Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Time-dependent perturbation theory

  1. Nov 20, 2009 #1
    Hi all

    Please look at this link (Search for the phrase "The quantum state at each instant can be expressed as a linear combination of the eigenbasis"): http://en.wikipedia.org/wiki/Perturbation_theory_(quantum_mechanics)

    If we write the wavefunction for the perturbed system as a (time-dependent) linear combination of the solutions to the unperturbed system, then do we assume that the perturbation is so weak that it does not change the energy-levels? If not, then I do not understand why we can write the perturbed wavefunction as a linear combination of the unperturbed wavefunctions. Because the probability of finding the particle in some state A is the square of the amplitude in front of the wavefunctions for the state A, but all the states in the linear combination are the "old" unperturbed wavefunctions.
     
    Last edited: Nov 20, 2009
  2. jcsd
  3. Nov 20, 2009 #2
    Yes, the new wavefunction is a linear combination of the "old" (unperturbed) energy eigenfunctions. But the energy levels always change if you have a time-dependent Hamiltonian. So although we have an explicit linear combination for the wavefunction, it is actually not expanded in terms of eigenfunctions of the (time-dependent) Hamiltonian. You can still interpret the coefficients in front as "the amplitude that you find the particle in that state". But the state is simply not an energy eigenstate (which is ok).

    However, you can imagine that the perturbation is turned on and off again (slowly). In that case you do end up with an expansion in terms of eigenfunctions of the (new) Hamiltonian.
     
  4. Nov 20, 2009 #3
    One-time measurement does not say anything about the quantum state - you have to perform multiple measurements. Then you will obtain an average energy which does not coincide with any "old" eigenvalue. Even with constant coefficients such a superposition is a state without certain energy. An example is a coherent state of photons.
     
  5. Nov 20, 2009 #4
    But what is the idea behind this? So we find out that the particle is in some state A, which is only an eigenstate of the unperturbed Hamiltonian. Thus we don't know anything?
     
  6. Nov 20, 2009 #5
    Yes, you do actually. For instance, you can calculate transition rates from one state to another state. Take for example electron orbitals of an atom in the presence of an oscillating electromagnetic field (a light pulse or a laser beam). We can assume we have one electron which starts in the lowest state. At one point the light pulse is turned on (weakly). The orbitals are no longer the energy eigenfunctions. Using this technique we can calculate what the overlap of the wavefunction with the other orbitals is as a function of time. You will notice that the electron starts to oscillate between the ground state and the first excited states (depending on the energy beam though -- other orbitals can be involved as well)

    You can go as far as to determine how long the laser beam has to be turned on in order for the electron to fully transfer from the ground state to the first excited state. Sounds pretty practical to me!

    A more general result which follows from perturbation theory is known as Fermi's Golden rule. It tells you the transition rate from one energy eigenstate to other eigenstates -- very useful for optical purposes, but far more general than that.
     
  7. Nov 20, 2009 #6

    Ok, so the above oscillations mean that the true (perturbed) eigenfunctions are a superposition of the old (unperturbed) eigenfunctions? And also that these new, perturbed eigenfunctions change over time?

    And thank you!
     
  8. Nov 20, 2009 #7
    Yes, that statement is actually true in general when you add a peturbation: the eigenfunctions are a complete basis of the Hilbert space. You can view the new eigenfunctions as a change of basis, so naturally this basis is expandable in terms of the old one (the Hilbert space stays the same).

    Indeed they do. The point is that you do not know what the exact expression is of these time-dependent eigenfunctions. We simply can't solve the corresponding equations of motion. This is where perturbation theory comes into play.

    The number of Hamiltonians for which you can determine the exact form of the eigenfunctions is remarkably small. The whole point of perturbation theory is that we can still make useful statements about the system using these solved Hamiltonians.
     
  9. Nov 20, 2009 #8
    Thank you very much. Your answers cleared things up for me.

    Have a nice weekend.
     
  10. Nov 20, 2009 #9
    The perturbation theory gives approximate time-dependent coefficients, the exact time-dependent wave function contains exact time-dependent coefficients at the non-perturbed basis wave-functions.
     
    Last edited: Nov 20, 2009
  11. Nov 20, 2009 #10

    dextercioby

    User Avatar
    Science Advisor
    Homework Helper

    I have a question: is the perturbed (time-dependent) hamiltonian still a self-adjoint operator ?
     
  12. Nov 20, 2009 #11
    By the way, I thought of a question: Let us say that at some time t0 we have the perturbed wavefunction given by the following linear combination of unperturbed eigenstates

    [tex]\Psi(x,t_0) = 0.4\psi_1+0.6\psi_2[/tex].

    In this case, what do the amplitudes 0.4² and 0.6² represent? Nothing, right? It is only in the case when the coefficients are 1 and 0 respectively that it is interesting, because then we can say with 100% which state the particle is in, correct?
     
  13. Nov 20, 2009 #12
    No, you are wrong. The ratio (0.4/0.6)2 is the relative population of the corresponding states (your wave function is not normalized). Yes, the wave function can be a superposition of different eigenstates. The energy is not certain is such a state: there is an average energy and its dispersion. For example, after scattering atomic wave function becomes a superposition of different excited states.
     
  14. Nov 20, 2009 #13
    Regarding normalization: Yes, my fault. Sorry.



    I don't follow this argument. If e.g. we have [itex]\Psi(x,t_0) = \sqrt{0.05}\psi_1+\sqrt{0.95}\psi_2[/itex], and we look at one particle, then what can I say about this particle? That upon measurement, then there is 95% chance of the wavefunction collapsing to the perturbed state described by unperturbed state [itex]\psi_2[/itex]?
     
  15. Nov 20, 2009 #14
    No. In QM one looks at a quantum state, not at a particle.

    QM does not describe "a single particle" but a "single" quantum state Ψ. Ψ can be represented as a superposition of some basis wave-functions. Different measuring devices give different measured basis states. Think of an interference pattern. It consists of many points. The screen "measures" the particle coordinates. They may be different since the wave function is not an eigenstate of coordinate operator. The whole pattern describes an electron/photon quantum state behind the double-slitted diaphragm. There is no interest in one point on the screen. It is the pattern which is interesting to predict. The wave
    function does it.

    One point on a screen is like one letter in a text - it does not carry much information. We are interested in the whole text which is an ensemble of letters. This gives the description of a quantum state.
     
  16. Nov 20, 2009 #15
    I see your point. But I must admit, I do not have any idea of what those probability amplitudes mean, since they "belong" to the unperturbed states. That was why I wanted to look at the single quantum state: To make things more "down to earth".
     
  17. Nov 20, 2009 #16
    OK, it's easy: take an atom in its ground state Ψin (an eigenstate of non-perturbed Hamiltonian). When I say: "Take", it means your preparatory device supplies (or contains as a target) the atoms only in this state. You make sure by multiple measurements.

    Then you scatter fast electrons off such atoms. After scattering the target atom can transfer into another state Ψout which is a superposition of the ground and all possible excited states allowed with the energy conservation law: Ψout = ∑ Cnψn. The excited atoms radiate. Observing the spectral lines you note that they belong to different final states ψn. So Ψout is not a state with certain atomic energy. It is so because the projectile energy loss is distributed between the kinetic energy of the final atom and its "internal" energetic state ψn.

    Similarly, if you act with a variable electric filed on atom, its state cease to be stationary but becomes time dependent. Again it is a superposition of the basis wave functions but with time-dependent coefficients. It is natural: in a variable external field the energy is not conserved. So the energy measurements (spectral lines) give spread values (populations of certain excited eigenstates change with time). A resonant external filed may populate (pump) certain eigenstates and deplete the initial state.
     
    Last edited: Nov 20, 2009
  18. Nov 20, 2009 #17
    You're approaching it too practical. Try to look at it from a more formal point of view: we would like to able to keep track of the wavefunction: given a the state at some time, how does it evolve? Sure, we have the Schrodinger equation which dictates how states evolve, but that doesn't mean we're able to solve this equation.

    The method described here shows us how we can use a different basis for our Hilbert space and describe how the system evolves with respect to that state -- even though it's not the true eigenbasis of the Hamiltonian. This is why "solved" Hamiltonians are so useful, and are not just simple pedagogical examples. Try to appreciate that statement, cause it's quite a big one ;)

    Having said that: recall that in time-independent perturbation theory (present on the same wiki page you linked to) there is a treatment to solve the new eigenfunctions in terms of the old ones. So we can actually switch to the true eigenbasis at any timeslice (I'm not saying it's easy or the right approach, but it can be done, at least approximately). This only holds at that particular timeslice though.
     
  19. Nov 20, 2009 #18
    The reason why I am looking at it this way is because there is a plot in my book showing the square modulus of the two coefficients (they oscillate). And if I choose some time t0 where P2=0.8 and P1=1-P2=0.2, then I really can't see what the physical interpretation of these probabilities is.

    It might be that I am missing some point in both your replies, but I still cannot interpret the probabilities.
     
  20. Nov 20, 2009 #19
    Look at an interference pattern. Its picture is still, the probabilities do not depend on time. Then look at the interference picture while one varies the slit widths and the distance between the slits. The picture will be varying like in a movie. The probabilities depend on time now. Any time slice gives different picture.
     
  21. Nov 21, 2009 #20
    I must admit, the analogy with the interference pattern does not help me much. I want to keep this practical, so I will use [itex]
    \Psi(x,t_0) = \sqrt{0.05}\psi_1+\sqrt{0.95}\psi_2
    [/itex] again. Does this wave equation mean that 95% of the quantum states are in Ψ2, or that 95% are in the perturbed eigenstates, which can be written as Ψ2?
     
Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook