Motivation
A repeated characteristic root produces more than a repeated copy of the same exponential. It produces independent solutions with powers of time. Likewise, a forcing that resembles a homogeneous solution requires more than an arbitrary extra factor of time: the correct factor is determined by the root multiplicity.
These rules are often taught as a list of trial forms. The common mechanism is simpler to remember and more useful to check. Multiplication by an exponential shifts the differentiation operator, and differentiation acts as a nilpotent operator on a finite-dimensional polynomial space. This explains both the homogeneous basis and the forcing rule.
A polynomial trial is still only appropriate for certain inputs. We will also construct an integral formula that works for every continuous forcing, without differentiating that forcing. Its kernel needs a precise initial normalization; using any homogeneous solution as the kernel would generally give the wrong equation.
Definitions
Let and let
be a monic polynomial with real constant coefficients. Write and consider, for a fixed ,
where is continuous. A solution belongs to and satisfies the equation pointwise, with the usual one-sided endpoint interpretation. If an equation has a nonzero constant leading coefficient, divide every term, including the forcing, by it before using this monic convention.
Definition 1.12.1
(Root multiplicity and exponential-polynomial forcing)
A complex number has multiplicity as a root if with . We use when .
An exponential-polynomial forcing is , with a complex polynomial. Real forcings such as , with a real polynomial, can be obtained by taking real parts with . The forcing problem is resonant at precisely when .
Complex notation is a computational device. Because has real coefficients, taking the real part of a complex solution gives a solution for the real part of its forcing. A final real initial-value solution still has to satisfy the actual real initial data.
Introducing turns (1) into a first-order system. Its matrix has ones immediately above the diagonal and last row ; the forcing vector has only its last entry nonzero, equal to . The continuous-coefficient linear-system existence and uniqueness theorem gives exactly one solution for every initial vector on . In particular, the homogeneous solution space has dimension , and a homogeneous solution with all initial derivatives zero is identically zero. We will use this previously established initial-value theorem, rather than assume that guessing some exponentials proves completeness.
Theorem/Proposition
Theorem 1.12.2
(The complete homogeneous basis)
Let be the distinct complex roots of , with multiplicities ; thus . The functions
form a basis of the complex homogeneous solution space. For real solutions, a real root contributes . Each nonreal conjugate pair , with and common multiplicity , contributes
These real functions form a basis with exactly real members.
Theorem 1.12.3
(The resonance factor is the full root multiplicity)
Suppose is a polynomial of degree , and let be the multiplicity of in . There is exactly one polynomial of degree at most such that
In fact has degree . If is the leading coefficient of , write with . The leading coefficient of is
This is uniqueness within the stated trial space. All particular solutions differ from by a homogeneous solution. For a finite sum of forcings, add the separately constructed particular solutions.
Theorem 1.12.4
(The normalized homogeneous kernel gives continuous forcing)
Let be the homogeneous solution normalized by
For the zero conditions form an empty list and . For every continuous , the unique solution with all initial derivatives zero is
For general initial data, add the unique homogeneous solution with initial derivatives .
Proof
Shift the operator and count independent solutions
Proof
(Exponential shift, independence, and dimension count)
The product rule gives . Repeated application gives the more general identity
Here a polynomial in means its finite expansion as a constant-coefficient differential operator; its factors commute. If has multiplicity , then . For a polynomial of degree less than , this expression vanishes, proving that all functions in (2) are solutions.
To prove independence, suppose a linear combination is zero and group its terms as , with . Fix one root and apply
Every other group is annihilated. The remaining group becomes
On polynomials of degree at most , each operator with is invertible: in the monomial basis it is triangular with diagonal entries . Every factor in (9) therefore has an inverse on that same polynomial space. Thus . Repeating for each root proves independence. There are independent functions, and initial-value uniqueness bounded the dimension by , so the list spans every homogeneous solution.
Real coefficients make nonreal roots occur in conjugate pairs with equal multiplicities. Replacing each pair of complex functions by its real and imaginary parts is an invertible change of basis over the complex numbers. The functions in (3) are real and independent, and there are still of them. They therefore span the real solution space. One must choose one representative with from each pair; counting both signs again would double-count the same real modes.
Solve the polynomial problem after the shift
Proof
(Invert the shifted operator on polynomials)
Write for the polynomials of degree at most . On this space . The operator is invertible because its matrix is triangular with the nonzero constant on its diagonal. Explicitly, if
then is a polynomial in with no constant term, so , and
Set . Its leading coefficient is because derivatives lower polynomial degree. There is exactly one polynomial with : integrate exactly times and impose . If , simply take . The map sends to , so it is an isomorphism between these polynomial spaces.
The commuting factors give . Identity (8) proves (4). Integrating the leading term gives (5) and proves that the degree is exactly . Reversing these invertible operations proves uniqueness in .
Proof X-ray
Resonance is a failed inverse on a polynomial space
The proof moves the problem twice. Identity (8) first moves it from exponential-polynomials to polynomials, where is nilpotent. The factorization then separates an invertible part from a nilpotent part: is inverted by the finite series (10), which needs only , and is undone by integrations, which raise the degree by and land in . Multiplicity enters only as the exponent of that , so it alone fixes the resonance factor.
This explains the full trial form. For a degree- forcing, use all coefficients of , even when some forcing coefficients vanish. Differentiation can mix the lower powers. The factor supplies the missing derivatives caused by resonance; a single factor is sufficient only when the root is simple.
Differentiate the convolution only as far as the data allow
Proof
(Leibniz differentiation of the convolution)
Existence and uniqueness of follow from the homogeneous initial-value theorem. The function is smooth. For the integral (7), Leibniz's rule and the zero conditions in (6) give
At the next differentiation the first nonzero endpoint term appears:
These operations use continuity of , not its derivative. Combining them gives . All initial derivatives of order below are zero. Uniqueness then proves (7). The same calculation also explains why a kernel with the wrong initial normalization changes the forcing coefficient or introduces unwanted initial terms.
Worked examples
Comparing the three damping regimes
Let , and , and consider
Divide by and write and . The equation becomes . Its characteristic polynomial has roots
The sign of decides the form of the general solution. Throughout, we impose and .
Underdamped, . The roots are with , so . The condition gives , and gives . Hence
The case belongs here. Then and , so .
Critically damped, . There is a repeated root , so . The conditions give and , so
Overdamped, . The roots are real. Both are negative, because and . Write . The conditions and give
Long-time behavior. If then . In every case, each characteristic root then has negative real part. Hence every solution tends to as , not only the one above. In the critical case the factor does not spoil this. We still have , because exponential decay dominates polynomial growth. If , the roots are , and oscillates forever. The undamped equation is not a damped one.
Energy. Define . Along a solution,
So the energy never increases. However, vanishes at any instant where , such as a turning point. Thus its derivative need not be strictly negative at every instant. A nonincreasing function with has a limit as , but this alone does not show that the limit is . When , for example, is constant. For this linear equation, decay to zero is established by the roots.
Example. Take and . Then , which is the critical case. The solution is , with . Therefore
This is zero at , because the mass starts at rest, and negative for . Indeed , which decreases from toward .
Quick check
(For , , and , find and . Does ?)
Solution · Answer
The solution is . For all , , so . The solution oscillates with amplitude and does not tend to .
Worked example 1.12.5
(A double root requires two powers of time)
For , set . Equation (8) reduces it to . Choosing two zero integration constants gives
The complete trial was . Trying could produce only a constant after the two shifted derivatives and could not recover the linear term in the forcing. Here the displayed particular solution also has zero initial value and first derivative. For other initial data, and must be chosen; a particular solution does not automatically solve every initial-value problem.
Worked example 1.12.6
(A resonant real oscillation from one complex calculation)
To solve , first use forcing . The shifted polynomial is , so and . A particular complex solution is . Taking its real part gives
Indeed, . This particular solution has zero value and first derivative at zero. The linear amplitude growth expresses this specific undamped resonant equation; the word “resonance” alone does not prove unbounded growth in every forced system.
Worked example 1.12.7
(Repeated damping roots and the kernel normalization)
For , the normalized homogeneous kernel is , because and . The zero-initial-data response to is
Substitution gives , with . The other homogeneous solution would not work as this second-order kernel: its value at zero is nonzero, and its convolution already has a nonzero first derivative at zero for this forcing.
Counterexample mode
A finite polynomial-exponential trial does not cover every forcing
Consider on the whole half-line, with zero initial data. Every finite sum of polynomial-exponential terms, and the result of applying to it, is bounded in magnitude by for suitable constants. The forcing exceeds every such bound as . Thus no finite trial from that family solves this equation on the half-line. Nevertheless,
is a valid solution on every finite time interval. The kernel has value zero and derivative one at zero. The failure belongs to the restricted trial family, not to existence or uniqueness of the initial-value solution.
Common mistakes
Common mistake
(Counting the modes of a repeated root)
A root of multiplicity contributes independent exponential-polynomial modes, not copies of one exponential. A nonreal conjugate pair of multiplicity contributes real modes; using both signs again in the real list duplicates them. Check that the complete real list has exactly the order of the equation many independent members.
Common mistake
(Particular solutions and the resonance factor)
Forcing methods find a particular solution first. Add the homogeneous part before imposing general initial data. The resonance factor uses the multiplicity of the forcing exponent in the characteristic polynomial. It is not always one, and it does not mean multiply the homogeneous solution by the forcing.
Common mistake
(Annihilators enlarge the solution set)
Applying a differential operator that annihilates the forcing gives a necessary higher-order homogeneous equation. It can lose the forcing amplitude and add extra solution directions. Substitute back into the original equation before calling the enlarged homogeneous family a solution set. The final exercise makes this loss explicit.
Exercises
Quick check
(Solve with , . Explain why one exponential is insufficient to fit these three data.)
Solution
The homogeneous basis is , , . Write . The data give , , and . Hence
A single multiple of would have derivative two at zero once its value was fixed to one, so it cannot fit the prescribed first derivative. Root multiplicity supplies the other independent directions.
Quick check
(Find the full solution of . State the resonance multiplicity and explain why the particular polynomial has degree four after removing .)
Solution
The root has multiplicity two. With , the equation becomes , so
The forcing polynomial has degree two and two integrations raise the particular degree to four. The full allowed trial is ; coefficient matching makes its first two coefficients zero, rather than permitting us to omit them without checking.
Quick check
(Solve with zero initial value and first derivative. Give the complete real homogeneous basis before selecting the data.)
Solution · Answer
The roots are , each simple. The real homogeneous basis is and . Put ; the equation becomes . Thus a particular solution is . Its value and first derivative are zero at zero, so the given initial data select exactly this solution. Exponential decay can dominate the polynomial resonant factor in this damped example.
Quick check
(For arbitrary continuous , derive the kernel formula for with . Check the kernel's three normalization values, and explain why no derivative of is required.)
Solution
The homogeneous kernel is . It satisfies and , . Therefore
The first two differentiations have zero endpoint terms. The third gives multiplied by . At no step is differentiated. The initial conditions mean , not a condition on the third derivative.
Quick check
(Apply to . Solve the resulting fourth-order homogeneous equation, then impose the original equation on its coefficients. Which two coefficients does the original equation fix that the annihilated equation left free?)
Solution
The annihilated equation is , whose real general solution is
Substitution into the original equation gives . Matching requires and . The original solution set is therefore . The annihilated equation had left and unrestricted and thus forgot two forcing constraints. The free coefficients are precisely the original homogeneous freedom.
Nearby frequencies, resonance, and damping
The resonant calculation above becomes more informative when compared with a nearby frequency. Let and start an undamped oscillator from rest:
If , a cosine particular solution and the two initial conditions give
The first expression is a difference of the forced and natural oscillations. When the frequencies are close, the second shows a rapidly oscillating factor multiplied by a slowly varying one: this is the beating pattern. For each fixed unequal pair, (13) is bounded in time, although its bound may become large as the frequencies approach each other.
At equality the trial coefficient in (13) is undefined, but the initial-value solution has a finite limit for every fixed . Differentiating numerator and denominator with respect to , or using the derivative of cosine, yields
Direct substitution verifies this solves (12) at , with both initial values zero. Its envelope grows linearly. There is no contradiction with the boundedness for each unequal frequency: the bound in (13) is not uniform as , and the limit is taken at fixed time, not uniformly for all .
Proof X-ray
Why the apparent singularity cancels
Both cosine terms in the numerator of (13) agree at . Their first-order difference in carries a factor of , while the denominator has first-order change . Keeping the natural oscillation required by the initial data is what makes the finite limit visible; the particular solution alone generally has no finite limit.
Now add positive damping and consider
Try . Equating cosine and sine coefficients gives
Writing , we obtain
Every solution is plus a homogeneous solution. Both characteristic roots have negative real part because ; when the root is repeated, the homogeneous term is , which tends to zero as . Thus the difference tends to zero for any initial data. The periodic term is the steady response and the decaying part is the transient. The amplitude in (16) is that of the steady response, not necessarily the largest displacement during the transient.
Worked example 1.12.8
(A damped response at the natural frequency)
For , (16) gives , : the steady response is . With , the homogeneous correction is , so
The forcing frequency equals the undamped natural frequency, yet this response is bounded. Damping changes the equation and removes the undamped linear growth; equality of these frequencies alone does not imply an unbounded response.
Quick check
(Solve with . Then find the pointwise limit of the solution of with the same initial conditions as .)
Solution · Answer
Formula (13) gives . Substitution and the initial values check it. Formula (14), or the derivative of cosine in the frequency, gives the limit . This limiting solution is unbounded in time even though the solution at every fixed is bounded.
An electrical network: a series RLC circuit
Consider one ideal series loop containing an inductor, a resistor, a capacitor and a voltage source. Choose a direction around the loop and let be the current in that direction. Define as the charge on the capacitor plate into which the chosen current enters, so . With the passive sign convention, the voltage drops across the components are , and . Kirchhoff's voltage law balances their sum against the signed source voltage rise in the chosen direction:
Here , and are constant, and the lumped, ideal component model is assumed. Charge and current are the two initial data: and . Reversing the chosen direction is allowed, but the signs of current, charge and source voltage must then be changed consistently. The equation follows from the stated model; it does not describe every real circuit without approximation.
Worked example 1.12.9
(Switching on a constant source)
Take henry, ohms, farad, and connect a constant volt source at . Suppose the capacitor is initially uncharged and the current is zero. For the equation and data are
A constant particular solution is . The characteristic polynomial is , so the complete solution is . The first initial condition gives ; since , the second gives . Thus
The charge approaches coulomb while the current approaches zero. This is consistent with the capacitor eventually sustaining the source voltage in this ideal DC model. Direct substitution gives , and the two initial values are satisfied. The transient is critically damped because the characteristic root is repeated; zero final current does not mean zero final charge.
There is a useful independent check. The stored energy is
Multiplying (17) by gives : the signed source power enters the circuit, and is dissipated in the resistor. If , energy is returned to the source. Therefore need not decrease while the source is on. If , then ; this inequality alone should not be confused with a proof of a particular rate of decay.
Quick check
(Keep and , but set , , and . Find and and check the energy balance.)
Solution · Answer
The repeated-root solution is . The initial conditions give , so and . Here , as required. The negative current means the capacitor discharges opposite to the chosen positive direction.