Single molecule detection of nanomechanical motion

Vadim Puller1
(Dated: May 23, 2012) Correlation function of the photon counts, when the photon emission rate is modulated.
PACS numbers: 73.23.-b, 73.23.Hk, 73.63.Kv

I.

POISSON STATISTICS OF PHOTONS

a. Rate equations for photon emission by a two level system (TLS) We consider a two-level system (TLS) driven by a laser and re-emitting photons into the luminescence eld. This TLS-photon system can be described by the probabilities that the system is in state a/b with n photons added to the eld due to TLS being rst excited from state a to state b with the absorption rate abs and then emitting the photon with the rate em : Pa,n /Pb,n . The corresponding equations for the probabilities are b,n = P a,n = P abs Pa,n em Pb,n em Pb,n1 abs Pa,n , (1a) (1b)

(we start with a state that has no photons, therefore all the probabilities with negative indices are identically zero.) Eqs. (1) can be solved using the Laplace transform, which we dene as + L [x(t); s] = x (s) = dtx(t)est , L [x (t); s] = sx (s) x(0). (2)
0

The Laplace transformed equations for the probabilities are [ ] 1 b,n (s) = a,n (s) P Pb,n (0) + abs P s + em [ ] 1 a,n (s) = b,n1 (s) . P Pa,n (0) + em P s + abs Taking the initial conditions Pb,n (0) = 0 and Pa,n (0) = n,0 we can solve these equations iteratively: b,n (s) = P with the result b,n (s) = P 1 em abs 1 a,n (s) = em P b,n1 (s), P a,0 (s) = Pa,n (s), P , s + em s + abs s + abs em abs (s + em ) (s + abs ) ]n+1 a,n (s) = , P ( [ ]n 1 em abs . s + abs (s + em ) (s + abs )
1 )k 2

(3a) (3b)

(4)

(5)

One can use the fact4 that (k 1)! is the Laplace transform of (s + a)k (s + b)k t ab e 2 (a+b)t Ik 1 2
1

) ab t 2

(6)

and the properties of the Laplace transform to nd the time dependence of Pa,n , Pb,n . b. Reduction to a Poisson process We use a simplifying approximation that the emission rate is very fast: em abs , which leads to b,n 0, P a,n P and corresponds to a usual Poisson process Pn (t) = which is described by equation Pn (t) = (Pn1 Pn ) . (From now on we omit subscript a in Na,n and subscript abs in abs .) (9) (t) t e , n!
n

n abs

(s + abs )

n+1 ,

(7)

(8)

2 c. Average and variance We denote the random number of the photons emitted during time t by N (t). The average and variance can be easily calculated:
+ + + n n+1 n (t) (t) t (t) t N (t) = e = e = et = t, nPn (t) = n n! (n 1)! n! n=0 n=1 n=1 n=0 +

(10)

since the expansion of the exponent is ex =

+ n x . n! n=0

(11)

[N (t)] = in particular

+ n=0

n2 Pn (t) =

+ n=1

n(n 1+1)Pn (t) =

+ n=2

n(n 1)Pn (t)+

+ n=1

nPn (t) = (t) +t = N (t)2 + N (t), (12)

[N (t)] N (t)2 = N (t) = t and the signal-to-noise ratio decreases with time as N (t) SN = = N (t) = t. 2 [N (t)] N (t)2

(13)

(14)

d. Correlation function To calculate the correlation function N (t)N (t ) we need the joint probability that n photons have been emitted during time t, given that m photons were emitted during time t (for certainty we will further always consider t t : N (t)N (t ) =
+ n n=0 m=0

nmP (n, t; m, t ) =

+ n n=0 m=0

nmP (n, t|m, t )Pm (t ),

(15)

where P (n, t|m, t ) is the corresponding conditional probability, given by P (n, t|m, t ) = Pnm (t t ). (16)

This can be easily shown when, for example, deriving the joint probability from the binomial distribution. Intuitively this follows from the independence of the photon emission events. One can also explicitly verify that: Pn (t) = Now we have N (t)N (t ) =
+ n n=0 m=0 + + m=0 k=0 n m=0

P (n, t|m, t )Pm (t ).

(17)

nmPnm (t t )Pm (t ) =

+ + m=0 n=m

nmPnm (t t )Pm (t ) =

(k + m)mPk (t t )Pm (t ) = N (t t )N (t ) + [N (t)] = (18)

tt + t = N (t)N (t ) + N (t ). Thus, N (t)N (t ) N (t)N (t ) = N (t ) = t .

(19)

3 e. Photon ux and photon counts The number of photons emitted by the molecule/TLS per unit time is I (t) = (t). The detector typically counts the number of photons that arrive during the detector delay time, which is N equivalent to the photon ux averaged over this time: IT (t) = 1 T
t t+T

dt1 I (t) =

1 T

t+T

(t) = dt1 N

N (t + T ) N (T ) . T

(20)

We further implicitly keep in mind that time t is changing in increments of T . The average photon ux, i.e. the average number of counts per unit time is given by IT (t) = N (t + T ) N (t) = . T (21)

The correlation function of the photon counts is given by: IT (t)IT (t ) = 1 [N (t + T )N (t + T ) + N (t)N (t ) N (t + T )N (t ) N (t)N (t + T )] . T2 (22)

If t > t we have (given that we discretize time in units of T this means that t > t + T ) IT (t)IT (t ) = 2 , which means that IT (t)IT (t ) IT (t)IT (t ) = 0, i.e. the photon uxes at dierent instants of time are completely uncorrelated. If t = t we have IT (t)IT (t ) = 2 + or IT (t)IT (t ) IT (t)IT (t ) = IT (t) = . T T (26) , T (25) (24) (23)

By taking the limit T 0 we can obtain the correlation function of the instantaneous photon uxes: I (t)I (t ) I (t)I (t ) = (t t ).
II. INHOMOGENEOUS POISSON PROCESS

(27)

Let us now consider the situation when the photon emission rate (i.e. the TLS absorption rate) is time-dependent. This is so-called inhomogeneous Poisson process1 . The equation for the probabilities in this case is n (t) = (t) (Pn1 Pn ) , P and its solution is [ Pn (t) =
t 0

(28)

]n dt1 (t1 ) n!

t
0

dt1 (t1 )

(29)

0 (t) = (t)P0 (t), (This equation can be obtained by integrating the dierential equations one by one, starting with P or by the limiting transition from the binomial-like distribution, which will be discussed below for a more complex case.) It is convenient to introduce notation t n [(t, t )] (t,t ) Pn (t, t ) = e , where (t, t ) = dt1 (t1 ). (30) n! t

4 Using the manipulations similar to those in the previous section we obtain t N (t) = (t, 0) = dt1 (t1 ), IT (t) = N (t)N (t ) N (t)N (t ) = IT (t)IT (t ) IT (t)IT (t ) = For T 0: I (t)I (t ) I (t)I (t ) = (t) (t t ).
III. DOUBLY STOCHASTIC POISSON PROCESS

(31) (32) (33) (34)

(t + T, 0) (t, 0) 1 t+T = dt1 (t1 ), T T t t N (t ) = (t , 0) = dt1 (t1 ), 0 { 0, if t > t + T IT (t) , if t = t T

(35)

We now would like to consider the case when (t) changes randomly. The process is known as doubly stochastic Poisson process2 , as it is described by a stochastic probability function: [ ]n t dt ( t ) t 1 1 0 Pn (t) = e 0 dt1 (t1 ) . (36) n! Ambiguity arises depending on at which point one decides to average over the realizations of (t) are taken: if one averages the expressions obtained in the end of the previous section over the realizations of (t), no new information comes about. In particular, the photon uxes at dierent times remain uncorrelated. This situation will arise, for example, when one has a long time realization of the photon emission process modulated by a deterministic signal. If one cuts this realization into many pieces in order to simulate the ensemble averaging by averaging over the pieces, this will amount to randomizing the initial conditions for the deterministic process. On the other hand, one obtains a dierent result by averaging over the realizations of at an earlier stage, simultaneously with averaging over the number of the emitted photons. Let us denote by the angular brackets the averaging both over the photon number and the realizations of the photon emission rate, whereas ... will designate the averaging over the realizations of (t) only. For convenience we will also decompose (t) = (t) + (t), where (t) is the uctuating part of the photon emission rate, which has zero average. We now have t N (t) = (t, 0) = dt1 (t1 ) , (37a)
0

IT (t) = whereas N (t)N (t ) =

(t + T, 0) (t, 0) 1 = T T

t+T

dt1 (t1 ) .

(37b)

(t, 0)(t , 0) + (t , 0) = t t (t, 0) (t , 0) + dt2 dt1 (t2 ) (t1 ) + (t , 0) = N (t)N (t ) +

0 0 t

0 t

dt2

dt1 (t2 ) (t1 ) + N (t ),

(38)

i.e. the correlation functions take form t t N (t)N (t ) N (t)N (t ) = dt2 dt1 (t2 ) (t1 ) + N (t ), 0 0 { t+T t +T 1 dt2 t dt1 (t2 ) (t1 ) , if t > t + T 2 T t IT (t)IT (t ) IT (t)IT (t ) = t + T t + T 1 dt2 t dt1 (t2 ) (t1 ) + ITT(t) , if t = t T2 t (39a) (39b)

5 In the limit T 0 we also obtain the correlation function for the instantaneous photon uxes I (t) = I (t)I (t ) I (t)I (t ) =

(t) (t) (t ) + (t) (t t )

(40a) (40b)

IV.

FULL DERIVATION FOR THE DOUBLY STOCHASTIC POISSON PROCESS

To justify the order of the averages used in the previous section, we sketch here the derivation of the Poissonian distribution with a stochastic transition rate, (t). Let us split the time interval [0, t] into N short time intervals of length = t/N , each stretching from tn1 = (n 1) to tn = n. The probability that a photon is emitted in such a time interval is p(tn ) = dw1 (, tn ) = (tn ) , (41) where w1 (, t) is the one-time probability density for stochastic process (t). The probability that the photon is not emitted during the same interval is p(tn ) = dw1 (, tn ) (1 ) = 1 (tn ) .

(42)

If, for example, we want to consider the probability that no photons are emitted during the time interval [0, t], we need to use a joint probability density for the values that (t) takes at dierent times, since these values may be correlated: P0 (t) = dN dN 1 ...d1 wN (N , tN ; N 1 , tN 1 , ..., 1 , t1 ) (1 N ) (1 N 1 ) ... (1 1 ) (43) Using the expansion of the exponent: e 1 we can write N P0 (t) = dN dN 1 ...d1 wN (N , tN ; N 1 , tN 1 , ..., 1 , t1 ) e i=1 N

(44)

Taking formally the limit N and using notation (tn ) = n we re-write the last expression in the form of a continuum integral: t t P0 (t) = D [(t)] W [(t)] e 0 dt1 (t1 ) = e 0 dt1 (t1 ) , (45)

where the integral is over all possible realizations of the random process (t) weighted with the probability density W [(t)]. The probability of emitting one photon is obtained by summing the probabilities of the photon emitted at dierent times tn : P1 (t) = D [(t)] W [(t)]
0

dN dN 1 ...d1 wN (N , tN ; N 1 , tN 1 , ..., 1 , t1 )

N i=1

N j =1,j =i

(1 j ) =

dN dN 1 ...d1 wN (N , tN ; N 1 , tN 1 , ..., 1 , t1 )
t

dt1 (t1 )e

t
0

dt2 (t2 )

N i e j =1 j 1 i i=1 t t 0 dt2 (t2 ) . = dt1 (t1 )e 0

(46)

(To the rst order in : /(1 ) .) For the probability of emitting n photons we obtain ]n [ t dt ( t ) t 1 1 0 0 dt2 (t2 ) e . Pn (t) = n!

(47)

6 Assuming gaussian uctuations of (t), characterized by average (t) and variance (t) (t ) ( (t) = (t) (t)), we can evaluate the generating functional t t t t 1 [ (t)] = e 0 dt1 (t1 )(t1 ) = e 0 dt1 (t1 )(t1 ) e 2 0 dt2 0 dt1 (t2 )(t2 )(t1 )(t1 ) (48)

and obtain all the probabilities Pn (t) by taking the variational derivatives in respect to (t) and then setting (t) = 1. For the purpose of evaluating the average N (t) and the correlation function N (t)N (t ) it is however easier to carry out rst the part of the averaging involving dierent number of the photon emission events, and only afterwards perform the averaging over the realizations of (t), as it was done in the previous section.
V. APPLICATION TO THE TLS COUPLED TO A STOCHASTIC OSCILLATOR

Dynamics of a TLS interacting with electromagnetic eld and illuminated by a laser is described by means of the following equations for the TLS density matrix: d ab (t) = i ( 0 ) ab (t) ab (t) + ig [bb (t) aa (t)] , (49a) dt 2 [ ] d (0) bb (t) = 2g [ab (t)] + bb bb (t) , (49b) dt where is the level splitting of the TLS, is the natural width of the luminescence line, 0 is the laser frequency, g is the strength of the coupling between the TLS and the laser. The other two components of the density matrix are given by ba = (ab ) and aa = 1 bb . The equations are written in the rotating wave approximation with the time dependence due to the laser already absorbed in the denition of the non-diagonal matrix element. The rst equation can be formally solved as t t1 t t i(0 )t i(0 )t 2 2 ab (t) = ab ()e +e dt1 ig [bb (t1 ) aa (t1 )] ei(0 )t1 + 2 . (50)

If the diagonal density matrix elements change much slower than , we can approximately evaluate the above expression as (in addition we assume ab () = 0): ab (t) = ig [bb (t) aa (t)] , i( 0 ) + 2 (51)

and substitute it to the equation for bb with the result d (0) bb (t) = bb + abs aa (t) em bb (t), dt where abs = g2 ( 0 ) +
2 2 4

(52)

, em = + abs .
(0)

(53)

Neglecting the thermal population of the excited TLS state, bb , we obtain d bb (t) = abs aa (t) em bb (t), dt d aa (t) = em bb (t) abs aa (t). dt The rates in these equations are identied with those in Eq. (1) in the obvious fashion: aa (t) =
+ n=0

(54a) (54b)

Pa,n (t), bb (t) =

+ n=0

Pb,n (t).

(55)

Let us now comment on the regime of validity of Eqs. (1): in our derivation we have assumed that bb (t) and aa (t) change slowly compared to the rate . This means that abs em - this is the very condition that justies the reduction to the pure Poisson process that we used in section 1. This condition also means that we assume weak coupling to the laser: abs 4g 2 g . 2 (56)

7 f. including coupling to the oscillator In the case when the TLS level spacing is perturbed slower than the rate 108 s1 we can account for this perturbation by simply replacing + x(t), which allows us to write the absorption rate as abs (t) = where abs =
(0)

g2 [ + x(t)] +
2 2 4

abs + abs (t),

(0)

(57)

g2 2 2 + 4

(58a) (58b)

g 2 abs (t) = [ x(t). 2 ]2 2 + 4

g. Correlation function of a periodic process If one wants to track the signal in real time, one has to ensure a sucient signal-to-noise ratio, i.e. one has to wait for the sucient time for every data point. For example, in order to have signal-to-noise ration of 100, one needs to collect 104 photons. With about 106 photons collected per second this means that one records only 100 data points per second, i.e. the frequency resolution of the real-time measurement is 100Hz . If one is interested only in determining the correlation function of the signal, the detection time per every data point can be reduced. Theoretically this means averaging over the ensemble of the realizations of the signal. In this ideal situation one need not worry about signal-to-noise ratio, since one has the innite number of the realizations of the random process and the averaging over this realizations lters out the noise. For a deterministic signal the ensemble average equals to the signal itself, x(t) = x(t) and the correlation function is not needed. Moreover, this correlation function is identically zeros, since there are no uctuations in respect to the average: x(t)x(t ) x(t)x(t ) = 0. For a random signal averaging over ensemble produces real correlation function. In practice, however, one averages records a long realization of the signal, which is then cut into pieces, which compose the ensemble for the numerical averaging. A deterministic signal in this case is randomized and cannot be ltered by simple averaging. However, in this case its correlation function becomes meaningful. Let us consider a periodic signal, e.f. the rf eld. In this case each of the realizations used in the averaging has form x(t) = x0 cos (t + ), where phase is uniformly distributed in the interval [0, ], i.e. it is characterized by the probability density w () = 1/(2 ). The conditional probability densities of the signal for a given value of phase are w(x, t|) = [x x0 cos (t + )] , w(x, t; x , t |) = [x x0 cos (t + )] [x x0 cos (t + )] . The full probability densities are then d 1 1 , w(x, t|) = 2 2 x0 x2 { [ ] 2 2 x x cos (t) + sin (t) x0 (x ) + w(x, t) = [ ]} )2 x x cos (t) sin (t) x2 ( x . 0

(59a) (59b)

(60a)

w(x, t; x , t ) =

d 1 1 w(x, t; x , t |) = 2 2 2 x0 (x )2

(60b)

(x, x take values only in the interval [x0 , x0 ]) The average signal is now zero: x(t) = 0, whereas the correlation function is (it may be easier to carry out rst integration over x, x and only then after ) x(t)x(t ) = x2 0 cos [(t t )] , 2 (61)

which means the periodicity of the signal can be detected by the correlation analysis, even if the signal is faster than the one allowed by the time necessary to achieve high signal-to-noise ratio.

8
VI. REFORMULATION IN TERMS OF DEGREE OF SECOND ORDER COHERENCE

When the radiation is treated classically, as described above, one can dene the degree of second order coherence as3 I (t)I (t + ) g (2) ( ) = . (62) I (t)I (t + ) In our case (...x now denotes averaging over the uctuations of x) I (t) = abs (t)x = abs , I (t)I (t + ) = I (t)I (t + ) + I (t) ( ) + abs (t) abs (t + )x , which leads to g (2) ( ) = 1 + ( ) abs
VII.
(0) (0)

(63a) (63b)

( 0 ) + 2 ( 0 )2 + 4

]2 x(t)x(t + )x . (64)

EVALUATING THE g (2) ( ) FUNCTION USING THE QUANTUM REGRESSION THEOREM A. Bloch equations and their solution

Hamiltonian:

) ( ( ) d a + a d , k a a + H = [ + x(t)] d d + g d ei0 t + dei0 t + k k k k k

k k

(65)

where d is the dipole moment operator, n = d d is the population of the excited state: ( ) ( ) 0 0 1 0 d= , n= . 1 0 0 0 The Bloch equations are d d(t) = i [ + x(t)] d(t) d(t) igei0 t [1 2n(t)] , dt 2 [ ] d n(t) = 2g d(t)ei0 t n(t), dt where we assumed zero equilibrium occupation of the excited state of the TLS. The correlation functions are C12 (t, t ) = d (t )d(t)d(t ), C22 (t, t ) = d (t )n(t)d(t ). Equations for the correlation functions are C12 (t, t ) = i [ + x(t)] C12 (t, t ) C12 (t, t ) igei0 t [n(t ) 2C22 (t, t )] , t 2 [ ] C22 (t, t ) = 2g ei0 t C12 (t, t ) C22 (t, t ). t To order g 2 in coupling to the laser we have [ t 2 n(t) = 2g dt2
t2 t2
t1

(66)

(67a) (67b)

(68)

(69a) (69b)

dt1 e

(tt2 )i( i2 )(t2 t1 )i

]
d x( )

=
2

g2

C22 (t, t ) = 2g n(t )

2

t t

dt2

t2

i i (tt1 )i dt1 e ( 2 )

t
t1

d x( )

(70a)

dt1 e
2

(tt2 )i( i2 )(t2 t1 )i

t2
t1

]
d x( )

=
2

g n(t )
t

dt1 e

i( i2 )(tt1 )i

t
t1

d x( )

(70b)

9 where we introduced detuning = 0 , and omitted angular brackets symbolizing quantum averaging for the TLS. Further we will use angular brackets, ...x to denote averaging over the realizations of x(t). The degree of second order coherence is dened as g (2) (t, t ) = C22 (t, t )x , n(t)x n(t )x (71)

where we assume that the experiment measures C22 (t, t )x and n(t)x which are then divided numerically.
B. No coupling to the oscillator ( = 0)

In this case the excited state population is n0 = and the degree of second order coherence is g (2) (t, t ) = 1 + e(tt ) 2 cos [ (t t )] e 2 (tt ) .

g2 2 , 2 + 2

(72)

(73)

For t t 1 we have g (2) 1, whereas for t = t we have antibunching eect: g (2) (0) = 0. (See Fig. 1, left panel, and Fig. 2.)
C. Adiabatic limit

When the oscillator dynamics is much slower than the rate , one can simplify the expressions as: n(t) = and C22 (t, t )x = g2 [ + x(t)] +
2 2 4

(74)

1 g 2 n( t ) + x(t)

i 2

ei(tt ) 2 (tt )i + x(t )

d x( )

,
x

i 2

(75)

Expanding this result to the second order in produces a rather cumbersome expression, therefore we note here only that at t t 1 we have ( )2 2 C22 (t, t )x = n(t )n(t)x n2 x(t)x(t )x , (76) 2 x 1+ 2 + 4 i.e. we obtain the degree of second order coherence evaluated before from more intuitive considerations (for unequal times): ( )2 2 (2) g (t, t ) = 1 + x(t)x(t )x . (77) 2 2 + 4 The fuller expression is g [
(2)

(t, t ) = 1 +

( )] 2 x(t)x(t ) + x2 e 2 (tt )+(tt ) + 2 1 2 2 2 + 4 [ ]} { tt ( 2 ) 1 2 n1 sin [ (t t )] 2 d x( )x(0) + , cos [ (t t )] x x(t)x(t ) 2 n0 2 2 + 0 4 ( ) 2 x2 x(t)x(t ) ( n1 n0 )2

n1 n0

)2
2

[ x(t)x(t ) + 1 +

n1 n0

)2

] x e(tt )
2 2

(78)

10 where 2 (t t ) = 2

d2

d1 x(2 )x(1 ).

(79)

[ n(t) n0

] 2 3 2 4 1 + x ( . 2 )2 2 + 4
2 2

(80)

n0 =

n1 2 g2 , = 2 . 2 2 2 n + 4 + 0 4

(81)

We show later (Eq. 96) that some of the terms in Eq. 78 are of higher order of smallness and can be omitted.

D.

Time-independent limit

In order to clarify the meaning of the dierent contributions to g (2) -function, let us consider the limit where the molecular frequency shift, x, is time-independent, but randomly distributed according to the Gaussian law with variance x2 . The following expression is exact to order 2( x2 or, if ) we assume that uctuations of the level population are also 2 Gaussian (i.e. n n0 + n1 x, where n0 = g / 2 + 2 /4 and n1 /n0 = 2/( 2 + 2 /4)): [ ( )2 ] { 2 x2 2 n1 (2) 2 2 g (t, t ) = 1 + x 1 + e(tt ) 2e 2 (tt ) 2 (tt ) n0 [ ]} 2 2 n1 cos [ (t t )] 2 x (t t ) sin [ (t t )] (82) n0 Note that at t t this expression does not go to unity, which is an artifact of our assumption that x is constant in time (otherwise the expression would include decaying term x(t)x(t ).)
E. Full time dependence to order 2 x2

It is convenient to introduce notation t t i 1 + B ( t) i i (tt1 )i tt d x( ) 1 ) A(t) = ( dt1 e ( 2 ) , B ( t ) = i d x( )ei( 2 )(t ) . i i 2 We now can write ( ) 2 2 2 n(t)x = g 2 |A(t)| x = n0 |1 + B (t)| x = n0 1 + |B (t)| x , n(t )n(t) = g 4 |A(t )A(t)| x = n2 {1 + 2 [B (t )B (t)x + B (t )B (t)x ]} , { x [ ]} 2 4 2 2 [n(t )] = g 4 |A(t )| x = n2 , x 1 + 2 [B (t )] x + |B (t )| x
2

(83)

(84a) (84b) (84c)

(84d) [ ] where n0 = g 2 / 2 + 2 /4 . The correlation function is now t 2 i 2 2 C22 (t, t )x = g 4 |A(t )| A(t) A(t )ei( 2 )(tt )i t d x( ) n(t )n(t)x + [n(t )] e(tt ) x { [( ) 2 i( i2 (tt ) 2 2 ) n0 2 e 1 + 2|B (t )| x + 2B (t )B (t)x + [B (t )] x + B (t )B (t)x ]} t t (85) ei t d x( ) + (2B (t ) + B (t ) + B (t)) ei t d x( )
x x

11 ( ) 2 2 To order 2 x2 we can take n2 0 1 + |B (t )| x nx out of the brackets: g (2) (t, t ) = C22 (t, t )x n(t )n(t)x [n(t )] (tt ) + e n2 n2 n2 x x { x i [( ) t 2 2 ei( 2 )(tt ) 1 + 2B (t )B (t)x + [B (t )] x + B (t )B (t)x ei t d x( ) + x ]} i tt d x( ) (2B (t ) + B (t ) + B (t)) e (86)
2 x

We further evaluate (the approximate inequality in the expressions below corresponds to the adiabatic limit) ] [ + i 2 x2 1 2 (87a) |B (t)| x = 2 d 2 x( )x(0)ei(+ 2 ) 2 , 0 2 + 4 i 1 + B (t )B (t)x = d 2 x(t t + )x(0)ei(+ 2 ) + 0 i 2 x(t)x(t ) 1 + , (87b) d 2 x(t t )x(0)ei( 2 ) 2 0 2 + 4 + i i ) B (t )B (t)x = ( d 2 [x(t t + )x(0) + x(t t )x(0)] ei( 2 ) i 2 2 0 2 x(t)x(t ) ( ) , 2 i2 + i i 2 x2 = d 2 x( )x(0)ei( 2 ) ( ) , i 2 2 0 i2 (87c) (87d)

[B (t )] x
2

where we have been assuming throughout that the correlation function x(t)x(t ) is a symmetric function of its arguments and depends only on their dierence. Further we have: t 2 t (tt ) i tt d x( ) = e , (t t ) = e d2 d1 x(2 )x(1 ), (88a) 2 t x t { } tt + t i i2 e(tt ) = B (t)ei t d x( ) , d x ( ) x (0) d [x(t t )x(0) x( )x(0)] ei(+ 2 ) (88b) i x + 2 0 0 } { + tt 2 (tt ) t i i e = , B (t )ei t d x( ) d x( )x(0) + d [x(t t + )x(0) x( )x(0)] ei( 2 ) (88c) x i2 0 0 { } + tt i2 e(tt ) i( + i2 i tt ) d x( ) B (t )e = d x( )x(0) + d [x(t t + )x(0) x( )x(0)] e (88d) . x + i2 0 0 The combination entering the g (2) (t, t ) is thus [2B (t ) + B (t ) + B (t)] e
i t
t

d x( )

{ = 2i

i2 e(tt ) + i2 We now obtain n(t )n(t) = n2 x {

2i2 e(tt ) i2

[
0

tt

d x( )x(0)+
i( i2 )

]} +
i

d [x(t t + )x(0) x( )x(0)] e

0 +

d [x(t t + )x(0) + x(t t )x(0) 2x( )x(0)] e+i(+ 2 )

(89)

2 1 + 2 i2

]} d (x(t t + )x(0) + x(t t )x(0)) e

i( i2 )

(90)

12 The coherence function is g

(2)

n(t)n(t ) + [n(t )] e(tt ) (t, t ) = n2 { [ ( )] n(t)n(t ) 2 (tt )+(tt ) 2 cos [ (t t )] + B (t )B (t) (B (t )) 2e n2 n1 tt sin [ (t t )] 22 d x( )x(0) + n0 0 [ 2 sin [ (t t )] B (t )B (t) (B (t )) + ]} + 22 + i2 i( i2 ) d (x(t t + )x(0) x(t t )x(0)) e , i2 0
2

(91)

where i2 ) B (t )B (t) (B (t )) = ( 2 i2
2

d [x(t t + )x(0) + x(t t )x(0) 2x( )x(0)] ei( 2 ) .

i

(92) Comments (i) Eq. (91) explicitly takes zero value at t = t (all the terms vanish, except those containing n(t)n(t ) and 2 [n(t )] which cancel out with each other.) (ii) In case when x is random but constant in time we obtain the correct answer (to order x2 ) (iii) In the adiabatic limit, |x/x | , when we take the correlation functions out of the integral, the term proportional to ( + i/2)/( i/2) vanishes, whereas [ ] 2 x(t)x(t ) x2 2 B (t )B (t) (B (t )) . (93) ( ) 2 i2 It is instructive to consider the adiabatic approximation:
0 +

d y ( )ep

+ (n) y (0) y (0) y (0) + 2 , n +1 p p p n=0

(94)

where the adiabaticity implies that |y (n) /y | |p|n . For a typical average in our calculations t t t B (t )ei t d x( ) i x = d1 d 2 x( )x(1 )ei( 2 )(t 1 ) = t t ei t d x( ) x ] [ [ t ] + t 2 ( ) i i i( i2 )1 d1 e x(t)x(t ) x2 (95) d 2 x( + 1 )x(0) d 2 x( )x(0) + i2 i2 0 t t t 2 Thus, B (t )B (t) (B (t )) is a correction of higher order in the adiabaticity parameter than t d 2 x( )x(0), and therefore can be omitted. We thus have the following answer in the adiabatic limit (which is a simplied version of Eq. 78): g (2) (t, t ) = n(t)n(t ) + [n(t )] e(tt ) n2 } { tt n n ( t ) n ( t ) 1 2e 2 (tt )+(tt ) cos [ (t t )] sin [ (t t )] 22 d x( )x(0) n2 n0 0
2

(96)

13
F. Application to a stochastic oscillator 1. Some re-denitions

For convenience, let us dene integrals + i( i2 2 2 ) I0 = d x( )x(0)e , I = I = 2

0 0 tt 0

d x(t t )x(0)ei( 2 ) ,
i

(97a) (97b)

d x( )x(0)

Now

[ ] [ ] n(t)n(t ) I+ + I n2 2I0 = 1 + 2 , = 1 + 2 , n2 n2 i2 i2 g (2) (t, t ) = ]] [ n(t)n(t ) i (I+ + I 2I0 ) ( ) 2e cos [ (t t )] + + n2 2 i2 ]} [ ( ) 2 + i2 4iI i (I+ + I 2I0 ) ( ) ) (I+ I ) sin [ (t t )] + + ( i2 2 i2 i2

(tt ) +(tt ) 2

(98a)

n(t)n(t ) + n2 e(tt ) n2 {

(98b)

Note that for t t = 0 we have I = I0 , and (0) = 0, so that the coherence function takes zero value, which corresponds to antibunching.

2.

Stochastic oscillator

The correlation function for a stochastic oscillator is given by { } |tt | x2 [ |tt | ] cos[(t t )] + e = x2 e 2 sin[|t t |] , 2 2 where = /2 + , with = 2 0 /4. So that (for clarity we consider only times t t > 0) x(t)x(t ) = I =
2

(99)

I0

d x( )x(0) = {( ) [ ]} (tt ) (t t ) x 2 2 2 2 0 e sin [(t t )] + 1 e cos ((t t )) , 2 2 0 [ ] + i i i2 x2 2 2 = + , i(+ ) i(+ ) 2 +

0 2 2 2 2

tt

(100a) (100b)

I+

i2 x2 (tt ) 2 e = 2 i2 x2 (tt ) 2 e 2

[(

i 2

ei(tt )

( +

i 2

ei(tt )

] , (100c)

I =

+ + ) ) i(2 + i(2 [ ( ) ( ) ( i + + i i2 x2 i( i )(tt ) 2 2 2 e + i (+ ) i (+ ) 2 + 2 2

) + i(+ 2 [( ) i(tt ) + i 2 e

) i(+ 2 ) i(tt ) ] ( i 2 e

i 2 i( ) 2

i 2 i( ) 2

] .(100d)

14
g () with no interaction, =0 3 /=0 /=1 /=2 /=3 3 /=0 /=1 /=2 /=3
(2)

Approximate g () (<<)

(2)

2.5

2.5

2 g(2)() g(2)() 0 1 2 3 4 5 6 7 8 9 10

1.5

1.5

0.5

0.5

10

FIG. 1: Degree of second order coherence, g (2) ( ), as a function of time for various values of detuning, . Left: no coupling to oscillator ( = 0). Right: coupled to an oscillator ( x2 / = .3, oscillator frequency = 2)

FIG. 2: Degree of second order coherence, g (2) ( ), as a function of time and the detuning, , for a TLS not coupled to an oscillator.

Finally, ( ) = )] 2 x2 [ 3 ( ( ) e 1 . 4 0 (101)

The degree of second order coherence is presented in Fig. 1 as a function of time for TLS not coupled to the oscillator (left) and with account of the oscillator-induced corrections (right). All energies are measured in the units of the TLS excited state broadening, . In particular, the coupling energy is taken to be x2 = .3, whereas the oscillator frequency is = 2. The oscillator damping for the right panel of the gure was taken negligibly small (see more about it below). The oscillations in the coherence functions are noticeably perturbed when the detuning is close to the oscillator frequency. For more clarity we illustrate the degree of second coherence in colorplots, showing its dependence on time and the detuning: Fig. 2 shows the case with no coupling to the oscillator, whereas Fig. 3 presents it with account for the coupling to the oscillator, for negligibly small oscillator damping (left) and for a nite damping rate, = .2 (right). The role of the oscillator damping is to suppress the resonant features.

3.

Negligibly small oscillator damping

In case of negligibly small oscillator damping, , one can obtain very simple expressions for the degree of second order coherence. In particular, x( )x(0) = x2 cos( ), ( ) = 2 x2 [1 cos( )] . 2 (102)

15

I = I0 I The level population is now g 2 n(t) = n0 (1 ) + 2

2 x2 sin [(t t )] , [ ] i2 x2 1 1 = + , 2 + i2 i2 [ ] i2 x2 ei(tt ) ei(tt ) = + , 2 + i2 i2 [ ]

2 4

(103a) (103b)

(103c)

1 ( + )2 +

2 4

1 + ( )2 +

, n0 =

g2 2 x2 , = . 2 2 2 + 4

(104)

The correlation function of the excited state occupation number is { [ 1 1 n(t)n(t ) = n2 1 + cos( ) 2 2 ( ) + 4 ( + )2 + where = t t . The degree of second order coherence can be now written as g (2) ( ) = 1 + A cos( ) + (1 + A)e 2e where
2

]}
2 4

= n2 [1 + A cos( )] ,

(105)

+( )

{cos( ) [1 + B + C cos( )] + sin( ) [D D cos( ) + F sin( )]} , [ ]

2 4

(106)

1 A = ( )2 + [ B= 2 ( )2 + [ 3 C= 2 ( )2 + [ 1 D= 4 ( )2 + [ 2 F = ( )2 +

2 4

1 ( + )2 + + ( + )2 + 3 + ( + )2 + 1 ( + )2 + 2 + + ( + )2 +

(107a) ] (107b) ] (107c) ] (107d) ] (107e) (107f)

2 4

2 4

2 4

2 4

2 4

2 4

2 4

2 4

The appearance of resonant features at = is now obvious. One can roughly estimate the time when the resonance becomes pronounced by equating the decaying oscillatory term with A cos( ) at = . Keeping only the largest contribution of in the oscillatory term (since at this point e 2 ) we have ( ) 2 A A cos( ) = 2e 2 cos( ) = log . (108) 2

4.

Negligibly small oscillator damping: arbitrary coupling strength

In the case of very small oscillator damping one can carry out averaging without expanding the exponentials and then use the standard expansion in terms of modied Bessel functions e
z cos

+ n=

In (z )ein ,

(109)

16

FIG. 3: Second order degree of coherence, g (2) ( ), as a function of time and the detuning, , for small coupling between the TLS and the oscillator, 1. Left: negligibly small oscillator damping, ; Right: nite oscillator damping, = .2.

which allows to obtain n(t) = g 2

+ n=

e In ( ) ( + n) +
2 2 4

(110)

C22 (t, t ) = g 4

e2 (1)r+s In ( )Im ( )Ip ( )Iq ( )Ir ( )Is ( )

[ ][ ] i i ei(m+pr)(tt ) ei(+ 2 )(tt ) ei(m+q+s)(tt ) ei( 2 )(tt ) [ ][ ][ ][ + (m + p r) + i2 + (m + q + s) i2 + (n + q + r) + i2 + (n + p s)

n,m= p,q = r,s=

i 2

].

(111)

The series of the modied Bessel functions can be now summed numerically to required order, which allows to address the strong coupling regime. The left panel of Fig. 4 presents the results obtained with the same parameters as previously, but with coupling given by x2 = . One can see the development of a second resonance feature at . It is also instructive to write down the average C22 (t, t ) n(t ) = g2
+ n=

e In ( ) ( + n) +
2 2 4

{ } (tt ) 1 + e(tt ) 2e 2 cos [( + n) (t t )] .

(112)

While this equation is of not much practical importance, it illustrates one of the features of the degree of second order coherence at strong coupling: superimposing of degree of second order coherence functions, eectively corresponding (tt ) to two-level systems at detunings + n: 1 + e(tt ) 2e 2 cos [( + n) (t t )]
5. TLS driven by a periodic (rf ) force

Let us also consider the case when the TLS is driven not by a stochastic oscillator, but by a periodic rf signal, x(t) cos(t + ). We will assume that the initial phase of the signal, , is randomly distributed in the interval [0, 2 ] due to uctuations of the signal source or due to the statistical averaging used in the measuring procedure. The averaging over this phase is further denoted by an overline. The level population is now n(t) = g 2
+ n=

[Jn ( )]
2

2 2 4

( + n) +

(113)

17

(2) FIG. 4: Second order degree of coherence, g ( ), as a function of time and the detuning, at high coupling strength x2 = . Left: TLS coupled to a stochastic oscillator with negligibly small damping, ; Right: TLS driven by a periodic force.

The occupation correlation function is n(t)n(t ) = g 4

+ s=

eis(tt )

+ m=

Jm ( )Jm+s ( ) )( + (m + s) + m + i2

) i
2

=
2

g whereas C22 (t, t ) = g 4

+ s=

+ s=

Jp ( )Jp+s ( )eip(tt ) ( )( + (p + s) + i2 + p p=

i 2

(114)

] [ (tt ) ip(tt ) + J ( )J ei(tt ) 2 p p+s ( ) e ( )( ) + (p + s) + i2 + p i2 p=

(115)

In the limit of small TLS-rf coupling, 1 we obtain the result identical with that for the case of oscillator with the correspondence = 2x2 (keeping in mind that in Eq. (106) e( ) 1 + ( )). The right panel of Fig. 4 shows the resulting degree of second order coherence obtained the the value of corresponding to x2 = .

3 4

B. Reien and H. Sherman, An optimum demodulator for Poissonian process: Photon source detectors, Proc. IEEE, vol. 51, Oct. 1963, pp. 1316-1320. D. L. Snyder, Filtering and detection for doubly stochastic Poisson process, IEEE Trans. Inform. Theory, vol. IT-18, Jan. 1972, pp. 91-102. Loudons book Abramovitz and Stegun