving: the regime of continuous-wave (cw) spectroscopy
V. Dobrovitski, L. 2
π
β
Spins resonant
with driving
Strong
driving:
π
β
Weak
driving:
Spins resonant
with driving
Typical initial state: along Z-axis, with ππ₯(0) = ππ¦(0) = 0
ππ§ π‘ = ππ§0 β 1 β 2 ππ₯
2 sin2 Ξ€π 2
π = Ξ©π‘ , Ξ© = β2 + π΅2 , ππ₯ = β/Ξ©
Driving at frequency π : can use previous results
Rabi oscillations: weak driving
ππ§(π‘) = ππ§(0) β 1 β β2 π π½ ππ½ β 2sin2
π‘
2β2+π½2
β2+π½2
We focus on βperturbative longβ times: βπ‘ βͺ 1, but ππ‘ β« 1
Consider general line shape π(π½) with characteristic width π
π π½ is assumed to vary smoothly with π½, on a scale π½~π
In contrast, 2sin2
π‘
2β2+π½2
β2+π½2at large π‘ (but small βπ‘!) has a
sharp peak of small width (β π‘β1) and large height (β π‘) at π½ = 0;
decays fast as |π½| grows.
V. Dobrovitski, L. 2
If you have a dΓ©jΓ vu feeling β you are right.
This is just standard time-dependent perturbation theory,
derivation of the Fermi Golden Rule
If you want more mathematical clarity (optional).
Important: π(π½) is dimensional quantity, with dimensionality [rad/s]β1
Because π π½ ππ½ = 1 , and ππ½ = π½ =rad
s
Thus, π π₯ = π β π(ππ₯) and π π₯ ππ₯ = π π½ ππ½
1) Set βπ‘ = 0 for convenience: this just means that we are working in the leading
order w.r.t. the small parameter βπ‘. We should study πΌ = π π½sin2 π½π‘/2
π½2ππ½.
2) π(π½) has a typical scale π: means that it depends on dimensionless quantity
π₯ = π½/π, and there are no other parameters in this function, large or small.
So we re-write πΌ = π π½sin2 π½π‘/2
π½2ππ½ = π π₯
sin2 π₯βππ‘/2
(π₯βπ)2ππ₯
3) Use large dimensionless parameter π = ππ‘, consider π β β.
πΌ = π π₯sin2 π₯βππ‘/2
(π₯βπ)2ππ₯ =
π‘
πβ π π₯
sin2 πβπ₯/2
πβπ₯2ππ₯ , and Ξ€π‘ π is a parameter
(it is dimensional, so it is neither small nor large β just a quantity)
4) Can show: π π₯ = 2sin2 πβπ₯/2
πβπ₯2=
1βcos ππ₯
ππ₯2β π β πΏ(π₯) when π β β
Consider the integral πΌ at π β β, extend it to complex plane, calculate the residue at π₯ = 0.
Get the answer πΌ = π Ξ€π‘ π β π 0 = ππ‘ β π(0)
Note: condition ππ‘ β« 1 is the key here; βπ‘ βͺ 1 is not crucial, can set βπ‘ = 0
V. Dobrovitski, L. 2
ππ§(π‘) = ππ§(0) β [1 β π β2π‘ π(0)]
Weak driving and cw spectroscopy
So we obtain:
Physical meaning? Initially, the spin is along quantizing field
Start driving it, ππ§ decreases: spin flipped, energy absorbed.
Absorbed power π β Ξ€π ππ§ ππ‘
Measuring absorption while continuously driving spins at
frequency π directly measures the line shape π(π)
π ππ§ππ‘
= βπβ2π(0) or, in the lab frame:π ππ§ππ‘
β π(π)
π
βContinuous-wave (cw) spectroscopy
π
π(π)
This is for π‘ βͺ 1/β . What happens later, when βπ‘~1 ?
V. Dobrovitski, L. 2
Most basic characterization of noise and
relaxation.NMR, ESR, optics, IR,β¦ β idea is the same
V. Dobrovitski, L. 2
Weak driving and cw spectroscopy
Energy absorbed β from where? From the cw driving field
π» π‘ = Τ¦ππ₯ β 2β cos(ππ‘ + πΌ)
Absorbed power, average per oscillation period: π = βπ β Ξ€ππ» ππ‘I.e., there must be non-zero components ππ₯(π‘) and ππ¦(π‘)
More detailed theory is needed, will discuss later (Bloch-Redfield)
Our system : ensemble of non-interacting energy-conserving spins
Initially absorb energy as described, but this cannot go on forever
Equlibrium state is not well defined. Spins keep precessing,
unhindered perpetual rotation: energy flows back and forth.
There must be some process that βresetsβ ππ§(π‘)
It could be slow fluctuations of π½, interaction between spins, coupling to
environment, β¦ The same problem as in standard statistical physics: origins of
stochasticity, thermalization, etc.
Serious problems with the current treatment
V. Dobrovitski, L. 2
Phenomenological Bloch equations
Phenomenological approach: there must be some relaxation process
that continually resets the spins and steers them to equilibrium.
Postulate that such a process is linear and memoryless, so the
systemβs relaxation towards equilibrium (without driving) :
πππ§
ππ‘= β
1
π1(ππ§βππ§
0)πππ₯,π¦
ππ‘= β
1
π2ππ₯,π¦
Relaxation time π1 β rate of longitudinal relaxation
Relaxation of the energy of spins in the quantizing field along Z.
How fast is the energy brought to or taken away from spins.
ππ§0 β equilibrium magnetization along the quantizing field
Relaxation time π2β² β transverse relaxation. Does not involve energy
exchange, purely internal spin relaxation (e.g. spin-spin coupling)
1
π2=
1
2π1+
1
π2β² i.e. π2 β€ 2π1 β to ensure correctness
V. Dobrovitski, L. 2
Phenomenological Bloch equations
Example:
two-level system
(e.g. spin 1/2)
| Ϋ§π
| Ϋ§π
Ξππ Ξπππ
ππ‘
ππππ
=βΞππ ΞππΞππ Ξππ
ππππ
Rate equation:
Gives Bloch equation for ππ§. But may not work for more levels!
Depends on rates, things may get complicated.
But the main idea holds: equilibrium requires equilibration mechanism
| Ϋ§β
| Ϋ§βπ
π π | Ϋ§β
| Ϋ§β
Many lines
of zero
width
Many lines
of width π1β1
Simple spectroscopy (FT or cw) gives π π but often does not
say much about individual lines:
In general, response to driving depends on π, β, π1, line shape,β¦
Gives rise to many useful effects (hole burning, spin echo,β¦)
A few notes on NMR/ESR experimental settings
Large quantizing field π΅0, much larger than any other relevant
energy scale. But still πππ΅π΅0 βͺβͺ ππ, so the initial state is:
π β1
πexp β
πππ΅π΅0 ππ§
ππβ
1
π1 β πππ§ , i.e. πrelevant β ππ§
The signal from πβ² = 1 is zero: this part of density matrix contributes
only to noise.
Irrelevant for the signal, important for analyzing signal-to-noise ratio.
because the identity part of the density matrix is not affected much.
V. Dobrovitski, L. 2
Dynamics in relevant experiments mostly unitary: π 1 πβ = ππβ = 1More generally: dynamics in NMR experiments is mostly unital,
i.e. maps 1 β 1
Also, other ways to initialize spins are used more and more often:
optical (e.g. NV centers), dynamic nuclear polarization (DNP), etc.
Often the first step in qubit characterization
0) Qubit is initialized along the Z-axis
1) Apply resonant Rabi driving for a short time, rotate spins from the
Z-axis to the Y-axis (so-called π/2 pulse).
2) Measure ππ₯(π‘) and ππ¦ π‘ : often called free induction decay (FID) (historical reasons: design of traditional NMR experiments)
or free coherence decay, or Ramsey measurement (in modified version)
3) Analyze π(π‘): find π2β , examine the form of decay, deduce which
noise dephases the qubit and what are its properties.
V. Dobrovitski, L. 2
FT spectroscopy: qubit characterization
Note: We do not have to rotate the spin all the way to Y-axis, it is
enough to just provide non-zero ππ₯0 and/or ππ¦
0 .
The form of decay will be the same, only the overall amplitude of
the signal will scale up or down. Will be important later.
NMR/ESR and the rotating frame: comments
Rotating frame: potential danger from non-seqular terms.
Contribution decreases slowly, as |π β π0|β2. If the spectral
density grows faster β can accumulate and /diverge. Be careful!
Relatively rare in NMR/ESR, but often happens in quantum optics.
NMR experiments: Ξ€π β π0 π0~10β3 β 10β5, measured in
ppm (part per million = 10β6). Secular approximation in rotating
frame works very well.
But for other spins and qubits more care may be required.
V. Dobrovitski, L. 2
One can apply π/2 pulse along X- or Y-axis, just by
choosing the phase of driving. Will be important later.
Can use either cw spectroscopy, measuring π(π), or FT
approach, measuring π(π‘). Usually, the two are related via
Fourier transform. But depends on the system (e.g. unusually long
π1 or π2, so that driving saturates the line)
Beyond rotating-frame secular approximation
Secular approximation is of utter importance: spin motion in a
general time-dependent field is not analytically solvable.
Even computers may be useless: for instance, motion of a spin under driving that is not
ideally periodic (e.g. with another harmonic of incommensurate frequency) can exhibit
deterministic chaos.
What lies beyond secular approximation? Two examples
1. Bloch-Siegert shift
Oscillating field
co-rotating
(resonant)
counter-rotating
(negligible)
πxy
Rotating frameπ
Influence of the counter-
rotating component?
Use the counter-rotating
coordinate frame, neglect
the co-rotating component
In this frame: quantizing field π΅π β large positive offset equal to
the Larmor frequency ππΏ
V. Dobrovitski, L. 2
Compare:
Co-rotating frame: π΅π β negative offset ππΏ β π»π§ = ππΏ βπ ππ§
Counter-rotating frame: π΅π β positive offset ππΏ β π»π§β² = ππΏ + π ππ§
I.e. in the counter-rotating frame: π»πΆπ = ππΏ + π ππ§ + βππ₯
This Hamiltonian corresponds to precession around β Z-axis
(corrections to the axis are nonsecular, neglect them)
with the frequency
Ξ©πΆπ β ππΏ +π + 12 β Ξ€β2 ππΏ +π β ππΏ + π + Ξ€β2 (4π)
Transform back to the co-rotating frame: this rotation corresponds to
a frequency shift βπ = Ξ€β2 (4π)
V. Dobrovitski, L. 2
1. The shift is always positive (similar to the βeffect of inverted pendulumβ)
2. Note the difference with ac Stark shift: Bloch-Siegert shift comes
from the counter-rotating field
2. Sub-harmonic resonances
Main spin resonance: π β ππΏ.
Moreover, spin is a nonlinear system, so there are also resonances at
π β 2ππΏ , 3ππΏ , β¦
But there are also sub-harmonic resonances at π β 1
3ππΏ ,
1
5ππΏ , β¦
I.e. spin is not only nonlinear, but also a parametric system
Beyond rotating-frame secular approximation
Suggested projects:
Project 2: sub-harmonic resonances (based on the textbook by A. Abragam,
βPrinciples of Nuclear Magnetismβ)
Project 1: accurate analysis of the Bloch-Siegert shift(based on the paper of Bloch and Siegert, Phys. Rev. 57, 522 (1940), and on the textbook
by A. Abragam, βPrinciples of Nuclear Magnetismβ)
Brief summary:
1). Rotating frame, secular approximation. Transformation to the
rotating frame is different from the rotating-wave approximation or
interaction representation.
2). Rotating frame, strong driving: Rabi oscillations.
Pointer states along the total effective field, rotating in the lab frame.
Dephasing time depending on the frequency detuning ππΏ βπ.
3). Strong resonant driving: extremely narrow-line pointer states
along the X-axis, power-law decoherence β π‘β1/2
4). Weak driving: cw spectroscopy, measurement of π(π)
5). FT spectroscopy: use of pulsed Rabi driving. Measuring coherence
decay π(π‘), related to π(π) via Fourier transform.
6). Need a theory for relaxation. Bloch equations - phenomenological
7). Physics beyond rotating-frame secular theory
V. Dobrovitski, L. 2
V. Dobrovitski, S. 2
πππ§
ππ‘= β
1
π1(ππ§βππ§
0)πππ₯,π¦
ππ‘= β
1
π2ππ₯,π¦
Bloch equations, weak driving: lineshape
πππ§
ππ‘= βππ¦β β
1
π1(ππ§βππ§
0)
With field: add the Landau-Lifshits term αΆπ = π Γ π»
Let us use this phenomenology in the rotating frame:
driving β along the X-axis, detuning π΅ = (ππΏ β π) along the Z-axis
πππ₯
ππ‘= ππ¦π΅ β
1
π2ππ₯
πππ¦
ππ‘= βππ₯π΅ +ππ§β β
1
π2ππ¦
Qualitative consideration:
driving tilts quantization axis away from Z by small angle (of the order Ξ€β π΅ or βπ1,2)
ππ§ βππ§0 is of the second order in this small tilt (cosine of small angle), i.e. of the
order of ( Ξ€β π΅)2 or (βπ1,2)2. If we look at linear terms in β, we can take ππ§ = ππ§
0.
Our goal: consider very weak driving, β βͺ π΅, π1,2β1and find the equilibrium
values of the transverse components ππ₯ and ππ¦
and assume π2 ~ π1
V. Dobrovitski, S. 2
Bloch equations, weak driving: lineshape
πππ₯
ππ‘= ππ¦π΅ β
1
π2ππ₯
πππ¦
ππ‘= βππ₯π΅ +ππ§
0β β1
π2ππ¦
π+ = ππ₯ + πππ¦ : ππ+
ππ‘= βπ π+π΅ β
1
π2π+ + π ππ§
0β = βππ+ + π ππ§0β
π+ π‘ = πΆπβππ‘ +π ππ§
0β
πSo that in equilibrium, at π‘ β β , π+ = Ξ€π ππ§
0β π
ππ₯ β = β ππ§0π2 β
(ππΏ β π) π2
1 + (ππΏ β π)2 π22
ππ¦ β = β ππ§0π2 β
1
1 + (ππΏ β π)2 π22
π» π‘ = Τ¦ππ₯ β 2β cosππ‘Driving in the lab frame:
π π‘ = Τ¦ππ₯ β ππ₯ cosππ‘ + Τ¦ππ¦ β ππ¦ cosππ‘Magnetization in the lab frame:
Power absorbed per period: π = βπ β Ξ€ππ» ππ‘ β β β ππ¦ β
V. Dobrovitski, S. 2
Bloch equations, weak driving: lineshape
Absorption line π(π) β πβ²β²(π) β ππ¦ β Dispersion line πβ²(π) β ππ₯ β
π
π π
ππΏ
π
ππΏπβ²(π)
(ππΏ β π) π2
1 + (ππΏ β π)2 π22
1
1 + (ππΏ β π)2 π22
Lorentzian
π π =1
1 + (ππΏ β π)2 π22
What is the corresponding π(π)?
Free decay, no driving, initial state ππ₯ 0 β 0, ππ¦ 0 = ππ§ 0 = 0
πππ₯
ππ‘= ππ¦π΅ β
1
π2ππ₯
πππ¦
ππ‘= βππ₯π΅ β
1
π2ππ¦ β
π π‘ = eβπ‘/π2 cos(ππΏ β π)π‘
ππ₯ π‘ = ππ₯ 0 eβπ‘/π2 cos π΅π‘
Note: this is no longer a static noise field along Z, but still π π‘πΉπ
π(π)
Fourier Transform