Effects of a non-linear coupling term in both classical
Transcription
Effects of a non-linear coupling term in both classical
Effects of a non-linear coupling term in both classical and quantum discrete non-linear Schrödinger equation ——— Etude des conséquences de l’ajout d’un terme de couplage non linéaire dans les équations de Schrödinger discrète non linéaire, classiques et quantiques Lionel Jimenez Supervised by J.C. Eilbeck Heriot Watt University April-July 2006 Contents 1 Introduction 3 2 Classical approach 2.1 Classical dimer . . . . . . . . . . . . 2.1.1 Stationary solutions . . . . . 2.1.2 Stability . . . . . . . . . . . . 2.1.3 Bifurcation diagram . . . . . 2.1.4 Non stationary solutions . . . 2.2 Trimer . . . . . . . . . . . . . . . . . 2.2.1 A new parametrization . . . . 2.2.2 Interpretation . . . . . . . . . 2.3 Conclusion on the classical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Quantum approach 3.1 Operators and the Hamiltonian . . . . . . . . 3.2 Quantum dimer . . . . . . . . . . . . . . . . . 3.2.1 The Hamiltonian . . . . . . . . . . . . 3.2.2 Energies of the |0, ni and |n, 0i states 3.2.3 Interpretation . . . . . . . . . . . . . . 3.3 Quantum lattice . . . . . . . . . . . . . . . . 3.3.1 The Hamiltonian . . . . . . . . . . . . 3.3.2 Behaviour when 1 = 0 . . . . . . . . . 3.3.3 Resonance behaviour . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 5 6 7 8 9 10 11 11 . . . . . . . . . 13 13 14 14 15 16 17 17 18 18 21 Bibliography 22 A Details 24 A.1 Eigenvalues of the stability matrix . . . . . . . . . . . . . . . 24 A.2 Calculation details of the trimer parametrization . . . . . . . 24 1 B Figures 25 B.1 Behavour near the bifurcation point . . . . . . . . . . . . . . 25 B.2 Trimer bifurcation diagram when 1 = 0 . . . . . . . . . . . . 26 B.3 Energies when 1 = 0 . . . . . . . . . . . . . . . . . . . . . . . 27 2 Chapter 1 Introduction The discrete nonlinear Schrödinger equation(DNLS) has been studied a lot over the last 20 years. The main reason is that it provides a good explanation for phenomenon such as localisation of energy(Self-Trapping) in certain molecules, that is why it is sometimes called Discrete Self Trapping(DST) equation. For example the study of the C − H stretching in the dihalomethanes such as CCl2 H2 or CBr2 H2 can be described using a discrete nonlinear Schrödinger equation(DNLS) with a linear coupling. A non linear coupling could refine the model and reveal new behaviours for particular values of the parameters that can be obtain arranging or creating molecules in different configurations. But this is a very general equation that can be also found in optics for example, so the parameter we will use will be dimensionless, and if taken a dimension that will be cm−1 ,as it would be for the C − H stretching study. The purpose will be to study the effects of this new term theoretically. We will try to get analytical results and to compare it to the usual results obtained using a linear coupling. So we will first try to add such a term in a classical model, as classical approach usually provides interesting general results before considering a quantum case(QDNLS) that is supposed to better describe the problems as we are working at a quantum scale. For both approach we will begin with only a few degrees of freedom f , that is consistent with the fact that for example the study of the C − H stretching in the dihalomethanes involves only 2 degrees of freedom. After that, when possible, we will try to generalise the results to more degrees of freedom. 3 Chapter 2 Classical approach The theoretical study of the DST equation finds applications in many cases in the quantum world(e.g. crystals with special properties), it seems therefore natural to study the quantum equivalent of the DST equation. But as usual in physics we will start with the classical approach that provides interesting general results and we will later discuss the quantum case in section 3. Although physical system such as nano-crystals include more than 2 or 3 degrees of freedom, you can model some molecules[5] using simple 2 or 3 degrees of freedom models known as dimers and trimers. The advantage is clearly that f degrees of freedom models are not possible to be solved analytically whereas the dimer with its two conserved quantities(energy and number) is fully integrable. Previous studies were mainly about the nonlinear dimer with on-site nonlinearity [6], [7] i.e. with a classical Hamiltonian of the form: H= f X i=1 (ω0 |Ai |2 − X γ A∗i Aj |Ai |4 ) − 0 2 i,j i6=j The object of this work is to study the effect of adding a nonlinear coupling or nonlocal nonlinearity as in [4]. The new classical Hamiltonian will now be: f X X 1 γ 0 A∗i Aj − |Ai Aj |2 (2.1) (ω0 |Ai |2 − |Ai |4 ) − H= 2 2 i=1 2.1 i,j i6=j Classical dimer We start with the simple case of f = 2 degrees of freedom. Using the usual dynamic equations dAj ∂H = i dt ∂A∗j 4 The Hamiltonian 2.1 leads to the dynamic equations: dA1 = ω0 A1 − γ|A1 |2 A1 − 0 A2 − 1 |A2 |2 A1 dt dA2 i = ω0 A2 − γ|A2 |2 A2 − 0 A1 − 1 |A1 |2 A2 dt i (2.2) Defining the norm or number N by N= f X i=1 |Ai |2 it is straightforward to check that Ṅ = 0 and Ḣ = 0, thereby giving the two conserved quantities needed for this new dimer to remain integrable. 2.1.1 Stationary solutions Let us first discuss stationary solutions of the form: A1 = φ1 e−iωt and A2 = φ2 e−iωt Using 2.2 and the definition of N leads to this set of 3 coupled equations with 3 unknown quantities(φ1 , φ2 and ω) φ2 + 1 |φ2 |2 φ1 φ1 ω0 − ω = γ|φ2 |2 + 0 + 1 |φ1 |2 φ2 N = |φ1 |2 + |φ2 |2 ω0 − ω = γ|φ1 |2 + 0 (2.3) Subtracting the first two equations and assuming that φ1 and φ2 are real, one finds: 0 (γ − 1 ) − (φ22 − φ21 ) = 0 (2.4) φ1 φ2 Combining this with the definition of N obviously gives 3 different solutions: The symmetric solution(↑↑) r r N N and φ2 = φ1 = 2 2 with ω↑↑ = ω0 − N (γ + 1 ) − 0 2 The antisymmetric solution(↑↓) r r N N N φ1 = and φ2 = − with ω↑↓ = ω0 − (γ + 1 ) + 0 2 2 2 5 The so-called soliton solution(↑ .) φ1 = " N 2 φ2 = " N 2 s !# 12 420 1+ 1− 2 N (γ − 1 )2 s !# 12 420 1− 1− 2 N (γ − 1 )2 ω↑. = ω0 − γN Interpretations The first two solutions are the usual results for coupled oscillators. However the so-called soliton solution seems really different. First of all it exists only 0 for γ > 2 N +1 , that is for a on-site nonlinearity strong enough. Then, if the nonlinearity increases until γ − 1 0 then φ1 → N and φ2 → 0 and all the energy is located on just one oscillator(this explains the designation ’↑ .’, the name soliton coming from the fact that is a typical localised nonlinear effect.) All these solutions are very similar to the case when there is only a linear coupling and we can point out that replacing γ − 1 by a new γ 0 in the expression of φ1 and φ2 leads to the exact same results. This does not work with the frequencies though, but we will show after 2.1.4 that this γ0 remains important. 2.1.2 Stability To study the stability of such a problem we have now to introduce small variations in the amplitudes as: φ01 = φ1 + δu1 φ02 = φ2 + δu2 δu1,2 are both complex and depend on time and φ1,2 being solution of 2.3. The frequency is unchanged. This leads(keeping only first order terms) to an evolution equation for the variations: ˙ <(δu) 0 B <(δu) = =(δu) A 0 =(δu) δu being a vector and A and B being 2x2 matrices defined below: δu1 W = ω0 − ω δu = δu2 A= −W + γφ21 + 1 φ22 0 0 −W + γφ22 + 1 φ21 6 B= −0 − 21 φ1 φ2 W − 3γφ21 − 1 φ22 −0 − 21 φ1 φ2 W − 3γφ22 − 1 φ21 For the solutions to be stables we want the eigenvalues of the matrix AB to be purely real and negatives([1], [7]). The calculation(A.1) of the eigenvalues gives the following results: 0 The ↑↓ solution is stable for γ > 1 − 2 N 20 The ↑↑ solution is stable for 0 < γ < N + 1 0 The ↑ . solution is stable for γ > 2 N + 1 These results are close to the linear coupling solution, and again replacing γ by γ 0 leads to the same result 2.1.3 Bifurcation diagram Figure 2.1 shows the bifurcation diagram(i.e. γ vs W = ω0 − ω) of the modified dimer in red and the usual dimer in blue(Unstable solution are in doted line). We have the same global behaviour except that there’s a shift in the epsilon1=0 400 Gamma 300 200 100 0 0 100 0 100 200 300 W epsilon1=2*epsilon0 400 400 Gamma 300 200 100 0 200 W 300 400 The highest branch is the ↑↓ branch. The one that becomes dashed is the ↑↑ branch. The one with the different steep is the ↑ . branch. Figure 2.1: Bifurcation diagram for different values of 1 (0 = 30cm−1 ) bifurcation point. This shift is due to the non local nonlinearity that tends 7 to reduce the effects of the local nonlinearity by a −1 term, so replacing γ by γ 0 we have the same coordinates for the bifurcation point. 2.1.4 Non stationary solutions As the dimer is integrable, it is interesting to search for non-stationary solutions. It will inform us about the behaviour of the 2 oscillators and will highlight the bifurcation showing how we move from one solution to another. For the usual dimer this has been done in [6], we basically follow the same method using formula on elliptic integrals([8]) Major steps of the calculation Using 2.2 and introducing p = |A2 |2 − |A1 |2 one can find: d2 p = 20 p [(γ − 1 )(A∗2 A1 + A∗1 A2 ) − 20 ] dt2 (2.5) We have now to find out an expression for A∗2 A1 + A∗1 A2 . Equations 2.2 and their conjugates give us a straightforward to solve differential equation that leads to A∗2 A1 + A∗1 A2 = γ − 1 2 (pt=0 − p2 ) + (A∗2 A1 + A∗1 A2 )t=0 40 (2.6) Then combining 2.5 and 2.6 one finds a Duffing’s equation d2 p dt2 = Ap − Bp3 A = 20 (γ − 1 )(A∗2 A1 + A∗1 A2 )t=0 + 12 (γ − 1 )2 p2t=0 − 20 B = 21 (γ − 1 )2 For simple initial conditions(e.g. A2 (t = 0) = N ) this can be solved in terms of elliptic functions depending on the value of the different parameters: 0 If 2A < N 2 B, that is γ < 2 N + 1 γ − 1 p = N cn −0 t, k = 20 0 If 2A > N 2 B, that is γ > 2 N + 1 20 γ − 1 N t, k = p = N dn − 2 γ − 1 dn and cn being the Jacobi’s elliptic functions of parameter k. The difference between the two behaviours is obvious on figure 2.2, while the green plot is always strictly positive, the red one oscillates between 1 and -1. This means that in the first case the energy is localised mainly on the first oscillator, whereas in the second case the energy oscillates between the two oscillators. This confirms the conclusion of the bifurcation diagram. If we check what is 8 Figure 2.2: behaviour of the dimmer for γ < 1 (green) 20 N + 1 (red) and γ > 20 N + happening at the bifurcation(B.1), we see that the green plot tends to have a longer period and a higher amplitude of oscillation and stay mainly near 0 and that the red plot is radically changing to be mainly near 0 except for some peaks. One must also notice that when γ = 1 i.e. γ 0 = 0 the behaviour is strictly sinusoidal, just as if there was no non-linearity at all. 2.2 Trimer We propose a first step into many degrees of freedom by treating the case of f = 3 which is called the trimer. We will only study the stationary solutions, as the time dependent solutions seem much more complicated to find. We now have a new set of equations for the amplitudes and the frequency: φ2 φ3 ω0 − ω = γ|φ1 |2 + 0 (2.7) + 1 |φ2 |2 |φ3 |2 + φ1 φ1 φ3 φ1 2 ω0 − ω = γ|φ2 | + 0 + 1 |φ1 |2 |φ3 |2 + φ2 φ2 φ1 φ2 + + 1 |φ1 |2 |φ2 |2 ω0 − ω = γ|φ3 |2 + 0 φ3 φ3 N = |φ1 |2 + |φ2 |2 + |φ3 |2 There are 2 trivial cases to start with. First if one of the amplitude is 0, we are back to the already solved dimer case, then if the amplitudes are all equal, obviously we have φ1 = φ2 = φ3 = √N3 and W = ω0 − ω = (γ − 21 ) N3 + 20 . Just as in the usual trimer we have a single complex solution([1]). For the 9 other solutions a parametric solution has been found in the linear coupling case([1]). As this can not be used(or adapted easily) for the modified trimer, we propose a new parametrization that can be use with 1 6= 0 as well as with 1 = 0(usual case). 2.2.1 A new parametrization We first assume that 2 amplitudes are equal, that is φ1 = a, φ2 = φ3 = b. After some algebra(A.2) starting from 2.7 we find that b− a 2 a2 W + γN − 0 = − 2 4 γ − 1 (2.8) Therefore giving us a parametrization in terms of hyperbolic sine and cosine instead of the sine and cosine obtained in the usual case([1]). x will be −0 our new parameter, and let W +γN be written as A. We can’t make any γ−1 assumption on the sign of A so we will discuss both signs separately. A>0 Using the fact that ch(x)2 − sh(x)2 = 1 we have: √ √ a = 2 Ash(x) and b = Aex (2.9) And we have to keep in mind that A depends on the parameter as well. To be able to use this parametrization we have to find which range of values correspond to which behaviour and to plot the bifurcation diagram including all the branches, that means being able to find W (x) and γ(x). From 2.7 first and last equation we find two non linear coupled equations: γN − 0 − W ex γN − 0 − W 2x sh(x)2 + 0 + 21 e γ − 1 sh(x) γ − 1 1 1 N γN 1 − = 0 + W − 4sh(x)2 + 2e2x 4sh(x)2 + 2e2x W = 4γ (2.10) (2.11) given the fact that combining 2.10 and 2.11 leads to a second order equation on γ we are able to find(using Maple if necessary) two different expressions for γ(x) and then W (x), expressions we can substitute in 2.9 to have a(x) and b(x). We will not give the expressions here as they might not fit in a page and they are easy to find using Maple. A<0 We proceed exactly the same way but the results are modified. Let us just give the equations, A0 = −A: √ √ a = 2 A0 ch(x) and b = A0 ex 10 W = 4γ 2.2.2 ex −γN + 0 + W 2x −γN + 0 + W ch(x)2 + 0 + 21 e (2.12) γ − 1 ch(x) γ − 1 1 1 N γN 1 + (2.13) = 0 + W + 2 2x 4ch(x) + 2e 4ch(x)2 + 2e2x Interpretation After verifying that putting 1 = 0 in the previous equations gives the same behaviours and branches as in [1], we are now able to plot everything we need to understand the behaviour of the modified trimer. Figure 2.3 shows the amplitude of a(red) and b(blue) while the parameter x scales from -4 to 4. First we didn’t use the full range(theoretically ] − ∞; ∞[ of the parameter as it appears that we can see some asymptotic behaviour already in this range. We can guess that the range between -4 and -1 on figure 2.3 correspond to the (↑ ..) as |a| |b|, a closer analysis of the data shows that on (a) this corresponds to a branch with negative values of γ, whereas (b) correspond to the desired (↑ ..) branch. Between -1 and 0 in (a) we have opposite sign for a and b but rather the same amplitude, this corresponds to the (↑↑↓) branch. Between 0 and 1 in (a) we have both a and b positive but this time b a, so that this is the (↑↑ .) branch. In both figure, we can see for x > 1 that we have a = b which should represent the (↑↑↑) branch, but again a closer analysis shows that it corresponds only to one or 2 points on the bifurcation diagram, which is good because we assumed that a 6= b, excluding this branch in the parametrization. If we now plot the bifurcation diagram(Figure 2.4) we are able to identify the branches, and tell that there is no major difference with the usual case(B.2), except again for a shift. 2.3 Conclusion on the classical approach After the study of the f = 2 and 3 cases, we are now able to draw some conclusion about the general behaviour and the influence of the parameter 1 . We have shown that there are solutions with localised energy in both cases and that the influence of 1 is to weaken the local nonlinearity. This new term does not sharply modify the behaviour which makes it simple to solve analytically, but does not show some interesting new behaviour. We must then generalise the results we have about the f = n case that is mainly the existence of a localised solution of the type (↑ .....etc) even though the 1 parameter tends to weaken the local nonlinearity. 11 (a) 1 0.5 0 −0.5 −1 −4 −2 0 x (b) 2 4 −2 0 x 2 4 1 0.5 0 −4 Figure 2.3: Amplitudes of a(red) and b(blue) function of the parameter x in the case A > 0(a) and A < 0(b) 600 540 480 gamma 420 360 300 240 180 120 60 0 0 60 120 180 240 300 W Figure 2.4: Trimer branches(no stability has been considered) 1 = 0 = 30cm−1 The yellow curve is the ↑↑↑ branch The highest red curve is the ↑↑↓ branch The other red one is the ↑↑ . branch The green red is the ↑ .. branch 12 Chapter 3 Quantum approach The classical approach gave us some interesting results about this new nonlinear coupling term, but the behaviours we predicted are very similar to the usual linear coupling case. The study of the usual quantum case leads to the quantum breathers phenomenon([5],[9]), that is a localised state in terms of number of excitation that latter progress through the lattice while delocalising. A recent study([4]) has tried to look at non linear coupled Hamiltonian and finds out a new resonance-like behaviour. This behaviour is also theoretically studied in the case of low number of quanta and big lattice, but in a certain approximation. The object of this chapter will be to get some exact result on the quantum dimer and compare the predictions to the classical ones, and after that to look at the big lattices case but without the approximation made in [4]. 3.1 Operators and the Hamiltonian We have to quantise the problem, we will proceed as in [5]. That is: A∗i ↔ b†i Ai ↔ bi bi and b†i respectively being the usual lowering and raising operators defined using Dirac’s notation as: √ bj |n1 , n2 , ..., nj , ..., nf i = nj |n1 , n2 , ..., nj − 1, ..., nf i p † nj + 1|n1 , n2 , ..., nj + 1, ..., nf i bj |n1 , n2 , ..., nj , ..., nf i = We now quantise N N̂ = f X N̂j N̂j = b†j bj j=1 nj being the eigenvalues of N̂j and describing the number level on site(oscillator) P j, and n = fj=1 nj being the total excitation number. And finally we quan13 tise H according to [5] Ĥ = − γ − ω0 N̂j + b†j bj b†j bj + 0 (b†j bj+1 + b†j bj−1 ) + 1 b†j bj b†j+1 bj+1 2 2 f X γ j=1 (3.1) 3.2 Quantum dimer 3.2.1 The Hamiltonian The dimer’s Hamiltonian is now: i γ γh † 2 Ĥ = ω0 − N̂ − (b1 b1 ) + (b†2 b2 )2 − 0 (b†1 b2 + b†2 b1 ) − 1 b†1 b1 b†2 b2 2 2 As N̂ and Ĥ obviously commute, an eigenfunction of N̂ is an eigenfunction of Ĥ. If we take a fixed value of number of excitation n, the most general eigenfunction of N̂ will be |ψi = c0 |n, 0i + c1 |n − 1, 1i + ... + cn |0, ni where the cj are complex constants we will determine using the fact that |ψi has to be an eigenfunction of Ĥ as well. This is H~c = E~c where ~c is the column vector(c0 , c1 , ..., cn ) and H = H0 − 0 V is the matrix associated to Ĥ in the |n − j, ji base. Let us now calculate the elements of the matrices H0 and V . It is well known([2],[5]) that: γ Ĥ0 |n − j, ji = ω0 n − (n + n2 − 2nj + 2j 2 ) − 1 b†1 b1 b†2 b2 |n − j, ji 2 γ 2 (3.2) = ω0 n − (n + n) + (γ − 1 )(n − j)j |n − j, ji 2 Thereby showing that H0 is diagonal of values given by 3.2. We can also find the matrix V : p b†1 b2 |n − j, ji = (n − j + 1)j|n − j + 1, j − 1i p b†2 b1 |n − j, ji = (n − j)(j + 1)|n − j − 1, j + 1i Which gives a tridiagonal matrix: √ n 0 ... 0 0 p 0 √n 2(n − 1) p 0 ... 0 p 0 2(n − 1) 0 3(n − 2) ... 0 0 ... . . . . . . . . . . .. V = ... p p 0 ... 3(n − 2) p 0 2(n − 1) 0 √ 0 2(n − 1) 0 n ... 0 √ 0 ... 0 0 n 0 14 To find the quantum analogy of the classical soliton solutions we are interested in finding the energies of states involving |0, ni and |n, 0i. 3.2.2 Energies of the |0, ni and |n, 0i states Since the (1, n) and (n, 1) positions in the matrix V are 0, one could think these two states are uncoupled, but in fact they are through the coupling between each adjacent state. To find out how, we will use the perturbation theory just as in [2].Using this theory supposes that γ − 1 0 . We have some data(3.3) about γ and 0 , but 1 has still to be measured, so we will 0 ). As we can see on 3.3 we have assume that it is small(typically 1 ∼ 10 γ − 1 ∼ 40 , that we will taken as enough to make the approximation. It is important to see that at orders lower than n both states have the same energy, so the calculation will require a nth order perturbation theory. Molecule 0 γ CCl2 H2 29.54 127.44 CBr2 H2 32.80 125.45 CI2 H2 33.69 124.25 V alues1 of γ and 0 (in cm−1 , taken in [5]) (3.3) We will use the following notations for the approximated energies and corresponding eigenvectors: E = E0 + 0 E1 + . . . + k0 Ek + . . . ~c = c~0 + 0 c~1 + . . . + k0 c~k + . . . Now that we know we are talking about vectors for c1 . . . ck we will omit the symbol ~. Defining the operator L = H0 − E0 I and using the perturbation theory we obtain the set of equations: Lc0 = 0 Lc1 = (E1 − V )c0 Lc2 = (E1 − V )c1 + E2 c0 .. . (3.4) Lck = (E1 − V )ck−1 + E2 ck−2 + . . . + Ek c0 As we are interested in the |0, ni and |n, 0i states, 3.2 gives us E0 = ω0 n − γ 2 0 2 (n + n). And one can remark that the operator L now depends on γ = γ − 1 : L = diag(α0 , α1 , . . . , αn ) αi = (γ − 1 ) 1 multiply it by hc e to have it in Joules 15 So formally, it is just like the usual case, but with γ 0 instead of γ. We will not develop the entire detailed calculation as it would take pages and it is all done in [2]. The important point is the result: n0 2nno (3.5) +o ∆E = (n − 1)!(γ − 1 )n−1 (γ − 1 )n−1 3.2.3 Interpretation It as been show([5],[2]) that this energy splitting correspond to eigefunctions of the form: 1 |ψi± = √ (|n, 0i ± |0, ni) 2 Comparing the results with the classical one, something seems surprising at first glance: nothing tells us about a localisation of the energy on one of the two oscillators as the probability of finding energy on one or another oscillator is equal(because of the symmetry). But there are experimental evidence that a localisation of energy can happen, even in the quantum case. Let us explain it by taking a solution that would be a combination of |ψi± 1 1 + − |ψ(t)i = √ |ψi+ e−iE t + √ |ψi− e−iE t 2 2 Then at t = 0 all the energy is located on the first oscillator, and most will remain there until the phase difference changes to π that is in a time τ : τ∼ π~ ∆E where ∆E = |E + − E − |. Giving us an estimation of the oscillation time: τ∼ π~(n − 1)!(γ − 1 )n−1 2nn0 This time grows rapidly2 with n , showing that the energy remains located in one oscillator in agreement with the classical case(2.1.1). Once again the term −1 tends to weaken the local non linearity and could even give a null time for γ = 1 , radically changing the behaviour of the oscillators, the energy not being localised anymore. Just like in the classical approach, we have a behaviour that is closer to a fully linear Hamiltonian for this particular value of 1 . 2 Numerically, this time becomes rapidly longer than any measurable time 16 3.3 Quantum lattice 3.3.1 The Hamiltonian On lattices we generally use a slightly different Hamiltonian coming from a quantisation done after a change of variables. Ĥ = − f X γ j=1 2 b†j b†j bj bj + 0 (b†j bj+1 + b†j bj−1 ) + 1 b†j bj b†j+1 bj+1 (3.6) We can notice that Ĥ is translationally invariant, so k is a good quantum number and defining the translation operator T̂ by: T̂ |n1 , n2 , . . . , nk−1 , nk , nk+1 , . . . , nf i = |nf , n1 , . . . , nk−2 , nk−1 , nk , . . . , nf −1 i Ĥ and T̂ commute. This Hamiltonian still commutes with N̂ , so as usual we will use eigenfunction of both T̂ and N̂ so that we can block diagonalise Ĥ and only study a block for fixed n and k. We will here only study the case n = 2 which is far less complex, we will also only consider the case f odd, since f even can be obtained with the same method. Let us introduce an eigenstate of both T̂ and N̂ : |ψτ i = c1 f X (T̂ /τ ) j−1 j=1 +c3 f X j=1 j=1 (T̂ /τ )j−1 |1, 1, 0, . . . , 0i (T̂ /τ )j−1 |1, 0, 1, . . . , 0i + . . . +c(f +1)/2 f X j=1 T̂ |ψτ i = τ |ψτ i |2, 0, 0, . . . , 0i + c2 f X (T̂ /τ )j−1 |1, 0, 0, . . . , 0, 1, 0, . . . , 0i τ = eik If we are interested in a particular block of Ĥ for a fixed τ (and n = 2) the energies can be determined by finding the eigenvalues of the matrix H2,τk that satisfies H2,τk ~c = E~c 17 Calculating the elements of H2,τk is straightforward and leads to the following (f + 1)/2 × (f + 1)/2 trigonal matrix: √ ∗ γ 2q 0 0 0 ... 0 √ 2q 1 q∗ 0 0 ... 0 .. ∗ 0 q 0 q 0 ... . .. ∗ q = 0 (1 + τ ) (3.7) H2,τk = − 0 0 q 0 q ... . .. .. .. .. . . . . 0 0 0 ... 0 q 0 q∗ 0 0 0 ... 0 q p 3.3.2 Behaviour when 1 = 0 This case is well known([5],[9]) and correspond to pthe quantum breathers. This means there is a new energy band E(k) = − γ 2 + 1620 cos2 (k/2) corresponding to a localised energy that remains here on a significant time scale before slowly delocalising. In the limit of f → ∞ there is also a continuum band located between ±20 cos(k/2) corresponding to non-localised states. The eigenvalues can be seen in B.3. 3.3.3 Resonance behaviour Resonance behaviour shown by [4] For arbitrary values of 1 , [4] showed numerically that we have the same global behaviour as in 3.3.2. But it highlights the fact that for 1 = γ, the propagation along the lattice is much faster and then the energy does not remain localised as we see on 3.1. This is this special behaviour for this particular value of 1 which we call resonance behaviour. [4] gives an expression for the energies involved in this resonance by considering only the localised states that are the one of the form(and their translations): |2, 0, . . . , 0i and |1, 1, 0, . . . , 0i Then the matrix 3.7 is only a 2 × 2 square matrix whose eigenvalues are easy to get: s γ − 1 γ − 1 2 E± = −γ − ± + 1620 cos2 (k/2) (3.8) 2 2 Without the approximation First we will assume that the matrix is semi-infinite that is f → ∞. Some corrections might be found in the large but finite case([10] for the usual 18 Figure 3.1: Numerical simulation of the evolution of nj . φt is for the time and n is for the lattice sites(taken from [4] thanks to C.Falvo) case) but I hadn’t enough time to prospect this. So let λ be an eigenvalue of matrix M = −H2,τk /|q|(to get the energy we will only have to multiply λ by −|q|) and let us introduce ~v = [v1 , v2 , . . . , vk , . . .] an associated eigenvector. We will also use g = γ/|q| and d = |q|/1 . We have then (M − λI)~v = 0, which can be written in term of coordinates: √ q∗ (g − λ)v1 + 2 v2 = 0 |q| √ q 1 q∗ 2 v1 + ( − λ)v2 + v3 = 0 (3.9) |q| d |q| q∗ q vk−1 − λvk + vk+1 = 0 k ≥ 3 |q| |q| The last equation is a two-step difference equation which general solution is: vk = Aµk+ + Bµk− ∗ q − λµ + |q| µ2 = 0. Then √ λ ± λ2 − 4 µ± = |q| 2q ∗ µ+ and µ− being the roots of q |q| We will now use a change of variables(J. Dorignac, private communication): λ = X + X1 so that: µ± = Xeiθ (+), 1 iθ e (−) using q = |q|eiθ X 19 As we want the solutions to be bounded(that is vanishing at infinity), let us choose vk = X k−2 ei(k−2)θ , k ≥ 2 with X between 0 and 1. The eigenvalues will be then determined using the first two equations of 3.9: X 3 + (d − g)X 2 + (dg + 1)X − d = 0 (3.10) Any solution of 3.10 with X between 0 and 1 will correspond to an energy of a bounded state. Even though this equation is not easy to solve analitically we can find some results. First, depending on the values of parameters d and g there is either 1 or 2 bounded solution. In the usual case when γ → 0 the breather band tends to go just at the border of the continuum, and with γ increasing the soliton band goes away of the continuum([5]). That gave us the idea to look for which values of the parameters we have a solution at the border of the soliton band. That is, changes in variables taken in account, when X = 1. This condition leads to the relation: 2 2|q|1 g= ⇔ γ= (3.11) 1−d 1 − |q| When this is verified we have 3 solutions: one is for X > 1, one is for 0 < X < 1 and of course X = 1. A numerical inquiry shows that for value of g greater than this limit we still have the first two solutions and the one that was X = 1 decrease to become a bounded solution(J. Dorignac, private communication). For values lower than the limit the last solution increases becoming a non-bounded solution. So we will take 3.11 as a limit between 1 and 2 bounded states. We now have to find an expression for these energies. We will use a low coupling approximation: |q| ∼ 0 γ ∼ 1 that is with the new parameters: g 1 and d 1. Then we can say that d − g ≈ −d − g, and d is an obvious root of 3.10 and we find the following roots (for g ≥ 2): p g ± g2 − 4 d, 2 Of course exactly two of them are between 0 and 1, as g is above the limit defined in 3.11. And if we go back to the energies we find for the bounded states: 2 + |q|2 ∼ −1 E1 = −γ E2 = − 1 1 So we have 2 breather bands located around −γ and −1. When γ = 1 , the energies are rather equal, which is important as the resonance behaviour is explained in [4] as a result of the hybridation of the 2 bands. This is another way to explain this hybridation qualitatively, and of course in the limit |q| = 0 these are exact results. 20 Chapter 4 Conclusion The classical part provided us with usual results, that is in fact we have the same behaviours except that γ becomes γ 0 . Such results are not very important since no new behaviour is found, but there is still one relevant thing: the fact that we can have γ 0 = 0 changes radically the behaviour for this value of the parameter. Concerning the quantum approach, in the dimer case we can fairly draw the same conclusion: there is no radical change of behaviour except for γ 0 = 0 that changes from a localised energy to a delocalised energy. About the lattice, we have one major change that is the resonance behaviour. But when looking closer, one could say that it is the same kind of conclusion as before, the fast delocalising energy being typically a linear effect and the localisation being typically nonlinear. So what is common to all the results is that for certain values of both the local and nonlocal nonlinearity, we see a surprisingly linear-like behaviour. It would be interesting to further study the lattice case to, for example, get better approximations(or exact) energies for the two breather bands. Other further works should also include 2 and 3 dimensional case as well as the study of non-translational invariant lattices. 21 Bibliography [1] J.C.Eilbeck,P.S. Lomdahl and A.C.Scott. The discrete self-trapping equation. Physica D, 16:318-338, 1985. [2] L.Bernstein, J.C.Eilbeck and A.C.Scott. The quantum theory of local modes in a coupled system of nonlinear oscillators. Nonlinearity 3 (1990) 293-323. [3] J.C. Eilbeck and M.Johansson. The discrete nonlinear Schrödinger equation-20 years on. Proceedings of the third conference: Localization & Energy Transfer in Nonlinear Systems (2002) 44-67 [4] C.Falvo, V.Pouthier and J.C. Eilbeck. Fast energy transfer mediated by multi-quanta bound states in a nonlinear quantum lattice. Submitted 2006 [5] A.C.Scott. Nonlinear Science (second edition) Oxford texts in applied and engineering mathematics, Oxford University Press (2003) [6] V.M.Kenkre and D.K.Campbell Self-traping on a dimer: Timedependent solutions of a discrete nonlinear Schrödinger equation Physical Review B, 34-7(1986) [7] J.C.Eilbeck,P.S. Lomdahl and A.C.Scott. Stationary solitons on finite lattices. Proceedings of the III International Symposium on Selected Topics in Statistical Mechanics, Dubna, USSR, 22-26 August 1984, 2, 328-339 (1985). [8] Derek F. Lawden. Springer-Verlag (1989) Elliptic functions and applications [9] J.C.Eilbeck Some exact results for quantum lattice problems. Proceedings of the Third Conference: Localization & Energy Transfer in Nonlinear Systems (2003) 177-186. [10] J.C.Eilbeck R.L.Pego On the eigenvalues and eigenvectors of some large trigonal matrices To be published 22 Thanks I would like to thank Chris Eilbeck for accepting to be my tutor for 4 months, for his disponibility, advice and his general kindness. I would also like to thank him for all the extra advice he gave me concerning living in Scotland and climbing. I also thank Jerome Dorignac for his precious contribution to the last part. I would like to thank Emily and Siobhan for helping me during the first days I was here, and I finally thank Nicol Craig and Scott Massie for beeing my flatmates and providing me interesting vues on the 8 = D problem. 23 Appendix A Details A.1 Eigenvalues of the stability matrix For the (↑↑) solution: λ=0 and − 40 − 20 1 N + 20 γN For the (↑ .) solution: λ=0 and − N 2 γ 2 + 2N 2 γ1 − N 2 21 + 420 For the (↑↓) solution: λ=0 A.2 and − 40 + 20 1 N − 20 γN Calculation details of the trimer parametrization Subtracting the first two equation of 2.7 leads to: W (a − b) = γ(a3 − b3 ) + 0 (b − a) + 1 (−b3 − a2 b + 2b2 a) assuming1 that a 6= b and using that −b3 − a2 b + 2b2 a = (a − b)(b2 − ab) we find that: W = γ(a2 + b2 + ab) − 0 + 1 (b2 − ab) If 1 = 0 you just have to ’complete the square’ in the a2 + b2 + ab expression to find a parametrization in term of sine and cosine. But here we introduce N to substitute a2 by N − 2b2 and find: b2 − ab = W + γN − 0 γ − 1 It is then straightforward to find 2.8 1 That explains why the parametrization does not take in account the ↑↑↑ branch 24 Appendix B Figures B.1 Behavour near the bifurcation point 1 = 0.9995(red) and Figure B.1: Behavour around the bifurcation( γ− 20 1.0001(green)) 25 B.2 Trimer bifurcation diagram when 1 = 0 600 540 480 Gamma 420 360 300 240 180 120 60 0 0 60 120 180 240 300 W Figure B.2: Trimer branches(no stability has been considered) 1 = 0, 0 = 30cm−1 The yellow curve is the ↑↑↑ branch The highest red curve is the ↑↑↓ branch The other red one is the ↑↑ . branch The green red is the ↑ .. branch 26 B.3 Energies when 1 = 0 Figure B.3: Eigenvalues for n=2, in grey the continuum and lower the breather band(taken from [9]) 27