In 6.1 The Laplace Transform we built the table and the derivative rule: . This lesson cashes them in. A simulation engineer meets IVPs with forcing all the time (a spring-mass damper under a push, a control surface under a command), and the method of 5.2 Constant Coefficients & the Characteristic Equation handles them in three stages: solve the homogeneous equation, guess a particular solution, then fix constants from the initial data. Laplace does everything in one stroke. Transform the equation, solve for by algebra, split it into table entries, invert. The initial data and the forcing both sit inside from the start. By the end you will see why 5.6 Forced Oscillations & Resonance had to treat resonance as a special case, and why here it arrives for free as a double pole.
Transcript
- Last lesson gave us the table and the derivative rule. Now we cash them in: transform, solve by algebra, split into partial fractions, invert. Data and forcing are inside Y from the start.
- Transforming gives P of s times Y equals G of s plus a s plus b times y of 0, plus a y prime of 0. Dividing by P, the initial-data piece is the zero-input response and the forcing piece is the zero-state response. Add them.
- Take y double prime plus 3 y prime plus 2 y equals 4, y of 0 equal 1, y prime of 0 zero. Solving gives s squared plus 3 s plus 4, over s times s plus 1 times s plus 2. Poles: 0 from the forcing, minus 1 and minus 2 from P.
- Cover up each factor: the rest gives 2 at s equals 0, minus 2 at minus 1, and 1 at minus 2. Each pole gives one term, in the same color on the pole plot: 2, minus 2 e to the minus t, e to the minus 2 t. Check: y of 0 is 1, y prime of 0 is 0.
- Zero-input response: 2 e to the minus t minus e to the minus 2 t, 0.60042 at t equals 1. Zero-state response: 2 minus 4 e to the minus t plus 2 e to the minus 2 t, 0.79915. Together, 1.39958. Only the second has the forcing's pole at 0, and its term 2 is where it settles.
- Now y double prime plus 2 y prime plus y equals 1, y of 0 equal 2. P is s plus 1 squared: a double pole. Minus F prime multiplies f by t, so 1 over s plus 1 squared inverts to t e to the minus t. At t equals 1: 1, 0.36788 and 0.36788, so 1.73576.
- Now y double prime plus 2 y prime plus 5 y equals 5, y of 0 zero, y prime of 0 equal 2. P is s plus 1 squared plus 4. Complete the square, writing s as s plus 1 minus 1: y is 1 minus e to the minus t cosine 2 t plus half e to the minus t sine 2 t. Poles at minus 1 plus or minus 2 i; y of 1 is 1.32035.
- One dictionary covers every example. A pole at sigma plus i omega, multiplicity m, gives t to the k, e to the sigma t, times cosine or sine of omega t. Sigma sets growth or decay, omega the frequency, multiplicity the power of t.
- Resonance: y double prime plus 4 y equals cosine 2 t. Y is s over s squared plus 4, squared: the forcing's poles at plus and minus 2 i land on the system's own, so each is double. So y is one quarter t sine 2 t: the factor t grows the envelope. At t equals 1, 0.22732.
- Forcing on a pole: the right side is e to the minus t. Its pole at minus 1 is a system pole, so it is double, and y is t minus 1 times e to the minus t, plus e to the minus 2 t. At t equals 1 the first two cancel: e to the minus 2, 0.13534.
- For a system, s I minus A, times X, equals x of 0. The determinant is s plus 1 times s plus 3, so X 1 and X 2 are halves over those poles. That determinant is the characteristic polynomial: the poles are the eigenvalues, minus 1 and minus 3. At t equals 1, 0.208833 and 0.159046.
- Example A the other way: roots minus 1 and minus 2, a constant guess of 2, and the data give c 1 equals minus 2, c 2 equals 1. Laplace wins when data and forcing are the whole problem; the characteristic equation wins on homogeneous ones. Neither handles tan t or nonlinear equations.
- On all five problems, the closed form and RK45 agree. Poles say which modes appear; the numerator says how much of each. Next: step functions, for forcing that switches on and off.
The pipeline, with forcing included
Take with data , , and let and . Transforming each term with the derivative rule gives , so
The first piece is what the system does with and only its initial state to drive it (the zero-input response). The second is what the forcing does starting from rest, (the zero-state response). Linearity means the true answer is their sum, and the algebra separates them without being asked. The factor multiplying the input is the transfer function, which 6.5 Convolution & Transfer Functions studies in earnest.
Example A (distinct real poles). , , . Here , so
- Transform: .
- Solve: .
- Partial fractions, by cover-up at each pole: at the rest gives ; at , ; at , . So .
- Invert term by term: .
Check the data: and . The split is also visible in the algebra: the first fraction inverts to (zero-input) and the second to (zero-state).
The lab runs the pipeline for five IVPs, one for each situation in this lesson. For each it assembles from the coefficients, lists the partial fractions and time terms, draws the poles of in the -plane in the same colour as the time terms they produce, and compares the closed form against an RK4 solution of the ODE itself. Press Next step to walk the five stages.
For Example A, set the probe time . The readout gives , made of (zero-input) plus (zero-state), which add to before rounding. The cross-check at gives from the rational function and from the three time terms, and from integrating the RK4 curve against , the definition from the previous lesson. So the algebra, the inverse and the definition agree, and the closed form sits within of RK4.
What the split is good for
Tick the lab’s split box and the two RK4 curves appear: the zero-input response starts at and the zero-state response starts at . They answer different questions. Double the forcing and only the zero-state part doubles; change the initial data and only the zero-input part moves. The zero-input part has only the poles that are roots of ; the zero-state part has those and also the forcing’s own poles, because it carries the extra denominator of . That is why the transient modes, set by , are the same whatever you push the system with, while the forced modes, set by , show up as extra poles. In Example A the pole at comes entirely from the constant forcing, and its term is the level the solution settles at, while the poles at and belong to and die out. A time stepper hands you one curve; the transform hands you the decomposition.
Repeated poles and the factor
Example B. , , . Now and
The cover-up rule finds at and the coefficient of () at the double pole; the remaining coefficient follows from matching the large- behaviour, , so .
The new entry is . Two routes reach it. The first shift theorem applied to gives . The second differentiates the definition with respect to :
This second route multiplies by for any , not only for exponentials. Either way , and at the lab’s three terms read , and , summing to (zero-input plus zero-state ). The mechanism is the same as in the characteristic-equation rule: a pole counted twice produces a term multiplied by one extra power of .
Complex poles: completing the square
Example C. , , . Here has no real roots, and
Split off the real pole with the cover-up rule, , then write the remainder over the quadratic: . Matching gives and matching gives , so
The trick is to rewrite the numerator as so that each piece matches a shifted table entry: and , which is why the last fraction needs a factor out front. Hence
Data check: . For the slope, at and at , total . The lab at gives , of which the zero-input part is and the zero-state part .
Pole location is mode
Every example shows the same dictionary. A pole at of multiplicity contributes terms and for (for real , just ). The real part sets growth or decay, the imaginary part sets the oscillation frequency, and the multiplicity sets the power of . A pole in the left half-plane decays, one on the imaginary axis neither decays nor grows, and one in the right half-plane grows exponentially, because does. The pole plot in the lab shows the first three: in Example C the pair at sits left of the dashed axis, matching the envelope and the frequency .
Two more presets use the same machinery on cases where the characteristic-equation method needs a special rule.
Resonance (preset 4). from rest. Then : the forcing’s poles land exactly on the system’s own poles, so each is double. Using with , ,
This is the growing oscillation of 5.6 Forced Oscillations & Resonance: the envelope is the factor from the double pole. No guessing rule is needed. At the lab gives .
Forcing on a pole (preset 5). from rest. The forcing pole at coincides with a system pole, so and . At the first two terms cancel and , which the lab prints.
First-order systems by the same algebra
A system transforms the same way. With the derivative rule gives , so , and the unknown is found by solving a small linear system whose entries are polynomials in . (A second-order ODE is the special case with and .)
Take and . Then with determinant . For a matrix the inverse is the swapped-and-negated matrix over the determinant, so
Inverting gives and . The eigenvalue method finds the same thing: the eigenvalues of are and with eigenvectors and , and supplies the two weights. The determinant is the characteristic polynomial, so the poles of a system’s transform are its eigenvalues. Laplace did it without ever mentioning an eigenvector, only a solve with polynomial entries. At the lab cell below prints and , and RK45 agrees.
Free sanity checks: initial and final values
Two standard facts about (stated here without proof, from memory) catch most algebra slips before you invert anything. The initial value theorem says . For Example A, tends to , matching ; for Example B it tends to , matching . The final value theorem says that if every pole of is strictly in the left half-plane, then . For Example A, at , and the dashed constant term in the lab is exactly ; for Examples B and C the limits are and , again the constant terms of the lab. The condition on the poles matters: in the resonant preset has poles on the imaginary axis, and does not settle at all, so the formula would claim a limit of that the solution never reaches.
The other habit worth building is to check the answer against the data every time: evaluate and from your closed form, as every example above does. A forgotten in the transform of , or a lost factor when matching , shows up immediately as a mismatch there.
Laplace against the characteristic equation
Solve Example A the other way. The characteristic roots are , so . A constant forcing suggests the constant (since ). Then gives and gives , so , : the same . The two methods perform the same linear algebra in different orders. Which should you reach for?
- Laplace wins when the data and forcing are the whole problem: initial conditions enter directly with no constants to fix afterward; and resonant or coinciding forcing (presets 4 and 5) needs no modified guess, because a repeated pole in makes the extra automatically. It wins even more when the forcing is switched on and off or is a sum of pieces, which 6.3 Step Functions & Discontinuous Forcing handles with a single new rule.
- The characteristic equation wins for a homogeneous equation: the roots give directly and the constants come from two equations, with no transform needed. It also extends to variable coefficients through series or other methods, where does not simplify (from memory: a coefficient like turns into , so the equation for becomes a differential equation for in , which is usually no easier).
- Both cost work in different places. Undetermined coefficients costs a guess table and a linear solve for the unknown constants. Laplace costs a partial-fraction expansion, which grows with the degree of the denominator: every pole the forcing adds raises the degree of the denominator of by one.
- Neither handles a forcing that has no rational transform or no neat guess (say ), and neither handles a nonlinear equation. For those, 5.5 Variation of Parameters or a time stepper from 4.3 Runge–Kutta Methods & Butcher Tableaux is the tool.
Computing it
Block A prints both ways, with the closed forms found above. Block B finds the factorisation , the eigenvalues and agreement with RK45. Block C confirms against the definition.
Now write the solver yourself. The exercise asks for the three steps in order: build , invert it, and compare against RK45.
With the cell completed, the Laplace solution and RK45 agree to roughly or better on both test problems.
Concept checks
Problems
Solve , , by Laplace. Then write the zero-input and zero-state responses separately.
Show worked solution
. Cover-up: at , ; at , ; at , . So . Check: , . The zero-input part is and the zero-state part is the remainder, .
Solve with , , and identify the pole structure of .
Show worked solution
. Write to get . By the shift theorem with , the inverses are , and , so . Check: . The pole has multiplicity three: two from the characteristic polynomial and one from the forcing , which is why the top power is .
Solve with . Use with .
Show worked solution
. With for , . Since (the bracket is ), . Check: and , so . The term grows without bound: resonance.
Key Takeaways
- Transforming gives , so is a rational function in which the initial data and the forcing both appear.
- The two pieces of are the zero-input response (initial data, no forcing) and zero-state response (forcing, from rest); their sum is the solution.
- Partial fractions then inverse: distinct poles give exponentials, a double pole gives a factor (shift theorem or ), a complex pair gives and after completing the square.
- Pole location is mode: real part is the decay rate, imaginary part the frequency, multiplicity the power of . Resonance is a forcing pole landing on a system pole.
- A first-order system is solved by ; its poles are the eigenvalues of .
- Laplace wins when initial data and discontinuous forcing are the problem; the characteristic equation is quicker for homogeneous equations; neither helps with nonlinear or variable-coefficient equations, where a time stepper takes over.
- Next, 6.3 Step Functions & Discontinuous Forcing adds switched forcing.