arXiv:1605.09102v3 [physics.class-ph] 5 Feb 2017 rmteprmtro ice emk defor- a make that We algorithm iterative an circle. apply a and different ansatz of too mation perimeter not the is which from Conserved (C.E.S.), a Surface along Energy happens geometric Motion spanned coordinates, apply. space momentum methods and phase two-dimensional position by a In sym- parity metry. with oscillator anharmonic dimensional, notecr urclmbcuete rsn se- fall present they not challenges because do technical curriculum but rious core physics, the into in elliptic interpretations of set many a defining of by functions solution motion the pendulum’s optimized the and standardized Jacobi condition tial renaissance. Galilean of to contrary musings predictions the support data Experiments digital yielding theory. perturbation log- in false. a alternative provides resounds ical mechanics now classical of amplitude flourishing on The pe- not the depends that riod pronouncement the initial time, Galileo’s characteristic period. one as manifest pendulums, nFg ,oclae through oscillates as describing 1, pendulum, for un- A Fig. long on in too depend conditions. heartbeats is while biological year apple, predictable an or of day fall A the measurement time? easy of of prevent difficulties Measurement What length. equations. most space and periments Poincar´e nteodro n eodwhen second one of order the on ouinmtos including: approximate of methods, variety solution a contains literature physics adn h iedpnetbhvo foscillating of behavior dependent pendulums time the garding g ≈ u oe ehdvestepnuu saone- a as pendulum the views method novel Our h iigo edlmde eedo t ini- its on depend does pendulum a of timing The pc n iepa onainlrlsi l ex- all in roles foundational play time and Space itr rdt aie iherydsoeisre- discoveries early with Galileo credits History 9 . 81( nyrqie h ento fastandard a of definition the requires only intere.Mn fteueu oin eengeneralize herein notions dimensions. useful higher the in comp of Many enables analogy theories. tion quantum/classical the experiment ap provides of analysis functions Development data elliptic recursive Jacobian with The equatio combined the precision. one-dimen produces arbitrary a algorithm to novel of Surfaces simple, incre A Energy with Conserved oscillator. decreases the or for increases ansatz tion period pendulum’s plane a 2,10,11 4,5 m/s h ml nl prxmto fe al oepanexperi explain to fails often approximation angle small The 1 h aoinelpi ucin have functions elliptic Jacobian The . entcdthe noticed He . 2 2,3 .INTRODUCTION I. n aoia etraintheory perturbation canonical and , ). ln edlmadByn yPaeSaeGeometry Space Phase by Beyond and Pendulum Plane fe eedeadAe,C.G.J. Abel, and Legendre After . eateto hsc,Uiest fAkna,Fayettevil Arkansas, of University Physics, of Department 6,7 sa lentv the alternative an As . isochrony dhoc ad time etn scale a setting , l ≈ 8,9 Lindstedt- , 1 fidentical of / Dtd eray7 2017) 7, February (Dated: 4( m and ) rde Klee Bradley 12 . xaso fteJcb litcfunctions series Elliptic a Jacobi obtain the we precision, of high expansion to equa- motion plagiarize pendulum of the tions to solving By temptation references. standard refuses and assumptions depen- time so readily. systems, follows dence integrable dimensional One are value. conserved oscillators one of to convergence energy force the to functions undetermined sets and I.1 ipepnuu n oriaegoer.Half geometry. coordinate and height pendulum Simple 1. FIG. edlmsmto hog n eids ogas long so obeys period motion one the through motion pendulum’s error. of free and arbitrary-precision as tion isnpecito ntecascllimit. Sommerfeld- classical the the to in quantum prescription equivalent Wilson a is derive which we oscillators, condition, Comparing quartic for energy. conserve results nearly wave- must approximate functions mechanics quantum In perturbation theories. time-independent between analogy cal agree closely predictions. derived that carefully parameters Anal- with extracted a data. utilizes digital yields setup produce ysis The to mouse range. USB this modified in place takes ment h rga fdrvto novsn mistaken no involves derivation of program The hr prxmto ucst eciethe describe to suffices approximation third A liaeyw eeltedtiso semiclassi- a of details the reveal we Ultimately † cn era pca ae hitexperiment Thrift case. special a as pear s so oinadtepro foscillation of period the and motion of ns oitgal n o-nerbesystems non-integrable and integrable to a lvrfiaino elkonpredictions. well-known of verification al ( sn mltd.W aeaperturba- a make We amplitude. asing ,α ϑ, eemnstepro foscillation. of period the determines rsno ieidpnetperturba- time-independent of arison inl aiysmerc anharmonic parity-symmetric, sional, etldt,de o vnpeitif predict even not does data, mental .Ti xriedsigihstederiva- the distinguishes exercise This ).
g m θ l e R72701 AR le, ∈ θ [ − π/ z 2 π/ , θ 0 ] hitexperi- thrift A 2]. b sn a ( ,α ϑ, x ) 2
II. DIMENSIONAL ANALYSIS gives a concise result
2 b dx T0 = lim 4 (2) a→0 0 2√g(a−z) The plane pendulum consists of a massive bob at- R tached by a string or a rod, assumed massless, to an = lim 2 l a dz =2π l , axle as in Fig. 13. Gravity acts on the bob with verti- a→0 g 0 √z(a−z) g q R q cal force mg, and the attachment applies a response which requires small-angle identity s x 2√lz force, the tension. As time elapses the bob swings to change from the circular line element≈ ds≈to the and executes a periodic motion along a circular tra- horizontal dx, and finally to the vertical dz. The jectory of radius l. In librational motion the sign of θ˙ simple result only applies in the limit a 0. alternates while the pendulum reaches a maximum In a general one-dimensional oscillation→ with deflection θ at regular intervals throughout the 0 small-amplitude period T , we usually have some- experiment.± The time of one complete oscillation, 0 thing along the lines say from θ to θ back to θ , is called the period. 0 − 0 0 Table I collects the relevant physical quantities, T (α, ǫ)= f(α, ǫ) T0, (3) read directly from Fig. 1. Dimensional symbols [ L ], with f(α, ǫ) a complicated function of dimensionless [ M ], and [ T ], denote length, mass, and time. The energy α, and ǫ, structure constants of the potential quantities l/g, a/g, b/g all have dimension { } energy. of time, [ T p]. Assumingp p Galileo’s observation cor- With the pendulum experiment the trouble is in rect, l/g must be the dimensional scale of time the initial conditions. Each initial condition deter- becausep quantities a,b depend on the amplitude mines one critical parameter of motion. { } 1 α = a/l = 1 cos(θ ) = sin2(θ /2), (4) 2 − 0 0 TABLE I. Dimensional Quantities. proportional to the total energy. In the simple har- 2 Symbol Dimension Trigonometric Form monic approximation α tends to zero as θ0. Consid- ering this fact, the hypothesis that factor f(α, ǫ) l [ L ] · → 2 f(α) has a non-terminating power series expansion a [ L ] l sin (θ0/2) seems likely. Referencing the expansion of K(α)13 b [ L ] l sin(θ0/2) cos(θ0/2) −2 we have g [ L ] [ T ] · ∞ T (α) 2 (2 n 1)!! 2 m [ M ] · f(α)= = K(α)= − αn T π (2 n)!! 0 nX=0 α 9α2 25α3 =1+ + + ... (5) A naive energy argument improves the estimation. 4 64 256 The maximum potential energy is 2mga. Assume Numerous approximation schemes (cf.8, Table III) that this energy converts entirely to kinetic energy as aim to simplify the description of a pendulum’s an- the mass m moves at constant velocity 2√ga through harmonicity, as measured by coefficients to the pow- a distance 8b, then T0 4b/√ga. In the small angle ≈ ers of α. At small α all formulae for T (α) must approximation, a l, b √la, and T0 4 l/g, asymptotically approach Eq.5, so exact and approx- ≪ ≈ ≈ certainly an underestimate. p imate agreement to (α) and (α2) respectively is a The exact period follows from a more sophisti- common feature ofO the manyO published results. For cated calculation, again based on conservation of en- example, the empirical Kidd-Fogg formula9 has ergy. At any half-height z = (l/2)(1 cos(θ)) < a, − 1 α 5α2 15α3 the kinetic energy equals 2mg(a z), the velocity f(α)= =1+ + + ... (6) − 1/4 equals 2 g(a z), and the period equals2,5 (1 α) 4 32 128 − − p Of course, other factors introduce uncertainty to 14 T (a/l) = T dt =4 lθ0 ds (1) physical experiments , and these uncertainties al- 0 0 2√g(a−z) R R ways cause the data to deviate from theoretical ex- =4 l π/2 dξ =4 l K( a ), pectations. Say that we write the standard deviation g 0 √1−(a/l) sin2(ξ) g l σ in units of T , then σ competes in order of mag- q R q 0 nitude with terms from the expansion of f(α) until n where ds goes along the arc of motion. The com- eventually, for some integer n, we have cn α σ. plete elliptic integral of the first kind, K, admits This logic is useful in data analysis and gives≪ some no simple closed-form. Alternatively, the small an- restraint to our exploration of approximate solu- gle approximation eliminates dependence on a and tions. 3
III. PHASE SPACE GEOMETRY need to apply a canonical transformation15
m2 g 1/4 A. Small Angle Approximation q q = q , (10a) → l e l 1/4 e p p = p , (10b) E → m2 g ω E(α) eE(α)= 0 p2 + q2 ,e (10c) → 2
with ω0 = g/l. Transformation Eq. 10 takes ellip- tical trajectoriesp into circular trajectories with equal energy and equal enclosed phase area
λ(α)= dq p(α, q)= dq p(α, q). (11) I I The period e e e ~p ~q m dq T (α)= dt = dq = , (12) I I p(α, q) I ω0 p(α, q) e also remains invariante undere the canonical trans- formation Eq. 10. To see this we use another FIG. 2. Total Energy Surface. Level sets of the total en- definition16 for the period ergy function project trajectories into the plane of phase −1 space, the Conserved Energy Surfaces. dλ dE dq T = = dq = , (13) dE I dp I ω0 p
In terms of the phase space coordinates, (q, p) = with α dependence suppressed. This equation proves (l θ,ml θ˙), the pendulum kinetic and potential en- a connection between the physical period of motion ergy are e e and the purely geometric phase area. As λ(α) and E(α) remain invariant under canonical transforma- T = p2/2m, V = mgl (1 cos(q/l)). (7) tion so too must T (α). − Circular trajectories transform invariantly under The potentiale expands in power series e rotations around the origin of phase space, which im- mediately implies φ¨ = 0. Then the time-dependent ( 1)n+1 solution to the equations of motion is V = mgl − (q/l)2n. (8) (2n)! nX=1 q = Ψ(α, 0) cos ω0(t t0) , (14a) e − − p = Ψ(α, 0) sin ω (t t ), (14b) In the small angle approximation we assume that − 0 − 0 q l throughout the experiment. Keeping only the with angular frequency ω =2 π/T , t an arbitrary first≪ potential term allows us to write the conserved, 0 0 0 e constant. total energy of the pendulum oscillator in the small The small angle approximation does not say any- angle approximation thing about the expansion for T (α). To illustrate the 1 1 mg dangers of approximation, let us work out a clever E(α)=2 mglα ( p2 + q2). (9) ruse. With q l = lθ 2b we have ≈ 2 m l ≪ ⇒ 0 ≈ 2 m g 1 e e 2 2 2 2 3 Clearly there exists a bijection between energies Ψ(α, 0)e = ( l ) (2b) =4 β m gl (15) and elliptical trajectories, depicted as a projection =4 α (1 α) m glp3, − in Fig. 2. Define radius Ψ and angle φ the po- p where β2 = (b/l)2 = α (1 α). The constant ra- lar coordinates of phase space. Each closed curve − Ψ(α, φ) Ψ(α) is alternatively a phase space tra- dius Ψ(α, 0) determines the phase area bounded by jectory →or a Conserved Energy Surface (C.E.S.). closed-curve Ψ(α), again assumed circular, It is much easier to determine time dynamics in a 2 π Ψ(α,0) system of measurement where the phase space tra- λ(α)= p(α, q) dq = 0 dφ 0 r dr (16) jectory takes the particular form of a circle, so we H = π Ψ(α, 0)R2 = λ (Rα α2), 0 − 4
1.20T to 0 ω ω ǫ E = 0 p2 +q2 + 0 n q2(n+1), (18) 2 (2(n + 1))! λn nX=1 π 1.15T 0 We make an ansatz of the form N → ∞ N = 3 Ψ(α, φ)= 2 λ α 1+ αn ψ (φ) , (19) N = 2 π n N = 1 p Xn Clever Ruse λ0 with λπ = 2 π . Our strategy is to substitute Ψ(α, φ) into the en- ergy equation, and determine the functions ψn(φ) in T terms of the expansion coefficients ǫn by forcing the 0 energy to equal λ ω α + (αN+2) for some integer θ π 0 O 0 N 0. As the C.E.S. more nearly obeys conserva- π/4 π/2 tion≥ of energy, the approximation improves.
FIG. 3. Comparison of Period Approximations. The small angle approximation doesn’t prevent wrong, diver- 1. N = 1, The O(α) Approximation gent predictions. Labels for convergent approximations follow naming convention of section IIIB. A first approximation only requires the first term of each sum in Eqs. 18 & 19. Applying (q,p) (Ψ(α, φ)cos(φ), Ψ(α, φ) sin(φ)) to the energy equa-→ tion and collecting terms by order, we have where λ0 = 2 mglT0 . We have yet to determine T (α), and do not assume that T (α) = T . Instead 0 α : λπω0α, (20a) we calculate T (α) by the beautiful formula Eq. 13. 2 2 ǫ1 4 Our phase space geometry consists of a triple α : λπω0α cos(φ) +2 ψ1(φ) . (20b) 6 E, Ψ, λ . In the small angle approximation, height { } E(α) is an exact function of α while perimeter Ψ(α) Setting terms at (α2) equal to zero and solving for O and area λ(α) are merely approximations, so we ex- ψ1(φ), we find pect to find inconsistency in the geometry wherever ǫ the assumptions break down, Ψ(α, φ)= 2 λ α 1 1 α cos4(φ)+ (α2) . (21) π − 12 O p 1 dα dλ 1 dλ f(α)= = = (1 2 α). (17) As in section III.A. the phase space geometry deter- T0 dE dα λ0 dα − mines approximate quantities
Comparing with Eq. 5, we see that T (α) in the 2 π ψ(α,φ) small angle approximation may give the wrong (α) λ(α)= p(α, q) dq = dφ r dr asymptote as depicted in Fig. 3. Worse, the smallO I Z0 Z0 2 angle approximation allows us to predict incorrectly ǫ1 α = λ ( α )+ (α3), (22a) that the period decreases with increasing total en- 0 − 16 O ǫ α ergy! f(α)=1 1 + (α2). (22b) − 8 O
The pendulum has ǫ1 = 2, which makes Eq. 22b B. Simple Anharmonic Approximation asymptotic with Eq. 5 to− (α). O The reductio ad absurdum of section III.A clearly states the need to find a better approximation of 2. N = 2, The O(α2) Approximation the exact phase space geometry. To present results in a more general fashion, we treat the pendulum as The approximation improves if we include sum- an anharmonic oscillator with a potential V . The mands for n = 1 and n = 2. Evaluation of the potential expands in power series around a position 2 energy yields the same constraints as above and ∂V ∂ V of stable equilibrium, i.e., q=0 =0, 2 q=0 > 0. ∂q | ∂q | α3 : λ ω α3 1 ǫ cos6(φ)+2 ψ (φ) (23) Imposing the symmetry constraint V (q) = V ( q), π 0 90 2 2 − 2 4 2 the most general form for the total energy reduces + 3 ǫ1 cos (φ) ψ1(φ)+ ψ1(φ) . 5
Setting (α3) terms equal to zero, substituting the C. Time Dependence O determined form of ψ1(φ), and solving for ψ2(φ) de- termines The phase space trajectory determines time evo-
ǫ1 4 lution Ψ(α, φ) = √2 λπ α 1 12 α cos (φ) (24) 2 − ′ 7 ǫ dq dφ Ψ (α, φ) + 1 α2 cos8(φ) ǫ2 α2 cos6(φ)+ (α3) . dt = = cot(φ) 1 , (27) 288 − 180 O ω0 p ω0 Ψ(α, φ) − The estimation of f(α) slightly improves, where prime indicates differentiation with respect to 2 φ. Again expand in powers of α ǫ1 α λ(α)= λ0 (α (25a) Ψ′(α, φ) − 16 = α ψ′ (φ)+ α2 ψ′ (φ) (28) 35 ǫ2 α3 ǫ α3 Ψ(α, φ) 1 2 − 1 2 4 + )+ (α ) ′ 3 2304 − 288 O ψ1(φ) ψ (φ) + (α ), ǫ α 1 O f(α)=1 1 (25b) − 8 Between two near points in phase space 2 2 2 35 ǫ1 α ǫ2 α 3 dφ 1 4 + + (α ). dt 1+ cos (φ) ǫ1 α (29) 768 − 96 O ≈ ω0 − 3 + ( 1 cos6 (φ) ǫ 1 cos8(φ) ǫ2 ) α2 , 30 2 − 6 1 Inserting pendulum values (ǫ1,ǫ2) = ( 2, 4) makes Eq. 25b asymptotic with Eq. 5 to (α−2). where we drop terms higher than (α2). Expanding O cosine terms (Cf. A273496) allowsO direct integration of dt; however, results at high order are not easy to 3. The O(αN ) Approximation express in concise form. To first order
t = t1 dt = φ1 (1 ǫ1 α ) (30) By iterating the procedure applied for N = 1 and 1 0 − ω0 − 8 N = 2, we obtain an approximation to arbitrary or- + α ǫ1 1 sin(2R φ )+ 1 sin(4φ ) + (α2), ω0 12 1 96 1 O der. Every ψn(φ) can be expanded in Fourier series or in even powers of cosine. For smooth potentials with limits φ(0) = 0 and φ(t1) = φ1. By Lagrange 19,20 with a single minimum, the approximation converges inversion we could in principle obtain φ1(t1) . according to Alternatively, we have
N+2 dφ 2 d lim E = lim λπω0α+ (α )= λπω0α. (26) = φ˙(φ) = cos (φ) tan(φ) . (31) N→∞ N→∞ O dt dt If the ǫ coefficients grow rapidly or contain a diver- Using the equations of motion and substituting an d 2 gence, then more detailed analysis is required. approximation for dt tan(φ) = (qp˙ qp˙ )/q , we ob- A simple symbolic computation calculates higher tain expressions for the phase space− angular velocity, order expansions by routine. Taking the pendulum such as n as an example with ǫn = ( 2) , we write a simple − dφ 1 4 code, and store expansion coefficients in the Online ω 1+ cos (φ) ǫ1 α (32) Encyclopedia of Integer Sequences17 (Cf. A273506, dt ≈− 3 A273507, A274130, A274131, A274076, A274078). 1 1 + cos6(φ) ǫ cos8(φ) ǫ2 α2 . Relaxing the condition V (q) = V ( q), we also cal- 30 2 − 18 1 culate various expansions for a potential− where the ǫ variables take on arbitrary values (Cf. A276738, The phase velocity φ˙ depends on the phase angle φ, A276814, A276815, A276816). as expected in any oscillation where the phase space Mathematica algorithms available via OEIS en- trajectory deforms away from elliptical or circular tries A273506 and A276816 give two different ways shape. Either limit (ǫ1,ǫ2) (0, 0) or α 0 re- → → to compute arbitrary precision expansions of phase covers constant φ˙, when again trajectories become space trajectories and K(α)18. Whenever N < 10, ellipses or circles. these algorithms operate in small time on a personal There are numerous well known methods for cal- computer. For moderate ranges of α, enumeration culating numerical time evolution of a Hamiltonian beyond N = 3 follows a law of diminishing returns. system including symplectic integration21. In the As can be seen in Fig. 3, the (α3) approxima- present case, time evolution occurs along the one- tion already captures to withinO 1%, the exact be- dimensional C.E.S. The standard Runge-Kutta al- havior of the pendulum in the range α [0, 1/2], gorithm applies; though, the task of integration re- θ [ π/2,π/2], where our experiment takes∈ place. quires only minimal complexity. The simple Euler’s 0 ∈ − 6
method (RK1) suffices. Iteration through time ac- The exact pendulum phase space trajectory is cording to φ˙ yields time-dependent predictions, as depicted in Fig. 4. This is the first plot to clearly q(α, ϑ)= 2 λπ arcsin(√α sn(ϑ, α)), (33a) p show anharmonicity as anisochronous motion of pen- p(α, ϑ)= 2 λπ α cn(ϑ, α), (33b) dulums with different initial conditions. p 2 2 Around α = 0 all approximations become indis- Ψ(α, ϑ)= q(α, ϑ) + p(α, ϑ) , (33c) p tinguishable. To α = 1/2, the (α) approximation where cn, sn are Jacobian elliptic functions of angu- closely matches the exact numericalO solution, which lar coordinate ϑ = K(α)+ ω0 t with period 4K(α). is nearly indistinguishable from the (α3) approx- O Substituting in time dependence such as Eq. 30, imation. Isochrony becomes more pronounced at it is possible to expand Ψ(α, ϑ) in powers of α and high α where, after 8 intervals of ∆t = T0/8, the prove, order-by-order, equivalence between the exact pendulum does not nearly reach its initial condition. solution and the approximate solution of III.B. We need not perform this tedious calculation, for any p solution that conserves energy must be equivalent to the exact solution. Rather, let us explore conver- gence by plotting the approximations of sn and cn near the divergence α = 1. Setting the left hand side of Eqs. 33a-b equal to an (αN ) approximation allows us to solve for an approximationO of both sn and cn18. Composing ap- proximate trajectories Ψ(α, φ) and time dependence q t(φ) gives parametric function graphs appropriately scaled for comparison, as in Fig. 5.
sn(θ,0.9) 1
K 0 3K θ
-1 cn(θ,0.9) N = 1 2 ∞ 1 FIG. 4. Time Evolution of Pendulums in Phase Space. The N = 10 ( black ), N = 1 ( gray ), and circu- lar ( dashed gray ) trajectories are plotted for α = K 2K 5K {0.01, 0.1, 0.2, 0.3, 0.4, 0.5}. The N = 10 trajectory com- 0 θ pletely overlaps the N = 3 trajectory for all α values, showing convergence. Small filed circles mark initial con- ditions, while open circles indicate the state of a system -1 at intervals of ∆t = T0/8 as it rotates clock-wise through phase space. Large open circles are calculated by the technique of symplectic integration. FIG. 5. Approximation Around the Divergence. Ap- proximations of sn and cn for N ∈ 1, 2, 3,... 10 are shown to approach the exact functions, even at α = 0.9. Vertical lines mark the end of one complete period of the N th approximation. IV. COMPARISON WITH STANDARDS Once we extend the approximation to functions A. Jacobian Elliptic Functions such as cn and sn, it becomes possible to treat other classical motions. We could solve Euler’s equations The Jacobian elliptic functions determine exactly for the free rotational motion of a rigid body2, or the phase space geometry of the simple pendulum. describe a photon orbit around a Kerr black hole23. The properties of these functions are well known and Using the inversion relations5,7, we could approxi- recorded in a number of standard resources13,20,22. mate rotational motion (α > 1) of a plane pendu- Paul Erd¨os gives a creative, geometric introduction lum. We follow Erd¨os7 by plotting a couple of Seif- via the Seiffert spirals7. fert spirals24, as in Fig. 6. 7
V. EXPERIMENTAL VERIFICATION
The experimental setup, procedure, and analysis for determining the period of a plane pendulum are among the most simple and ubiquitous in the physics classroom. A majority of physics students have com- pleted the basic experiment, while relatively few go on to measure amplitude dependence of the period. As is usual in measurement of small perturbations, more stringent precision goals require more sophisti- cated technology. To make fine measurements of the pendulum’s motion, we need to implement a system α=0.060584 α=0.628415 with digital data acquisition.
FIG. 6. Approximating Seiffert Spirals. For small α, N = 1 approximations closely follow exact Seiffert spi- A. Setup and Data Processing rals (left). Approximate trajectories (gray) for N = 1, 2, 3,... 6 approach the exact spiral when α ≈ 0.628 Experimental systems are available at cost from (right). The N = 7 approximation (black) appears nearly indistinguishable from the exact spiral. manufacturers of scientific classroom equipment, but a USB mouse device25,26 provides a thrifty DIY so- lution, our preference. After modifying the mouse into a digital pendulum, we connect it to a Linux 27 Comparing Figs. 4, 5, and 6 gives an idea of lim- work station running the X window system . The xdotool itations encountered when truncating an arbitrary utility program measures the cursor loca- precision result. In a ”small angle”, even a simple tion directly from the computer’s desktop environ- bash approximation works well. As α becomes large, more ment. Integrating with a script obtains data terms in the expansion need to be computed. Slow at nearly millisecond resolution while the pendulum convergence motivates nome expansions4, useful to goes through damping from maximum amplitude to know of, but unnecessary in the present context. stop, as in Fig. 7. The oscillation decays through time, only by a small amount per period. Partitioning at zero- amplitude intercepts obtains a division of the ampli- B. Alternative Approaches tude vs. time data into a number of non-overlapping, nearly-sinusoidal samples of one-period duration. 1. Canonical Perturbation Theory Averaging maximum and minimum amplitude, we then associate one-to-one a set of periods and a set of The Hamiltonian formulation of mechanics also al- amplitudes. Converting amplitude to energy yields lows one to obtain an arbitrary-precision expansion a set of period vs. energy values. Repeating this for the phase space trajectory by applying a suc- process for 100 separate trials, we obtain a dense cession of canonical transformations12. The method sample, as in Fig. 7. above is similar in spirit but with a gentle learning The bulk data obviously shows significant noise, curve. but the sheer number of data points, more than 2000, enables noise reduction by a procedure of bin- ning and averaging. We set meta-analysis parame- 2. Mistaken Expansions ters for bin width ∆α and a minimum energy cutoff αmin to obtain a more manageable data set, with no apparent noise problem. A great many authors2,10,11 recommend solv- ing anharmonic oscillations by some variant of the Lindstedt-Poincar´emethod. This method is error- prone, and usually a wrong assumption is made B. Recursive Data Analysis regarding time dependence (cf. Eq. 30), lead- 2 ing to something like ψ1(φ) cos (φ) rather than Of course we are not the first to obtain digital pen- 4 ∝ ψ1(φ) cos (φ). Taking the wrong ψ1(φ), it is still dulum data, or even the first to analyze amplitude possible∝ to compute the correct term of the f(α) se- dependence28–30. We fit K(α) using the period as a ries expansion, so the mistake often escapes notice. free parameter and observe good agreement over the 8
data range, as in previous investigations. To deter- mine just how well K(α) describes the data requires TABLE II. Cubic Best Fit Parameters. a novel analysis. {αmin, ∆α, αLQ, αQC } = {0.003, 0.013, 0.08, 0.22} It should be possible to extract expansion coeffi- Expectation Estimate Error 1 cients by fitting a cubic function to the data, but im- 4 ≈ 0.2500 0.2463 ± 0.0071 0.52σ , 1.5% mediately we encounter a covariance problem. The 9 64 ≈ 0.1406 0.1508 ± 0.0082 1.24σ , 7.2% slope of the data does not change sign, remains 25 ≈ 0.0977 0.1037 ± 0.0126 0.48σ , 6.2% nearly flat. One pass analysis yields inaccurate pa- 256 rameter estimation, a wide range of plausible fits. To improve accuracy and precision we take advan- tage of the data’s asymptotic nature by partitioning Choosing other vales for α , ∆α, α , α the entire set into three simply connected, disjoint min LQ QC we obtain similar best fit parameters,{ especially} subsets. This introduces two additional analysis pri- when heuristics are nearly obeyed. To facilitate com- ors, energy values, α and α , which demarcate LQ QC parison of various analyses, archival data and basic boundaries as in Fig. 7. The fit procedure first tools are available online31,32. determines the period and linear expansion coeffi- cient from linear data, subsequently determines the quadratic expansion coefficient from the union of lin- 0.4π ear and quadratic data, finally determines the cubic expansion coefficient from all data. 0.2π In total, the analysis depends on four meta- 20 analysis parameters: αmin, ∆α, αLQ, αQC . To set 0 these values we adopt{ the following heuristics:} time (s) Use as much data as possible. • amplitude (rad) Exclude noisy data around α = 0. -0.4π • Make the bin width as small as possible. • 1.10T Capture at least 10 data points per bin. 0 •
In the linear range [0, α ]: 1.05T 0 • LQ 2 1 2 T α K(α) (1 + α) < 0.001 K(α). 0 π − 4 × π 0.1 0.2 0.3
In the quadratic range [0, α ]: • QC FIG. 7. Measuring K(α). Above: Sample amplitude 2 1 9 2 K(α) (1 + α + α2) < 0.001 K(α). vs. time data. Below: Period vs. energy data points in π − 4 64 × π gray are binned and averaged into the black points. Se- quential linear and quadratic fits are depicted as dashed The extracted linear, quadratic, cubic coeffi- lines, while final cubic fit is a solid line going through • cients should have increasing uncertainty. all points. Dashed vertical lines mark the boundaries between data subsets. Minimize uncertainty where possible. • From these we have initial values αmin, ∆α, αLQ, αQC = 0.003, 0.013, 0.083, 0.21 . {Searching around, not} too{ far, we find the best fit} of Table II. Comparison of extracted parameters VI. QUANTUM CLASSICAL ANALOGY with coefficients of Eq. 6 leads to the humorous conclusion that the Kidd-Fogg formula—though false de facto—also adequately fits the data. That The choice of nomenclature in section III suggests is, the cubic fit does not distinguish between a quantum/classical analogy at work. The symbol competing Eqs.5-6. To decide against Kidd-Fogg by Ψ connotes a quantum wavefunction, but above it data alone requires an experiment with sufficient denotes a C.E.S. Our use of Ψ follows other semi- quality up to and beyond α = 0.6, i.e. 5 10 classical works15,33. In the sequel, we extend the expansion coefficients of K(α). − analogy to time-independent perturbation theory. 9
A. Conservation of Energy To find an analogous equation in quantum dynam- ics, we apply an infinitesimal time-translation by ex- Whenever we use approximate methods in the panding the Hamiltonian propagator i∆t analysis of physical systems, classical or quantum, − ~ H ψn,N (t + ∆t) = e ψn,N (t) (41) we also introduce terms of error at some level of | i | i (1 i∆t H) ψ (t) precision. For example, approximation of a pendu- ≈ − ~ | n,N i lum’s motion may only conserve total energy up to (1 i∆t N ǫi E ) ψ (t) + (ǫN+1). ≈ − ~ i=0 n,i | n,N i O some power of α. A similar situation often arises in P quantum mechanics. This equation shows that time-evolution acts on the We assume a Hamiltonian H = H +ǫ V for which approximate eigenfunctions as a change of complex 0 phase, but only to (ǫN+1). Complex phases cancel the eigenstates ψn are approximately known and O non-degenerate,| i in expectation products, so Eq. 41 implies no worse convergence than H ψn = En ψn (34a) d H | i | i h i =0+ (ǫN+1). (42) H0 ψn,0 = En,0 ψn,0 . (34b) dt O | i | i As ever, the analogy involves an obvious fallacy: The standard perturbation theory34 determines cor- α and ǫ are dimensionless quantities belonging to rections to the zero-order wavefunctions and ener- two separate classes. In the classical theory, we sup- gies. press dependence on the ǫ coefficients and implicitly To first order, the time-independent Schr¨odinger assume that a convergence criteria can always be equation becomes stated as a maximum value for α given an approx- 2 imation and a precision goal. The quantum theory H ψn,1 = (En,0 + ǫ En,1) ψn,1 + (ǫ ). (35) | i | i O requires quantization of α. After more exploration We make the ansatz and explicit calculation, we hope to gain a better understanding of the semiclassical analogy’s inner ψ = ψ + ǫc1 ψ , (36) | n,1i | n,0i n,i | i,0i workings. Xi=6 n and solve for B. Quantum Anharmonic Oscillator ψ V ψ c1 = h i,0| | n,0i, E = ψ V ψ . (37) n,i E E n,1 h n,0| | n,0i n,0 − i,0 As with classical perturbation theory, quantum 1 perturbation theory allows iteration to arbitrary 35 N , where the approximate eigenfunctions 2σ 4σ
N ψ = ψ + ǫj cj ψ , (38) | n,N i | n,0i n,i | i,0i Xj=1 Xi=6 n are nearly stationary with regard to energy
N FIG. 8. Perturbed oscillator wavefunction. From light i N+1 gray to black, the approximate wavefunctions for N = H ψn,N = ǫ En,i ψn,N + (ǫ ). (39) | i | i O 0, 1, 2 with parameter values ǫ1 = −2 and h/λ0 = 1. Xi=0 Though again a law of diminishing returns applies to higher order corrections. We consider the quantum anharmonic oscillator, The semiclassical analogy associates conservation with Hamiltonian of energy with the eigenvalue equation. In either ω ω ǫ H = 0 (p2 + q2)+ 0 1 q4. (43) theory, iteration of a recursive algorithm changes the 2 24 λπ shape of a C.E.S. or a wavefunction such that the perturbed solution becomes increasingly precise ( Cf. Using the technique of ladder operators, it is rela- tively easy to solve for energy to (ǫ), Fig. 8 ). O As time evolves the N th classical approximation 1 E = (2n + 1)~ω , (44a) satisfies n,0 2 0 dE 1 2 ~ =0+ (αN+2). (40) En,1 = (2 n +2 n + 1) ω0, (44b) dt O 32 10 where ~ = h/(2π) is the reduced Planck’s constant writing out a list of prior beliefs, we define an anal- h and ǫ = ǫ1 λ0 . ysis where the extracted parameters closely match Setting equal quantum and classical energies, we the expected values. Nothing precludes our analysis see that from applying to higher energy motions in the do- main α [0, 1]. It would be interesting to see how E 0+ǫ E 1 α = n, n, + (ǫ2) (45) many expansion∈ coefficients this method may accu- λπ ω0 O = 1 h (2 n + 1) rately and precisely determine. As many as five, 2 λ0 ten? + ǫ1 ( h )2(2 n2 +2 n +1)+ (ǫ2), Perturbation methods extend beyond the impor- 32 λ0 O tant but simple example of a plane pendulum. Ex- or equivalently, by Eq. 22a, tending the ansatz Eq. 19 to include half-powers of α allows arbitrary precision solution of any one- 1 ǫ h2 λ(n) = (n + )h + 1 + (ǫ2). (46) dimensional, power-series potential. This important 2 64 λ0 O elaboration leads to applications in mathematical biology38,39, classical astronomy40, and relativistic The quantum conditions Eqs. 44 & 45 recall astronomy41. In higher dimensional phase space, we the Sommerfeld-Wilson prescription of old quantum obtain angular equations of motion along a variety 36,37 mechanics of multidimensional C.E.S. Subsequent work will ex- plore the multidimensional generalization. λ(n)= p dq = (n + δ) h, (47) The classical/quantum analogy reveals fundamen- I tal principles that apply throughout physics. Time- with Maslov index independent perturbation theories subject phase space trajectories and wavefunctions to perturbative 1 ǫ1 h variations. Evolving through time, trajectories and δ = + , (48) wavefunctions meet the expectation that higher pre- 2 64 λ0 cision approximations more nearly conserve energy. slightly perturbed from δ = 1/2, the usual value Exploration of quantum conditions resolves a fallacy associated with harmonic vibrational motion. in the analogy by showing that quantum theory re- In the case of a ”quantum pendulum”, ǫ1 can not places continuous energy parameter α with a quan- be made small, so fidelity of approximate methods tum number n. We have yet to find any connections 42 depends entirely on the constant λ0. As l and m to the Matheiu functions , but speculate of their become increasingly small, λ0 approaches h. When- existence. ever h/λ0 < 1/10 and n < 5 then α < 1/2, a Ultimately we reach a detailed understanding of first or second approximation adequately describes the plane pendulum and its relation to time. The the quantized wavefunctions. The N = 0, 1, 2 ap- pendulum is a particular anharmonic oscillator, with proximations of the quantum anharmonic oscilla- a period that varies slightly as a function of am- tor’s groundstate wavefunction appear in Fig. 8, plitude. On the quantum scale atomic oscillations, with extreme expansion parameter h/λ0 = 1. for example in Cesium-133, provide the highest pre- cision time scales. On the astronomical scale, we also measure long times, even for the practical pur- VII. CONCLUSION pose of keeping a calendar. Seconds are useful units for time, but physics also needs nanoseconds and years. Wherever we find an oscillation, usually an- The pendulum takes an eminent place in the harmonicity is not too far behind, leading to unex- physics canon, not only as a measurement device but pected behaviors such as precession. also as an example of anharmonic oscillation. The simple harmonic approximation leaves open the pos- sibility of spectacular failure because it only reliably ACKNOWLEDGMENTS determines the overall scale of time. Apprehension of the dependence on θ0 or α requires more careful The author gratefully acknowledges dissertation and detailed analysis. Here we present a novel al- committee members, William Harter, Salvador gorithm, which solves equations of motion and pro- Barraza-Lopez, and Daniel Kennefick for helpful dis- duces the expansion coefficients of K(α). Calcula- cussions, comments on earlier drafts of the paper; tions require minimal technical skill and avoid any also, Wolfdieter Lang and many other volunteer edi- confusing artifice. tors for their work on the OEIS. This work was sup- Thrift experiment produces good enough data. ported in part by a Doctoral Fellowship awarded by Taking into account theoretical expectations by the University of Arkansas. 11
† [email protected], [email protected] 22 M. Abramowitz and I.A. Stegun. Handbook of math- 1 P. Ariotti. Galileo on the isochrony of the pendulum. ematical functions, volume 55. Courier Corporation, Isis, 59(4):414–426, 1968. 1964. 2 L.D. Landau and E.M. Lifshitz. Mechanics. Pergamon 23 L.C. Stein. Private communication, 2016. Press, 1976. 24 A. Seiffert. Uber¨ eine neue geometrische Einf¨uhrung 3 W.G. Harter. Classical mechanics with a BANG!, in die Theorie der elliptischen Funktionen. Wis- 2016. URL https://goo.gl/CCN11y. senschaftliche Beilage zum Jahresbericht der 4 C. G. J. Jacobi. Fundamenta Nova Theoriae Func- St¨adtischen Realschule zu Charlottenburg, Os- tionum Ellipticarum. K¨onigsberg, 1829. tern, 1896. 5 A.J. Brizard. A primer on elliptic functions with ap- 25 V. Gintautas and A. H¨ubler. A simple, low-cost, plications in classical mechanics. European Journal of data-logging pendulum built from a computer mouse. Physics, 30(4):729, 2009. Physics Education, 44(5):488, 2009. 6 T.E. Baker and A. Bill. Jacobi elliptic functions and 26 R.D. Peters and S.C. Lee. Pendulum sensor using an the complete solution to the bead on the hoop prob- optical mouse. arXiv:0904.3070, 2009. lem. American Journal of Physics, 80(6):506–514, 27 Arch Linux Collaboration. Xorg - ArchWiki. URL 2012. https://wiki.archlinux.org/index.php/xorg. 7 P. Erd¨os. Spiraling the earth with C.G.J. Jacobi. 28 T. Lewowski and K. Wozniak. The period of a pen- American Journal of Physics, 68(10):888–895, 2000. dulum at large amplitudes: a laboratory experiment. 8 C.G. Carvalhaes and P. Suppes. Approximations European journal of physics, 23(5):461, 2002. for the period of the simple pendulum based on 29 R.C. Nicklin and J.B. Rafert. The digital pendulum. the arithmetic-geometric mean. American Journal of American Journal of Physics, 52(7):632–639, 1984. Physics, 76(12):1150–1154, 2008. 30 A.E. Legarreta S. Gil and D.E. Di Gregorio. Mea- 9 R.B. Kidd and S.L. Fogg. A simple formula for the suring anharmonicity in a large amplitude pendulum. large-angle pendulum period. The Physics Teacher, American Journal of Physics, 76(9):843–847, 2008. 40(2):81–83, 2002. 31 B. Klee. Github - digital pendulum data analysis, 10 L.P. Fulcher and B.F. Davis. Theoretical and experi- 2016. URL https://goo.gl/juvySd. mental study of the motion of the simple pendulum. 32 B. Klee. AUR(en) - pendulumdata, 2016. URL American Journal of Physics, 44(1):51–55, 1976. https://goo.gl/m1JEMQ. 11 D. Park. Classical Dynamics and Its Quantum Ana- 33 W.G. Harter. Su(2) coordinate geometry for semi- logues. Springer, 2012. classical theory of rotors and oscillators. The Journal 12 J.H. Lowenstein. Essentials of Hamiltonian Dynam- of chemical physics, 85(10):5560–5574, 1986. ics. Cambridge University Press, 2012. 34 P.J.E. Peebles. Quantum Mechanics. Princeton Uni- 13 Wolfram Research. Wolfram functions: Elliptic func- versity Press, 1992. tions, 2016. URL https://goo.gl/bzBNqj. 35 V.S. Mathur and S. Singh. Concepts in quantum me- 14 R.A. Nelson and M.G. Olsson. The pendulum-rich chanics. CRC Press, 2008. physics from a simple system. American Journal of 36 S. Tomonaga. Quantum Mechanics: I, Old quantum Physics, 54(2):112–121, 1986. theory. North-Holland, 1968. 15 J.P. Ralston. Berry’s phase and the symplectic char- 37 M.S. Child. Semiclassical mechanics with molecular acter of quantum time evolution. Physical Review A, applications. Clarendon Press, Oxford, 2014. 40(9):4872, 1989. 38 A.J. Lotka. Analytical note on certain rhythmic rela- 16 V.I. Arnold. Mathematical Methods of Classical Me- tions in organic systems. Proceedings of the National chanics. Graduate Texts in Mathematics. Springer Academy of Sciences, 6(7):410–415, 1920. New York, 2013. 39 V.I. Arnold and R. Cooke. Ordinary Differential 17 The OEIS Collaboration. The on-line encyclopedia of Equations. Springer Textbook. Springer Berlin Hei- integer sequences, 2016. URL http://oeis.org. delberg, 1992. ISBN 9783540548133. 18 B. Klee. Approximating the jacobian elliptic func- 40 B. Klee. Estimating planetary perihelion precession, tions, 2016. URL https://goo.gl/o7Cg7x. 2016. URL https://goo.gl/wCrwx0. 19 W. Lang. Private communication, 2016. 41 B. Klee. Exact and approximate relativistic correc- 20 E.T. Whittaker and G.N. Watson. Modern analysis. tions to the orbital precession of mercury, 2016. URL Cambridge University Press, 1927. https://goo.gl/Y2k7yX. 21 E. Forest and R.D. Ruth. Fourth-order symplectic 42 M. Leibscher and B. Schmidt. Quantum dynamics of integration. Physica D: Nonlinear Phenomena, 43(1): a plane pendulum. Physical Review A, 80(1):012510, 105–117, 1990. 2009.