aa r X i v : . [ qu a n t - ph ] D ec QUANTUM HAMILTON-JACOBI THEORY
Marco Roncadelli
INFN, Sezione di Pavia, Via A. Bassi 6, I-27100 Pavia, Italy,and Dipartimento di Fisica Nucleare e Teorica, Universit`a di Pavia, Italy ∗ L. S. Schulman
Physics Department, Clarkson University,Potsdam, New York 13699-5820, USA † (Dated: October 31, 2018)Quantum canonical transformations have attracted interest since the beginningof quantum theory. Based on their classical analogues, one would expect themto provide a powerful quantum tool. However, the difficulty of solving a nonlinearoperator partial differential equation such as the quantum Hamilton-Jacobi equation(QHJE) has hindered progress along this otherwise promising avenue. We overcomethis difficulty. We show that solutions to the QHJE can be constructed by a simpleprescription starting from the propagator of the associated Schr¨odinger equation.Our result opens the possibility of practical use of quantum Hamilton-Jacobi theory.As an application we develop a surprising relation between operator ordering andthe density of paths around a semiclassical trajectory. PACS numbers: 03.65.Ca
Canonical transformations play a central role in classical mechanics [1]. From the earliestdays of quantum mechanics, the importance of quantum canonical transformations (QCT’s)has been recognized [2] and their properties have been systematically investigated by Jor-dan [3], London [4], Dirac [5] and Schwinger [6], among others. Schwinger’s framework—based on the Quantum Action Principle [7]—provides the most suitable context to define aQCT as ˆ q → ˆ Q, ˆ p → ˆ P , H (ˆ q, ˆ p, t ) → K ( ˆ Q, ˆ P , t ), where all canonical variables pertain to thesame dynamical system S with N degrees of freedom [8].Owing to the formal similarities between classical and quantum mechanics, QCT’s closelyresemble their classical counterparts. In particular, one of four possible sets of independentcanonical variables (ˆ q, ˆ Q ), (ˆ q, ˆ P ), ( ˆ Q, ˆ p ), (ˆ p, ˆ P ) must be selected to represent a QCT ex-plicitly, and we denote by W (ˆ q, ˆ Q, t ), W (ˆ q, ˆ P , t ), etc., the associated operator generatingfunctions. Choosing the set (ˆ q, ˆ Q ), a QCT can be written as (1 ≤ i ≤ N )ˆ p i = ∂∂ ˆ q i W (ˆ q, ˆ Q, t ) , (1)ˆ P i = − ∂∂ ˆ Q i W (ˆ q, ˆ Q, t ) , (2) ∗ Electronic address: [email protected] † Electronic address: [email protected] K ( ˆ Q, ˆ P , t ) = H (ˆ q, ˆ p, t ) + ∂∂t W (ˆ q, ˆ Q, t ) . (3)The presence of noncommuting operators in the generating function makes QCT’s differ-ent from the classical ones and is ultimately responsible for the difference between classicaland quantum mechanics [9]. As emphasized by Jordan and Dirac, the resulting operator-order ambiguity should be fixed by enforcing well-ordering : operators represented by capitalletters should always stay to the right of those labelled by lower case letters. This meansthat W (ˆ q, ˆ Q, t ) should have the structure W (ˆ q, ˆ Q, t ) = X α f α (ˆ q, t ) g α ( ˆ Q, t ) , (4)for suitable functions f α ( · ) and g α ( · ). Throughout, we will suppose that operator generatingfunctions are well-ordered . Note that with well-ordering a quantum generating function like W (ˆ q, ˆ Q, t ) is uniquely defined by the replacements q → ˆ q , Q → ˆ Q in a given c-numberfunction W ( q, Q, t ) [10].As in classical mechanics, the quantum time evolution is described by a canonical trans-formation bringing the canonical variables in the Heisenberg picture ˆ q ( t ), ˆ p ( t ) to constantvalues at some initial time t . In addition, ˆ q ( t ), ˆ p ( t ) can be derived from Eqs. (1) and (2),provided that the transformed Hamiltonian vanishes. As a consequence, the operator gen-erating function W (ˆ q, ˆ Q, t ) obeys the operator quantum Hamilton-Jacobi equation (QHJE) H (cid:18) ˆ q, ∂∂ ˆ q W (ˆ q, ˆ Q, t ) , t (cid:19) + ∂∂t W (ˆ q, ˆ Q, t ) = 0 . (5) W (ˆ q, ˆ Q, t ) should be a complete solution of Eq. (5), i.e., it should depend on N indepen-dent “integration constants” ˆ Q i . As in classical mechanics, the operator Hamilton-Jacobiequation, Eq. (5), provides an independent formulation of the theory. Yet the formidabledifficulty of finding solutions to this nonlinear operator partial differential equation hashindered progress along this otherwise promising avenue.Our aim is to show that this stumbling block can be sidestepped, thereby opening theway to exploiting the operator QHJE as a calculational tool. As we will demonstrate,the solutions to the operator QHJE arise by a simple prescription from the solutions ofthe Schr¨odinger equation for the same Hamiltonian. In particular, the operator generatingfunction W (ˆ q, ˆ Q, t ) arises from the quantum propagator. Implications of our result will bediscussed after we have completed its demonstration.We are concerned throughout with the general
Weyl-ordered
Hamiltonian [11] H (ˆ q, ˆ p, t ) = 12 a ij (ˆ q )ˆ p i ˆ p j + ˆ p i a ij (ˆ q )ˆ p j + 12 ˆ p i ˆ p j a ij (ˆ q )+ b i (ˆ q )ˆ p i + ˆ p i b i (ˆ q ) + c (ˆ q ) , (6)where a ij ( · ), b i ( · ), and c ( · ) are functions of ˆ q k , and summation over repeated Latin indicesfor the degrees of freedom of S is understood. Employing the shorthand ˆ W ≡ W (ˆ q, ˆ Q, t ),Eq. (5) reads 12 a ij (ˆ q ) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j + ∂ ˆ W∂ ˆ q i a ij (ˆ q ) ∂ ˆ W∂ ˆ q j + 12 ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j a ij (ˆ q ) + b i (ˆ q ) ∂ ˆ W∂ ˆ q i + ∂ ˆ W∂ ˆ q i b i (ˆ q ) + c (ˆ q ) + ∂ ˆ W∂t = 0 . (7)Since we are looking for the relationship between the operator QHJE and the Schr¨odingerequation, we turn Eq. (7) into a c-number partial differential equation. Hence we sandwichEq. (7) between h q | and | Q i , finding12 a ij ( q ) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i a ij (ˆ q ) ∂ ˆ W∂ ˆ q j (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + 12 h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j a ij (ˆ q ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + b i ( q ) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i b i (ˆ q ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + c ( q ) h q | Q i + h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂t (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = 0 . (8)To evaluate the matrix elements in Eq. (8), we make repeated use of the canonical commu-tation relations. In this connection, we recall that for an arbitrary function G ( · )[ G (ˆ q ) , ˆ p i ] = i ~ ∂G (ˆ q ) ∂ ˆ q i . (9)By inserting Eq. (1) into (9), we obtain ∂ ˆ W∂ ˆ q i G (ˆ q ) = G (ˆ q ) ∂ ˆ W∂ ˆ q i − i ~ ∂G (ˆ q ) ∂ ˆ q i . (10)We begin by taking G (ˆ q ) ≡ b i (ˆ q ), G (ˆ q ) ≡ a ij (ˆ q ) and G (ˆ q ) ≡ ∂a ij (ˆ q ) /∂ ˆ q j . Accordingly, Eq.(10) allows us to rewrite Eq. (8) as2 a ij ( q ) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + 2 (cid:18) b i ( q ) − i ~ ∂a ij ( q ) ∂q j (cid:19) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i + (cid:18) c ( q ) − i ~ ∂b i ( q ) ∂q i − ~ ∂ a ij ( q ) ∂q i ∂q j (cid:19) h q | Q i + h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂t (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = 0 . (11)At this point, we denote by W ( q, Q, t ) the c-number function that uniquely produces W (ˆ q, ˆ Q, t ) by the substitution q → ˆ q , Q → ˆ Q . Explicit use of Eq. (4) yields h q | ˆ W | Q i = W ( q, Q, t ) h q | Q i , (12) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂t (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = ∂W ( q, Q, t ) ∂t h q | Q i , (13) h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = ∂W ( q, Q, t ) ∂q i h q | Q i , (14)and furthermore h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = X α,β h q (cid:12)(cid:12)(cid:12)(cid:12) ∂f α (ˆ q, t ) ∂ ˆ q i g α ( ˆ Q, t ) ∂f β (ˆ q, t ) ∂ ˆ q j g β ( ˆ Q, t ) (cid:12)(cid:12)(cid:12)(cid:12) Q i . (15)What remains to be done is to disentangle Eq. (15). To this end, we first take G (ˆ q ) ≡ ∂f β (ˆ q, t ) /∂ ˆ q j in Eq. (10) to get X α ∂f α (ˆ q, t ) ∂ ˆ q i g α ( ˆ Q, t ) ∂f β (ˆ q, t ) ∂ ˆ q j = ∂f β (ˆ q, t ) ∂ ˆ q j X α ∂f α (ˆ q, t ) ∂ ˆ q i g α ( ˆ Q, t ) − i ~ ∂ f β (ˆ q, t ) ∂ ˆ q i ∂ ˆ q j . (16)We next multiply Eq. (16) by g β ( ˆ Q, t ) on the right and sum over α , thereby obtaining X α,β ∂f α (ˆ q, t ) ∂ ˆ q i g α ( ˆ Q, t ) ∂f β (ˆ q, t ) ∂ ˆ q j g β ( ˆ Q, t )= X α,β ∂f β (ˆ q, t ) ∂ ˆ q j ∂f α (ˆ q, t ) ∂ ˆ q i g α ( ˆ Q, t ) g β ( ˆ Q, t ) − i ~ X β ∂ f β (ˆ q, t ) ∂ ˆ q i ∂ ˆ q j g β ( ˆ Q, t ) , (17)which allows us to rewrite Eq. (15) as h q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂ ˆ W∂ ˆ q i ∂ ˆ W∂ ˆ q j (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q i = (cid:18) ∂W ( q, Q, t ) ∂q i ∂W ( q, Q, t ) ∂q j − i ~ ∂ W ( q, Q, t ) ∂q i ∂q j (cid:19) h q | Q i . (18)As a consequence, Eq. (11) takes the form2 a ij ( q ) (cid:18) ∂W ( q, Q, t ) ∂q i ∂W ( q, Q, t ) ∂q j − i ~ ∂ W ( q, Q, t ) ∂q i ∂q j (cid:19) +2 (cid:18) b i ( q ) − i ~ ∂a ij ( q ) ∂q j (cid:19) ∂W ( q, Q, t ) ∂q i + c ( q ) − i ~ ∂b i ( q ) ∂q i − ~ ∂ a ij ( q ) ∂q i ∂q j + ∂W ( q, Q, t ) ∂t = 0 . (19)This derivation makes it natural to regard Eq. (19) as the c-number QHJE associated withthe operator QHJE (5) for S described by the quantum Hamiltonian (6).The physical significance of Eq. (19) becomes clear by setting ψ ( q, Q, t ) ≡ exp { ( i/ ~ ) W ( q, Q, t ) } . (20)A straightforward (if tedious) calculation shows that ψ ( q, Q, t ) obeys precisely theSchr¨odinger equation associated with the quantum Hamiltonian (6) in the variables q, t [12].Hence – thanks to Eqs. (12) and (20) – starting from a solution W (ˆ q, ˆ Q, t ) we get a solution ψ ( q, Q, t ) of the corresponding Schr¨odinger equation depending on N independent constants Q i . We stress that this result holds true even for solutions W (ˆ q, t ) of the operator QHJEthat are independent of ˆ Q , since all equations from (7) onward could have been multipliedby R dQ φ ( Q ), with φ ( Q ) arbitrary. What is more important for us, the argument can beturned around, because W (ˆ q, ˆ Q, t ) can be uniquely obtained from W ( q, Q, t ) by enforcing well-ordering . Therefore, from a solution ψ ( q, Q, t ) of the Schr¨odinger equation dependingon N independent constants Q i we get W (ˆ q, ˆ Q, t ), and from any particular solution ψ ( q, t )we can construct a particular solution W (ˆ q, t ).So far, we have focused on showing that W ( q, Q, t ) satisfies a certain differential equation.As we demonstrate below, by use of appropriate boundary conditions we get more specificinformation. Namely, the solution of Schr¨odinger’s equation that results from the operatorgenerating function W (ˆ q, ˆ Q, t ) is precisely the quantum propagator K ( q, Q, t ) [13]. Sinceany solution of the Schr¨odinger equation arises by convolving an arbitrary wave functionwith the propagator, we conclude that any solution of the operator QHJE can ultimatelybe constructed in terms of the propagator.This will allow solutions of and approximations to the operator QHJE to be obtained,since a wealth of information is available on the corresponding solutions to Schr¨odinger’sequation. In particular, once an exact or approximate ˆ W has been constructed, one canobtain the time dependence of operators, using Eqs. (1) and (2).We proceed to prove that ψ ( q, Q, t ) = K ( q, Q, t ). Since both quantities satisfy the sameSchr¨odinger equation, which is first order in time, all we need show is that they have thesame boundary conditions at t = 0. The propagator of course is δ ( q − Q ) at t = 0. Toshow that ψ ( q, Q, t ) shares this property, we must look at the behavior of ˆ W for t →
0. Weexpect ˆ W to generate the identity transformation in the limit t →
0, but there is a slightcomplication: As in classical mechanics [1], the identity transformation using the (ˆ q, ˆ Q )variables does not have a simple form.A way out of this difficulty relies on the observation that for a nonsingular potentialthe solution to the classical Hamilton-Jacobi equation approaches that of the free particlefor t →
0. Thus, for sufficiently small t the classical generating function has the form F ( q, Q ) = m ( Q − q ) / t . We use this to guess the limit of the operator ˆ W for t →
0, andfrom that to obtain the corresponding limit of the c-number function W ( q, Q, t ). The firstobservation is that as a candidate for the small- t limit of ˆ W , the well-ordered operator formof F ( q, Q ) (which contains − q ˆ Q ) does not work, which is to say, it does not satisfy Eq. (5).To see this in detail—and to see the cure—we assume the following small- t limiting form forˆ W ˆ W = m t (cid:16) ˆ Q − q ˆ Q + ˆ q (cid:17) + g ( t ) . (21)Substituting into Eq. (5), the squaring of ˆ W generates a term − (ˆ q ˆ Q + ˆ Q ˆ q ), rather than − q ˆ Q , so that satisfying Eq. (5) requires0 = m t [ˆ q, ˆ Q ] + ∂g ( t ) ∂t . (22)For small t , one can again neglect the influence of the potential terms and the commutatorcan immediately be deduced from the relation ˆ q = ˆ Q + ˆ P t/m , the solution of the free particleHeisenberg equations of motion. Eq. (22) now becomes ∂g/∂t = i ~ / t and we obtainˆ W = m t (cid:16) ˆ Q − q ˆ Q + ˆ q (cid:17) + i ~ t , for t → , (23) ψ ( q, Q, t ) = const · r t exp (cid:18) i ~ m t (cid:0) Q − qQ − q (cid:1)(cid:19) , for t → , (24)with the constant in Eq. (24) arising from integrating ∂g/∂t = i ~ / t . It is remarkable that,aside from the constant (which is not fixed by ˆ W ), the ˆ q ˆ Q -commutation relation has givenus precisely the correct time-dependence of the propagator. This completes our proof.The c-number QHJE (19) has repeatedly attracted interest. For instance, Eq. (19) hasbeen derived from a diffeomorphic covariance principle based partly on an SL (2 , C ) algebraicsymmetry of a Legendre transform [14]. Alternatively, Eq. (19) has been taken as thestarting point of a classical-like strategy to define c-number quantum action-angle variablesin quantum mechanics [15]. We also remark that a variant of Eq. (19) has been derivedwithin the phase-space path-integral approach to quantum mechanics [16].We next show the power of the relation we have just developed between W ( q, Q, t ) and K ( q, Q, t ), using, as suggested above, known information about the propagator. Considera situation where the semiclassical approximation is valid and there is but one classicalpath between the initial and final points. Then in this approximation, as is well-known, K ( q, Q, t ) = const · p det ∂ S/∂q∂Q exp( iS ( q, Q, t ) / ~ ), with S ( q, Q, t ) Hamilton’s principalfunction (a solution of the classical Hamilton-Jacobi equation). It then follows from ourresult that W (ˆ q, ˆ Q, t ) | WO = S (ˆ q, ˆ Q, t ) | WO − i ~ log det ∂ ˆ S/∂q∂Q | WO , where “WO” stands for“well-ordered.” Now imagine that this expression is inserted in Eq. (5). If not for the wellordering, S alone would solve the equation. Therefore we conclude that the effect of the well-ordering is precisely to demand the presence of the additional term, i ~ log det ∂ S/∂q∂Q (where “WO” has been dropped because there is already an ~ in the expression). But thatadditional term (famously) has a meaning of its own: it goes back to van Vleck and representsthe density of paths along the classical path; it plays an essential role, for example, in theGutzwiller trace formula. What our result says is that this density of paths can be thoughtof as arising from the commutation operations necessary to bring S to well-ordered form.Thus the purely quantum issue of commuting operators produces a quantity that one wouldhave thought is exclusively derivable from classical mechanics.We remark that this relation took us completely by surprise. To check it, we worked thesimplest non-trivial example we could (our proof above comparing the boundary conditionsfor K and W already showed it to be true for the free particle case). Let H = p / V with V = V Θ( a/ − | x | ) and x in one dimension. To lowest order in V the action is S ( x, y, t ) =( x − y ) / t − V at/ ( x − y ) for y < − a/ x > a/
2. We checked our relation, with x → ˆ q and y → ˆ Q and with well-ordering implemented by h / (ˆ q − ˆ Q ) i WO = R ∞ du exp( − u ˆ q ) exp( u ˆ Q ).Using the Baker-Campbell-Hausdorff formula and other techniques and keeping only lowestorder in V and ~ , indeed the relation checked out!In conclusion, we have shown how to construct solutions to the operator quantum QHJEstarting from the quantum propagator K ( q, Q, t ) for the same Hamiltonian. Explicitly, once K ( q, Q, t ) is known we get its “complex phase” W ( q, Q, t ) via Eq. (20). Then, by demanding well-ordering , the replacement q → ˆ q , Q → ˆ Q uniquely produces the operator W (ˆ q, ˆ Q, t ).Alternatively, by convolving K ( q, Q, t ) with an arbitrary φ ( Q ) we produce any solution of theSchr¨odinger equation. Finally, by replacing q → ˆ q in its “complex phase” we get a solution W (ˆ q, t ) of the operator QHJE. While this is obviously true for exact propagators, it alsoenables one to find approximate solutions to the operator QHJE by exploiting approximatepropagators. In particular we used the semiclassical approximation to the propagator toshow that the commutation operations establishing well-ordering provide just what is neededto get the density of paths around the classical path. This density of paths satisfies acontinuity equation which, as O’Raifeartaigh and Wipf [17] emphasize, is in a sense of order ~ (even though it involves classical quantities only and has no ~ in it!). Although our proofestablishes this surprising relation, there remains the provocative question of understandingits intuitive basis. Acknowledgements.
We thank the Max Planck Institute for the Physics of Complex Sys-tems, Dresden, for kind hospitality. This work was supported by NSF Grant PHY 0555313. [1] H. Goldstein,
Classical Mechanics , 2 n d ed. (Addison-Wesley, Reading, Mass. 1966).[2] M. Born, W. Heisenberg and P. Jordan, Z. Phys. , 557 (1926). P. A. M. Dirac, Proc. Roy.Soc. A , 621 (1927).[3] P. Jordan, Z. Phys. , 383 (1926); , 513 (1926).[4] F. London, Z. Phys. , 915 (1926).[5] P. A. M. Dirac, Physik. Zeits. Sowjetunion , 64 (1933).[6] J. Schwinger, Phys. Rev. , 914 (1951); Phys. Rev. , 713 (1953).[7] J. Schwinger, Quantum Kinematics and Dynamics (Benjamin, New York, 1970).[8] As usual, quantum operators representing dynamical variables are denoted by a hat.[9] T. F. Jordan and E. C. G. Sudarshan, Rev. Mod. Phys. , 515 (1961).[10] We follow Dirac’s terminology, according to which ordinary, i. e., commuting, variables arecalled “c-numbers.”[11] See, e. g., T. D. Lee, Particle Physics and Introduction to Field Theory (Harwood Acad. Pub.,New York, 1981).[12] Verifying this assertion is a textbook exercise, one that is done for the case a = 1 / m , b = 0, c = V in Eq. (12), § Quantum Mechanics , 1 s t ed. (Benjamin, New York,1966).[13] See, e. g., L. S. Schulman, Techniques and Applications of Path Integration (Wiley, New York,1981; Dover, New York, 2005).[14] A. Faraggi and M. Matone, Int. J. Mod. Phys. A , 1869 (2000).[15] R. A. Leacock and M. Padgett, Phys. Rev. Lett. , 3 (1983); Phys. Rev. D , 2491 (1983).[16] V. Periwal, Phys. Rev. Lett. , 4366 (1998).[17] L. O’Raifeartaigh and A. Wipf, Found. Phys.18