The last lesson ended with a matrix pair, a stiffness matrix and a mass matrix , and the eigenproblem . That was a method for approximating a PDE. Here the same objects appear without any approximation: a handful of real masses on real springs. 5.3 Mechanical Vibrations: Spring–Mass–Damper solved one mass; this lesson solves many, and the answer turns out to be one mass many times over, in the right coordinates. Those coordinates are the eigenvectors of 1.6 Change of Basis & Diagonalization, called normal modes. Let the number of masses grow and the chain becomes a string obeying the wave equation of 9.4 The Wave Equation & Vibrating Strings, with an effect a finite-difference wave solver cannot avoid.
Transcript
- Two masses sit between two walls. Outer springs of stiffness k, and a weak coupling spring, k c, joins the masses. Each mass feels a force from each neighbor equal to the spring's stiffness times its stretch. Collect the two positions in one vector x, and the equations become M times x double dot equals minus K times x. The mass matrix M is diagonal. The stiffness matrix K has the total stiffness at each mass on its diagonal and minus the joining stiffness off it, so K is symmetric, and the walls make it positive definite. It is the same pair of matrices that finite elements gave us, now with real springs.
- Ask for motions where every mass oscillates at the same frequency, with a fixed shape u. Substituting gives K u equals omega squared M u, a generalized eigenproblem. For two masses, take m and k equal to 1 and k c equal to 0.05. The first shape is 1, 1: the masses move together, the coupling spring is never stretched, and the frequency is 1.0000. The second is 1, minus 1: the masses move oppositely, the coupling spring is stretched twice as much, and the frequency is higher, omega squared equals 1 plus 2 k c, which is 1.1, so omega is 1.0488. Any motion is a sum of these two modes, and each mode is a single independent oscillator.
- Pull mass 1 aside and let go. That start is half of each mode, so mass 1 moves as cosine of the mean frequency times a slow envelope, and mass 2 as sine of the mean frequency times the complementary envelope. These are beats. The two frequencies differ by 0.0488, so the slow envelope takes pi over delta, 64.37 time units, to hand the whole motion from mass 1 to mass 2, against a single oscillation period of 2 pi. Then it flows back. The weaker the coupling, the slower the exchange. Notice that the energy of each mode never changes. What moves is energy between the masses.
- Now let N equal masses sit on equal springs between fixed walls. K is k times the tridiagonal matrix minus 1, 2, minus 1: the very matrix of the finite difference second derivative. A sine is an eigenvector, because sine of i plus 1 times theta plus sine of i minus 1 times theta equals 2 cosine theta times sine of i theta, and the wall at the far end fixes theta equal to j pi over N plus 1. So the frequencies are 2 root k over m times sine of j pi over 2 times N plus 1. Mode 1 swings all masses one way, and the top mode alternates. For N equal to 5 the frequencies are 0.5176, 1.0000, 1.4142, 1.7321 and 1.9319. Splitting a motion into modes is the discrete sine transform.
- Let the number of masses grow, with spacing a, and the correction to the wave equation vanishes. A plane wave solves the chain exactly when omega equals 2 c over a times sine of kappa a over 2, below the straight line of the string, omega equals c kappa. For N equal to 24 the slowest mode is 0.12558 against 0.12566, a ratio of 0.9993. Mode 12 is 1.36909 against 1.5080, a ratio of 0.9079, and the top mode 1.99605 against 3.0159, a ratio of 0.6618. The harmonics go out of tune: the chain is not the string. The slope of the curve is the group speed, c cosine of kappa a over 2, which is 0.5 c at kappa a equal to 2 pi over 3. A finite-difference wave solver is exactly this chain.
- Add the same damper to every mass and each modal equation stays independent: q double dot plus gamma q dot plus omega squared q equals zero. Each mode is the damped oscillator we already know. For N equal to 24, gamma equal to 0.1, mode 12, the exact energy ratio at t equal to 20 is 0.131, against e to the minus 2, which is 0.135. At t equal to 40 it is 0.0187 against 0.0183. With gamma equal to 0.3, gamma over 2 is 0.15, larger than omega 1, which is 0.126, so the slowest mode no longer oscillates at all.
- The highest mode also sets the time step. A leapfrog step on one mode stays bounded only when h is at most 2 over omega. For the chain of 24 masses that is 2 over 1.99605, which is 1.0020, while the slowest motion has period 2 pi over 0.12558, which is 50. The ratio of fastest to slowest is 15.9, and it grows with N: add masses and the step must shrink. Next, the lesson on electrical circuits takes this equation into electronics. An RLC circuit is the same second-order equation, with inductance in the place of mass, resistance for the damper, and one over C for the spring. I expect coupled loops to give a matrix equation like this one, and so to have normal modes.
Masses, springs and two matrices
Put masses on a line, joined to each other and to two fixed walls by springs, and let each move a small distance sideways or along the line (the equations are identical). Mass feels a force from each neighbor equal to the spring’s stiffness times its stretch. For two masses between two walls, with outer springs and a coupling spring between the masses:
Collect the unknowns in . Then with
The pattern holds for any chain: is diagonal with the masses; is the total stiffness of the springs touching mass ; . So is symmetric. The walls make it positive definite: is twice the stored spring energy. Everything below uses just these two properties. The same and are the finite element matrices of 10.13 Introduction to Finite Elements: there the springs were elements and came from integrating products of hat functions.
Normal modes
A single oscillator has one natural motion, . Ask whether the coupled system has motions where every mass oscillates at the same frequency: with a fixed shape . Substituting gives
This is the generalized eigenproblem, with eigenvalue . For equal masses it is the ordinary eigenproblem of . In general, is diagonal and positive, so exists, and the substitution turns the problem into
is symmetric and positive definite. The spectral theorem (from memory; 1.5 Eigenvectors & Eigenvalues gives the geometry) says a real symmetric matrix has real eigenvalues and orthonormal eigenvectors . Positive eigenvalues mean is real: the system oscillates and never grows. Going back, the mode shapes are , and they satisfy : orthogonal in the mass-weighted sense.
Modal decomposition. Write . Because the diagonalize both matrices, the equations decouple completely:
independent one-mass oscillators. The solution is , with and . So every motion is a sum of normal modes, and the coupling has not disappeared: it has been moved into the choice of coordinates. In the first-order form of 7.2 Linear Systems & the Eigenvalue Method, these are the eigenvalues on the imaginary axis of 7.4 Complex & Repeated Eigenvalues, one pair per mode.
Energy. The total energy becomes in modal coordinates. Each term is conserved separately. Energy never moves from one mode to another. It moves between masses, which is what beats are.
Worked example: two masses by hand
Take and a weak coupling spring . Then has determinant , so , that is
The eigenvectors: for , reads , so . For it gives , so . In the first mode the masses move together and the coupling spring is never stretched, so the frequency is that of one mass on one spring. In the second they move oppositely and the coupling spring is stretched twice as much as either end, so the frequency is higher. So and .
Beats and energy exchange
Pull mass 1 aside by and let go: . Then
with and . This is an oscillation at the mean frequency with a slow envelope on mass 1 and on mass 2: beats. Mass 1 comes to rest and mass 2 takes over the whole motion at ; the energy then flows back. For , , so the transfer takes time units against a single oscillation period of . The weaker the coupling the slower the exchange. For small , , because . At the exact value is against from the approximation.
A chain of N masses
Now equal masses with equal springs and fixed ends. Then and : the same matrix as the finite difference second derivative of 10.1 Finite Differences: Discretizing Derivatives, and the same one whose eigenvectors the multigrid and spectral lessons found. Try a sine, . The identity gives
for every interior , so the sine is an eigenvector as soon as the end conditions hold. The wall at is satisfied automatically; the wall at needs , so . Therefore
Mode 1 swings all masses one way; mode alternates sign between neighbors. Modal decomposition of the initial displacement is a discrete sine transform: the coefficients are the sine transform of , the transform the spectral lesson used to solve Poisson’s equation on a box, and the frequencies are the square roots of the finite-difference Laplacian’s eigenvalues, up to the factor . For and , :
| 1 | 2 | 3 | 4 | 5 | |
|---|---|---|---|---|---|
With the consistent mass matrix of the finite element lesson instead, the same eigenproblem is the finite element one. Lumped beads are finite differences; smeared beads are finite elements.
From chain to string: the wave equation
Let with spacing . Newton’s law is . The Taylor series gives , so
Keep fixed while (more and more lighter masses on stiffer springs: for line density , for tension , so ) and the correction vanishes: the wave equation of 9.4 The Wave Equation & Vibrating Strings. Its fixed-end modes are with , equally spaced, which is why a string has harmonics. See how the chain differs: a plane wave solves the discrete equation exactly when
This is the dispersion relation of the chain. For long waves (), , the wave equation again. For short waves it bends below the line . The phase speed and the group speed are both below , the group speed by more. At the shortest wave the chain can carry, (neighbors alternating), reaches its maximum and the group speed is zero. There is no wave above that frequency at all.
The fixed chain’s modes sit on this curve at . For (sim values with ), against the continuum value , a ratio of ; against , a ratio of ; and the top mode against , a ratio of . The nine percent loss at mode 12 is the harmonic series going out of tune: the chain is not the string, and it is a fact about beads, not an error.
Phonons glimpse. A crystal is a chain of atoms, with the same maximum frequency and zero group speed at the zone edge . Quantum mechanics (from memory, not derived here) makes each mode a harmonic oscillator with energies , and one step is a phonon, a quantum of sound. Sound is the long-wavelength part, where the dispersion curve is straight.
Energy, damping and the highest frequency
Each mode keeps its own energy. For , mode started at peak amplitude has at every instant; the sim shows the ratio to the starting value as throughout. Add a damper of strength on each mass, with unit masses. In modal coordinates , still decoupled. Each mode is the damped oscillator of 5.3 Mechanical Vibrations: Spring–Mass–Damper with : underdamped when , so energy falls as on average, and overdamped for slow modes with . For , , mode (), the exact energy ratio at is against , and at it is against . With , exceeds , so mode no longer oscillates. A general damper couples the modes again; only (Rayleigh damping, from memory) keeps them separate.
Driving the chain at frequency forces each , and resonance appears whenever meets some (5.6 Forced Oscillations & Resonance): masses, resonances.
The lab
The top strip animates the masses. The panels show mode ‘s shape, the frequencies (dots, against the dashed continuum line) with each mode’s energy underneath, and two masses against time. The frequencies come from a numerical eigenvalue solver (Jacobi iteration), and for uniform springs agree with to at . The motion is the exact modal sum.
- With the frequencies are . Step through the modes.
- Set , the middle spring to , and start from “modes and ”. The readout gives and and a transfer time of : the worked example above.
- Set , uniform: , , , ratio fastest to slowest , step limit .
- Keep and set the middle spring to (a nearly cut chain). Mode stays at , since it never stretches the middle spring; mode drops from to . This near-degenerate pair has a beat transfer time of .
- Drag one mass and release: the energy bars show the displacement spread over many modes, each staying constant with no damping.
The exercise: normal modes from and
The playground below shows the beat envelope of the worked example. The exercise asks for the general routine: from masses and springs to frequencies and mode shapes.
Concept checks
Problems
Mass and mass hang in a row from a wall: wall, spring , mass , spring , mass , with the right end free. Find the frequencies and mode shapes, and check that the shapes are orthogonal with respect to .
Show worked solution
and . Then , so and , and and . The second row gives : for , (both move the same way, the lighter one more); for , . Check: .
For , , find the frequencies and the three mode shapes, then write for the start at rest. Check that the energy equals .
Show worked solution
gives , so . The normalized shapes are , , . The amplitudes are , so , which is at . The energy is , which equals .
Two identical oscillators with are coupled by . Mass 1 is released at with mass 2 at rest. After how long is mass 1 first (nearly) at rest with the motion on mass 2? Compare with the estimate .
Show worked solution
and , so and the transfer time is . The estimate gives , within .
Starting from , find the wavelength (in units of ) at which the phase speed is of , using , and give the group speed there.
Show worked solution
With , the phase speed ratio is , so and (the exact solution of is ), and (exactly ). The group speed is either way, so a packet at that wavelength travels at about of : the energy lags the crests.
What comes next
The next lesson, 14.2 Electrical Circuits, takes this equation into electronics. An RLC circuit obeys , the same second-order equation as a damped, driven mass on a spring, with inductance in the place of mass, resistance in the place of the damper and in the place of the spring. The syllabus lists that analogy with impedance, filters and multi-loop circuits as systems. I expect, from memory, that coupled loops give a matrix equation like this lesson’s, with inductance and matrices, and so have normal modes; that is for the next lesson to check.
Key Takeaways
- Masses on springs give , diagonal, symmetric positive definite. Trying gives ; symmetrizing shows the are real and positive and the shapes -orthogonal.
- Modal coordinates give independent oscillators ; every motion is a sum of normal modes. Each mode’s energy is conserved; energy moves between masses, not between modes.
- Two equal masses with coupling have modes at and at . With the frequencies are and , and the beat transfer time is .
- A fixed-end chain of equal masses has and sine shapes. It is the finite-difference Laplacian, modal decomposition is the discrete sine transform, and gives .
- The continuum limit is with . The chain’s dispersion relation gives phase speed and group speed , both below , and a maximum frequency with zero group speed.
- A finite-difference wave solver is this chain: numerical dispersion is the dispersion of beads. At mode runs at of its continuum frequency, mode at .
- Uniform damping keeps modes decoupled; slow modes with are overdamped. The highest frequency sets the explicit step limit ( at ), and the stiffness ratio is .
- Next: 14.2 Electrical Circuits, where an RLC circuit is the same second-order equation.