University Calculus · Calculus IV

Unit D4: Nonhomogeneous Second-Order ODEs

When a linear oscillator is pushed by an external force, the response splits into a complementary solution and a particular solution. This unit builds both pieces with undetermined coefficients and variation of parameters.

Calculus IV ODEs Differential Equations MIT 18.03 / GT 2552
Read me first. This unit assumes you can solve a homogeneous constant-coefficient equation from its characteristic equation (Unit D3). We first prove that the general solution is the complementary solution plus any particular solution, then build particular solutions two ways: undetermined coefficients for the standard forcing families, and variation of parameters for everything else. The resonance section explains the modification rule, and the applications section ties the algebra to driven oscillators and RLC circuits. Work each example by hand before opening the collapsibles, and confirm every quiz answer against its worked solution.

Structure of the General Solution

A linear second-order ordinary differential equation with constant coefficients has the form $a y'' + b y' + c y = g(t)$, where $a e 0$. When the forcing term $g(t)$ is identically zero the equation is homogeneous; when $g(t) e 0$ it is nonhomogeneous. This unit assumes the homogeneous theory (Unit D3) and shows how the forcing term enters the solution.

具有常系数的线性二阶常微分方程形如 $a y'' + b y' + c y = g(t)$,其中 $a e 0$。当强迫项 $g(t)$ 恒为零时,方程为齐次方程;当 $g(t) e 0$ 时,方程为非齐次方程。本单元以齐次理论(第D3单元)为前提,说明强迫项如何进入解的表达式。

Key idea. The general solution of a nonhomogeneous linear equation is the sum of two pieces: the general solution $y_h$ of the associated homogeneous equation (the complementary solution), and any single solution $y_p$ of the full equation (a particular solution). The complementary part carries all the arbitrary constants; the particular part pins the response to the forcing.

Structure of the general solution
$$ y(t) = y_h(t) + y_p(t), \qquad a y_h'' + b y_h' + c y_h = 0, \qquad a y_p'' + b y_p' + c y_p = g(t). $$

The reason this works is linearity. Let $L[y] = a y'' + b y' + c y$. The operator $L$ is linear, so $L[y_h + y_p] = L[y_h] + L[y_p] = 0 + g = g$. Conversely, if $y$ and $y_p$ both solve the full equation, then $L[y - y_p] = g - g = 0$, so their difference is a homogeneous solution. Every solution therefore has the stated form.

Complementary solution from the characteristic equation
$$ a r^2 + b r + c = 0 \;\Rightarrow\; y_h = \begin{cases} C_1 e^{r_1 t} + C_2 e^{r_2 t}, & \text{distinct real } r_1, r_2,\\[2pt] (C_1 + C_2 t)\,e^{r t}, & \text{repeated root } r,\\[2pt] e^{\alpha t}\!\left(C_1 \cos\beta t + C_2 \sin\beta t\right), & r = \alpha \pm i\beta. \end{cases} $$

Remark. A boundary or initial condition must be applied to the full solution $y_h + y_p$, never to $y_h$ alone. Imposing initial data on the complementary part before adding the particular part is a common error that produces the wrong constants.

注。 边界条件或初始条件必须代入完整解 $y_h + y_p$,绝不能仅代入 $y_h$。在加上特解之前就将初始数据施加到齐次解上,是一个常见错误,会得到错误的常数。

Worked Example 1.1: assembling a general solution

Given that $y_p = t - \tfrac12$ solves $y'' - y = -t + \tfrac12$, write the general solution.

The associated homogeneous equation is $y'' - y = 0$ with characteristic equation $r^2 - 1 = 0$, roots $r = \pm 1$. Hence $y_h = C_1 e^{t} + C_2 e^{-t}$.

$$ y(t) = C_1 e^{t} + C_2 e^{-t} + t - \tfrac12. $$

Check: $y_p'' = 0$ and $-y_p = -t + \tfrac12$, so $y_p'' - y_p = -t + \tfrac12$, matching the right side.

Worked Example 1.2: applying initial data to the full solution

Solve the initial value problem $y'' - y = -t + \tfrac12$ with $y(0) = 2$, $y'(0) = 0$, using the structure from Worked Example 1.1.

The general solution is $y(t) = C_1 e^{t} + C_2 e^{-t} + t - \tfrac12$. Differentiate: $y'(t) = C_1 e^{t} - C_2 e^{-t} + 1$. Now impose the data on the full solution, not on $y_h$.

$$ y(0) = C_1 + C_2 - \tfrac12 = 2 \;\Rightarrow\; C_1 + C_2 = \tfrac52, $$ $$ y'(0) = C_1 - C_2 + 1 = 0 \;\Rightarrow\; C_1 - C_2 = -1. $$

Adding and subtracting gives $C_1 = \tfrac34$ and $C_2 = \tfrac74$. The solution of the initial value problem is

$$ y(t) = \tfrac34 e^{t} + \tfrac74 e^{-t} + t - \tfrac12. $$

The particular part $t - \tfrac12$ contributes $-\tfrac12$ to $y(0)$ and $+1$ to $y'(0)$, which is exactly why the constants differ from what the bare homogeneous problem would give.

Worked Example 1.3: recovering the equation from a known solution

Suppose $y_p = \tfrac12 t^2$ is claimed to be a particular solution of $y'' + 4y = g(t)$ for some forcing $g$. Identify $g$ and write the general solution.

Apply the operator directly. Here $y_p' = t$ and $y_p'' = 1$, so

$$ g(t) = y_p'' + 4 y_p = 1 + 4\cdot\tfrac12 t^2 = 1 + 2 t^2. $$

The homogeneous equation $y'' + 4y = 0$ has characteristic roots $r = \pm 2i$, so $y_h = C_1\cos 2t + C_2\sin 2t$. The general solution of $y'' + 4y = 1 + 2t^2$ is therefore

$$ y(t) = C_1\cos 2t + C_2\sin 2t + \tfrac12 t^2. $$

This reverse reading is a useful check: applying $L$ to any candidate $y_p$ must reproduce the forcing exactly, or the candidate is wrong.

Common error. Students often apply the initial conditions $y(0), y'(0)$ to the complementary part $y_h$ before adding $y_p$, then tack on $y_p$ afterward. This is wrong: the particular solution generally contributes nonzero values to $y(0)$ and $y'(0)$, so the constants $C_1, C_2$ must be solved from the full expression $y_h + y_p$. The fix is to assemble $y = y_h + y_p$ first, differentiate that whole sum, and only then substitute the data.
Going deeper: why a linear operator forces the sum structure

Define $L[y] = a y'' + b y' + c y$. Linearity means $L[\alpha u + \beta v] = \alpha L[u] + \beta L[v]$ for constants $\alpha, \beta$, which follows because differentiation is linear and multiplication by the coefficients $a, b, c$ is linear.

Suppose $L[y_p] = g$. Let $y$ be any other solution, so $L[y] = g$. Then

$$ L[y - y_p] = L[y] - L[y_p] = g - g = 0, $$

so $w = y - y_p$ solves the homogeneous equation, meaning $w = y_h$ for suitable constants. Therefore $y = y_h + y_p$. Conversely, for any homogeneous $y_h$, $L[y_h + y_p] = 0 + g = g$, so every such sum is a solution. The solution set is the single particular solution shifted by the two-dimensional space of homogeneous solutions.

In linear-algebra terms, the solution set of $L[y] = g$ is an affine subspace: a fixed translate $y_p$ of $\ker L$. The kernel $\ker L = \{y : L[y] = 0\}$ is the two-dimensional space of homogeneous solutions (existence and uniqueness pins the dimension at exactly two). Choosing a different particular solution $\tilde y_p$ only re-labels constants, since $\tilde y_p - y_p \in \ker L$ is absorbed into $C_1 y_1 + C_2 y_2$. That is why we may take the simplest $y_p$ we can find.

The general solution of $a y'' + b y' + c y = g(t)$ with $g e 0$ is:方程 $a y'' + b y' + c y = g(t)$($g e 0$)的通解是:
1.1
$y_p$ alone
$y_h$ alone
$y_h + y_p$
$y_h \cdot y_p$
Correct. The general solution is the complementary solution plus any particular solution.
By linearity, every solution is a homogeneous solution plus a fixed particular solution: $y_h + y_p$.
The arbitrary constants $C_1, C_2$ in the general solution belong to which part?
1.2
the particular solution $y_p$
the complementary solution $y_h$
the forcing term $g(t)$
neither part
Correct. The complementary solution $y_h$ carries the two arbitrary constants; $y_p$ is one fixed function.
A particular solution is a single function with no free constants. The constants live in $y_h$.

Undetermined Coefficients

The method of undetermined coefficients finds a particular solution by guessing a form that mirrors the forcing term, then solving for the unknown coefficients. It applies when $g(t)$ is built from exponentials, polynomials, sines and cosines, and products of these, because such functions reproduce their own kind under differentiation.

待定系数法通过猜测一个与强迫项形式相同的函数形式来求特解,再解出未知系数。该方法适用于 $g(t)$ 由指数函数、多项式、正弦与余弦及其乘积构成的情形,因为这类函数在求导后仍属同类。

Key idea. Choose a trial $y_p$ of the same functional family as $g(t)$, with undetermined constants. Substitute into $L[y_p] = g$ and match coefficients of like terms. The forcing families and their trial forms are fixed in advance.

Trial forms for undetermined coefficients
$$ \begin{array}{l|l} g(t) & \text{trial } y_p \\ \hline P_n(t) \ (\deg n) & A_n t^n + \cdots + A_1 t + A_0 \\ e^{\alpha t} & A e^{\alpha t} \\ \cos\beta t \ \text{or} \ \sin\beta t & A\cos\beta t + B\sin\beta t \\ e^{\alpha t}\cos\beta t \ \text{or} \ e^{\alpha t}\sin\beta t & e^{\alpha t}(A\cos\beta t + B\sin\beta t) \end{array} $$

For a polynomial forcing of degree $n$, the trial must include all lower-degree terms, not just the highest power, because differentiation mixes them. For a trigonometric forcing, always include both $\cos$ and $\sin$ terms even if only one appears in $g$, since $L$ couples them.

对于 $n$ 次多项式强迫项,试解必须包含所有低次项,而不仅仅是最高次项,因为求导会将各次项混合。对于三角函数型强迫项,始终应包含 $\cos$ 和 $\sin$ 两项,即使 $g$ 中只出现一项,因为 $L$ 会将它们耦合。

Superposition. If $g(t) = g_1(t) + g_2(t)$, solve $L[y_1] = g_1$ and $L[y_2] = g_2$ separately and add: $y_p = y_1 + y_2$. This follows directly from linearity of $L$.

叠加原理。 若 $g(t) = g_1(t) + g_2(t)$,分别求解 $L[y_1] = g_1$ 和 $L[y_2] = g_2$,再相加:$y_p = y_1 + y_2$。这直接由 $L$ 的线性性得出。

Superposition for a sum of forcings
$$ L[y_1] = g_1, \quad L[y_2] = g_2 \;\Rightarrow\; L[y_1 + y_2] = g_1 + g_2. $$
Worked Example 2.1: exponential forcing

Solve $y'' - 3y' + 2y = 4 e^{3t}$.

Homogeneous: $r^2 - 3r + 2 = (r-1)(r-2) = 0$, so $y_h = C_1 e^{t} + C_2 e^{2t}$. Since $3$ is not a root, no overlap, so try $y_p = A e^{3t}$.

$$ y_p'' - 3y_p' + 2y_p = (9A - 9A + 2A)e^{3t} = 2A e^{3t} = 4 e^{3t}, $$

so $2A = 4$ and $A = 2$. Hence

$$ y = C_1 e^{t} + C_2 e^{2t} + 2 e^{3t}. $$
Worked Example 2.2: polynomial and trigonometric forcing

Find a particular solution of $y'' + y = 2t + \cos 2t$.

Split by superposition. For $g_1 = 2t$, try $y_1 = A t + B$. Then $y_1'' + y_1 = 0 + At + B = 2t$, forcing $A = 2$, $B = 0$, so $y_1 = 2t$.

For $g_2 = \cos 2t$, since $\pm 2i$ are not roots of $r^2 + 1 = 0$, try $y_2 = C\cos 2t + D\sin 2t$. Then $y_2'' = -4C\cos 2t - 4D\sin 2t$, so

$$ y_2'' + y_2 = -3C\cos 2t - 3D\sin 2t = \cos 2t, $$

giving $-3C = 1$, $D = 0$, so $C = -\tfrac13$. Therefore

$$ y_p = 2t - \tfrac13\cos 2t. $$
Worked Example 2.3: product forcing $e^{\alpha t}\cos\beta t$

Solve $y'' + 4y' + 5y = e^{-t}\cos 2t$.

Homogeneous: $r^2 + 4r + 5 = 0$ gives $r = -2 \pm i$, so $y_h = e^{-2t}(C_1\cos t + C_2\sin t)$. The forcing $e^{-t}\cos 2t$ does not appear in $y_h$, so there is no resonance. Use the product trial $y_p = e^{-t}(A\cos 2t + B\sin 2t)$. Write $u = A\cos 2t + B\sin 2t$ so $y_p = e^{-t}u$. Then

$$ y_p' = e^{-t}(u' - u), \qquad y_p'' = e^{-t}(u'' - 2u' + u). $$

Substitute into $y'' + 4y' + 5y$:

$$ e^{-t}\big[(u'' - 2u' + u) + 4(u' - u) + 5u\big] = e^{-t}\big[u'' + 2u' + 2u\big]. $$

With $u = A\cos 2t + B\sin 2t$: $u'' = -4u$ and $u' = -2A\sin 2t + 2B\cos 2t$. So $u'' + 2u' + 2u = -2u + 2u' = -2(A\cos 2t + B\sin 2t) + 2(-2A\sin 2t + 2B\cos 2t)$, which equals $(-2A + 4B)\cos 2t + (-2B - 4A)\sin 2t$. Setting this equal to $\cos 2t$:

$$ -2A + 4B = 1, \qquad -4A - 2B = 0. $$

The second gives $B = -2A$; substituting, $-2A - 8A = 1$, so $A = -\tfrac{1}{10}$ and $B = \tfrac15$. Hence

$$ y_p = e^{-t}\!\left(-\tfrac{1}{10}\cos 2t + \tfrac15\sin 2t\right). $$

The key technique is factoring out $e^{-t}$ and shifting the operator onto $u$, which turns a messy product into a plain trigonometric match.

Common error. For a sine-only or cosine-only forcing such as $g = \cos 2t$, students frequently try $y_p = A\cos 2t$ alone, dropping the $\sin$ term. The operator $L$ couples $\cos$ and $\sin$ through the first-derivative term, so the substitution produces a $\sin$ component that cannot be matched, and the system has no solution. Always include both $A\cos\beta t$ and $B\sin\beta t$ (one of the coefficients may legitimately come out zero, as in Worked Example 2.2, but you must let the algebra decide).
For $y'' + 4y = 5e^{t}$ (where $1$ is not a characteristic root), the correct trial particular solution is:对于 $y'' + 4y = 5e^{t}$(其中 $1$ 不是特征根),正确的试解特解是:
2.1
$A t e^{t}$
$A\cos t + B\sin t$
$A t + B$
$A e^{t}$

Correct. An exponential forcing with no resonance calls for $y_p = A e^{t}$.

正确。无共振的指数型强迫项对应 $y_p = A e^{t}$。

The forcing is a plain exponential and $1$ is not a root, so the trial is $A e^{t}$ with no extra factor of $t$.

强迫项为普通指数函数且 $1$ 不是特征根,因此试解为 $A e^{t}$,无需乘以额外的 $t$ 因子。

For a degree-2 polynomial forcing $g(t) = t^2$, the trial $y_p$ should be:对于二次多项式强迫项 $g(t) = t^2$,试解 $y_p$ 应为:
2.2
$A t^2 + B t + C$
$A t^2$
$A t^2 + B$
$A e^{t}$

Correct. A polynomial trial must include every lower-degree term, since differentiation mixes them.

正确。多项式试解必须包含所有低次项,因为求导会将各次项混合。

Use the full polynomial $A t^2 + B t + C$; dropping lower terms leaves the system unsolvable.

使用完整多项式 $At^2 + Bt + C$;舍去低次项将使方程组无解。

Resonance and the Modification Rule

The trial forms of Section 2 fail when the forcing term is itself a solution of the homogeneous equation. Substituting such a trial into $L$ yields zero, so no choice of coefficient can match a nonzero $g$. This is the phenomenon of resonance, and the fix is the modification rule.

当强迫项本身是齐次方程的解时,第2节的试解形式会失效。将此类试解代入 $L$ 后得零,因此无论如何选取系数都无法匹配非零的 $g$。这就是共振现象,而修正规则是解决之道。

Key idea.核心思想。

If any term of the chosen trial $y_p$ already appears in the complementary solution $y_h$, multiply the entire trial by $t$. If the offending term corresponds to a double root, multiply by $t^2$. The smallest power of $t$ that clears all overlap is the correct modification.

若所选试解 $y_p$ 中有任何项已出现在齐次解 $y_h$ 中,则将整个试解乘以 $t$。若问题项对应二重根,则乘以 $t^2$。能消除所有重叠的最小 $t$ 幂次即为正确修正。

Modification rule修正规则
$$ y_p = t^{s}\,(\text{base trial}), \qquad s = \text{multiplicity of the offending root} \in \{0, 1, 2\}. $$

Concretely, suppose $g(t) = e^{lpha t}$ and $lpha$ is a root of the characteristic polynomial. If $lpha$ is a simple root, use $y_p = A t e^{lpha t}$. If it is a double root, use $y_p = A t^2 e^{lpha t}$. The same rule applies to trigonometric and polynomial-exponential forcings.

具体地,设 $g(t) = e^{lpha t}$ 且 $lpha$ 是特征多项式的根。若 $lpha$ 为单根,用 $y_p = A t e^{lpha t}$;若为二重根,用 $y_p = A t^2 e^{lpha t}$。同样的规则适用于三角函数和多项式指数型强迫项。

Physical meaning. Resonance is what happens when an undamped or lightly damped oscillator is driven at its natural frequency. The factor of $t$ in the particular solution reflects the unbounded growth of amplitude that results from the system absorbing energy from a perfectly tuned drive.

物理意义。 共振是无阻尼或轻阻尼振荡器以固有频率受驱时发生的现象。特解中的 $t$ 因子反映了系统从精确调谐驱动中持续吸收能量所导致的振幅无界增长。

Worked Example 3.1: simple resonance with an exponential

Solve $y'' - 3y' + 2y = e^{t}$.

Homogeneous roots are $r = 1, 2$, so $y_h = C_1 e^{t} + C_2 e^{2t}$. The forcing $e^{t}$ matches the root $r = 1$ (simple), so the plain trial $A e^{t}$ would give zero. Modify to $y_p = A t e^{t}$.

Compute $y_p' = A(1 + t)e^{t}$ and $y_p'' = A(2 + t)e^{t}$. Then

$$ y_p'' - 3y_p' + 2y_p = A\big[(2+t) - 3(1+t) + 2t\big]e^{t} = A(2 + t - 3 - 3t + 2t)e^{t} = -A e^{t}. $$

Setting $-A = 1$ gives $A = -1$, so

$$ y = C_1 e^{t} + C_2 e^{2t} - t e^{t}. $$
Worked Example 3.2: pure resonance of an oscillator

Solve $y'' + \omega^2 y = \cos\omega t$, the undamped oscillator driven at its natural frequency $\omega$.

Homogeneous: $r^2 + \omega^2 = 0$ gives $r = \pm i\omega$, so $y_h = C_1\cos\omega t + C_2\sin\omega t$. The forcing $\cos\omega t$ already appears in $y_h$, so modify: $y_p = t(A\cos\omega t + B\sin\omega t)$.

Differentiating twice and substituting (the cosine-only terms cancel by the homogeneous structure) yields

$$ y_p'' + \omega^2 y_p = -2A\omega\sin\omega t + 2B\omega\cos\omega t = \cos\omega t. $$

Matching, $2B\omega = 1$ and $-2A\omega = 0$, so $A = 0$ and $B = \dfrac{1}{2\omega}$. Hence

$$ y_p = \frac{t}{2\omega}\sin\omega t. $$

The amplitude $t/(2\omega)$ grows linearly in time: the hallmark of resonance.

Worked Example 3.3: resonance with a double root

Solve $y'' - 2y' + y = e^{t}$. The characteristic equation $r^2 - 2r + 1 = (r-1)^2 = 0$ has the double root $r = 1$, so $y_h = (C_1 + C_2 t)e^{t}$.

The forcing $e^{t}$ matches the double root. Both $e^{t}$ and $t e^{t}$ already live in $y_h$, so the smallest clean modification multiplies the base trial $A e^{t}$ by $t^2$: take $y_p = A t^2 e^{t}$.

Differentiate using the product rule. With $y_p = A t^2 e^{t}$,

$$ y_p' = A(t^2 + 2t)e^{t}, \qquad y_p'' = A(t^2 + 4t + 2)e^{t}. $$

Substitute:

$$ y_p'' - 2y_p' + y_p = A\big[(t^2 + 4t + 2) - 2(t^2 + 2t) + t^2\big]e^{t} = A\cdot 2\,e^{t}. $$

The $t^2$ and $t$ terms cancel exactly (they must, since $(r-1)^2$ annihilates $e^{t}$ and $te^{t}$), leaving $2A e^{t} = e^{t}$, so $A = \tfrac12$. Hence

$$ y = (C_1 + C_2 t)e^{t} + \tfrac12 t^2 e^{t}. $$
Common error. When the base trial has several terms, students sometimes multiply only the offending term by $t$ rather than the whole trial. For $g = e^{2t}$ with $r = 2$ a simple root, the correct modified trial is $A t e^{2t}$, not a mixed $A e^{2t} + B t e^{2t}$, since the unmodified $A e^{2t}$ piece lies in $y_h$ and contributes nothing. Likewise the power of $t$ equals the multiplicity of the matched root, so a double root needs $t^2$, never $t^3$. Multiply the entire base trial by the smallest power $t^s$ that clears all overlap, and no more.
Going deeper: why multiplying by $t^s$ is exactly right

Write the operator in factored form using its characteristic roots: $L = a(D - r_1)(D - r_2)$, where $D = d/dt$. Suppose the forcing is $e^{\alpha t}$ and $\alpha = r_1$ is a simple root, while $r_2 \ne \alpha$.

The operator $(D - \alpha)$ has the shift property $(D - \alpha)\big(e^{\alpha t} h(t)\big) = e^{\alpha t} h'(t)$, which follows from the product rule since the $\alpha e^{\alpha t} h$ terms cancel. Apply $L$ to the candidate $y_p = t^s e^{\alpha t}$ (constant absorbed):

$$ (D - r_2)(D - \alpha)\big(t^s e^{\alpha t}\big) = (D - r_2)\big(e^{\alpha t}\,s\,t^{s-1}\big). $$

For this to produce a nonzero multiple of $e^{\alpha t}$ (matching the forcing $e^{\alpha t}$, which is $t^0 e^{\alpha t}$), we need $s\,t^{s-1}$ to contain a constant term, i.e. $s = 1$. With $s = 1$ the inner result is $e^{\alpha t}$, and applying $(D - r_2)$ gives $(\alpha - r_2)e^{\alpha t}$, a nonzero constant times $e^{\alpha t}$ because $\alpha \ne r_2$. So $s = 1$ works and $s = 0$ fails (it gets annihilated by $(D-\alpha)$).

If instead $\alpha$ is a double root, $L = a(D - \alpha)^2$, and $(D - \alpha)^2(t^s e^{\alpha t}) = e^{\alpha t}\,\dfrac{d^2}{dt^2}t^s = e^{\alpha t}\,s(s-1)t^{s-2}$. A constant term appears only when $s = 2$. This is precisely the modification rule: $s$ equals the multiplicity of the matched root, the smallest power that survives all the annihilating factors.

For $y'' - 3y' + 2y = e^{2t}$ with $y_h = C_1 e^{t} + C_2 e^{2t}$, the correct trial is:对于 $y'' - 3y' + 2y = e^{2t}$,其中 $y_h = C_1 e^{t} + C_2 e^{2t}$,正确的试解是:
3.1
$A e^{2t}$
$A t e^{2t}$
$A t^2 e^{2t}$
$A\cos 2t + B\sin 2t$

Correct. The forcing $e^{2t}$ matches the simple root $r = 2$, so multiply the base trial by $t$: $A t e^{2t}$.

正确。强迫项 $e^{2t}$ 与简单根 $r = 2$ 吻合,故将基础试解乘以 $t$:$A t e^{2t}$。

Since $2$ is a simple characteristic root, the plain trial fails. Multiply by $t$ once: $A t e^{2t}$.

由于 $2$ 是简单特征根,普通试解失效。将试解乘以一次 $t$:$A t e^{2t}$。

If the characteristic equation has the double root $r = 3$ and $g(t) = e^{3t}$, the trial $y_p$ is:若特征方程有二重根 $r = 3$ 且 $g(t) = e^{3t}$,则试解 $y_p$ 为:
3.2
$A e^{3t}$
$A t e^{3t}$
$A t^2 e^{3t}$
$A t^3 e^{3t}$

Correct. A double root has multiplicity $2$, so multiply by $t^2$: $A t^2 e^{3t}$.

正确。二重根的重数为 $2$,故乘以 $t^2$:$A t^2 e^{3t}$。

Both $e^{3t}$ and $t e^{3t}$ lie in $y_h$ for a double root, so the smallest clean factor is $t^2$.

对于二重根,$e^{3t}$ 和 $t e^{3t}$ 均在 $y_h$ 中,因此最小的清除因子为 $t^2$。

Variation of Parameters

Undetermined coefficients works only for the special forcing families above. Variation of parameters is a general method that produces a particular solution for any continuous $g(t)$, including $ an t$ or $1/t$ which have no finite trial form. It builds $y_p$ directly from the two homogeneous solutions.

待定系数法仅适用于上述特殊强迫函数族。参数变易法是一种通用方法,能对任意连续的 $g(t)$ 给出特解,包括 $ an t$ 或 $1/t$ 等不具有有限试解形式的强迫项。它直接由两个齐次解构造 $y_p$。

Key idea.核心思想。

Let $y_1, y_2$ be independent solutions of the homogeneous equation. Seek $y_p = u_1 y_1 + u_2 y_2$ where $u_1, u_2$ are functions of $t$. Imposing one convenient constraint reduces the problem to a $2 imes 2$ linear system whose determinant is the Wronskian.

设 $y_1, y_2$ 是齐次方程的线性无关解。令 $y_p = u_1 y_1 + u_2 y_2$,其中 $u_1, u_2$ 是关于 $t$ 的函数。施加一个方便的约束条件,将问题化为一个 $2 imes 2$ 线性方程组,其行列式即为朗斯基行列式。

Variation of parameters (equation in standard form $y'' + p y' + q y = f$)参数变易法(标准形式 $y'' + p y' + q y = f$ 下的方程)
$$ u_1' = \frac{-y_2 f}{W}, \qquad u_2' = \frac{y_1 f}{W}, \qquad W = \begin{vmatrix} y_1 & y_2 \\ y_1' & y_2' \end{vmatrix} = y_1 y_2' - y_2 y_1'. $$
Particular solution特解
$$ y_p = -y_1 \int \frac{y_2 f}{W}\,dt + y_2 \int \frac{y_1 f}{W}\,dt. $$

Standard form is required. Divide through by the leading coefficient $a$ so the equation reads $y'' + p y' + q y = f$. Then $f = g/a$, not $g$ itself. Forgetting this division is the most frequent mistake in applying the formula.

必须使用标准形式。 将方程除以最高次系数 $a$,使其化为 $y'' + p y' + q y = f$ 的形式。此时 $f = g/a$,而非 $g$ 本身。忘记这一步除法是应用该公式时最常见的错误。

Going deeper: deriving the variation-of-parameters formulas深入探讨:推导参数变易法公式

Start from $y_p = u_1 y_1 + u_2 y_2$. Differentiate:

$$ y_p' = u_1' y_1 + u_2' y_2 + u_1 y_1' + u_2 y_2'. $$

We have two unknown functions but only one equation to satisfy, so we are free to impose one constraint. Choose

$$ u_1' y_1 + u_2' y_2 = 0, $$

which kills the messy terms and leaves $y_p' = u_1 y_1' + u_2 y_2'$. Differentiate once more:

$$ y_p'' = u_1' y_1' + u_2' y_2' + u_1 y_1'' + u_2 y_2''. $$

Substitute $y_p, y_p', y_p''$ into $y'' + p y' + q y = f$ and group by $u_1$ and $u_2$. Each group is $y_i'' + p y_i' + q y_i = 0$ because $y_1, y_2$ solve the homogeneous equation, so everything cancels except

$$ u_1' y_1' + u_2' y_2' = f. $$

Together with the constraint we have the linear system

$$ \begin{pmatrix} y_1 & y_2 \\ y_1' & y_2' \end{pmatrix}\begin{pmatrix} u_1' \\ u_2' \end{pmatrix} = \begin{pmatrix} 0 \\ f \end{pmatrix}. $$

By Cramer's rule, with $W = y_1 y_2' - y_2 y_1'$,

$$ u_1' = \frac{-y_2 f}{W}, \qquad u_2' = \frac{y_1 f}{W}. $$

Integrate to recover $u_1, u_2$. Since $y_1, y_2$ are independent, $W \ne 0$, so the system is always solvable.

Worked Example 4.1: a forcing with no trial form例题 4.1:无有限试解形式的强迫项

Solve $y'' + y = \sec t$ on an interval where $\sec t$ is continuous.

Homogeneous solutions: $y_1 = \cos t$, $y_2 = \sin t$. The equation is already in standard form, so $f = \sec t$. The Wronskian is

$$ W = \cos t\,(\cos t) - \sin t\,(-\sin t) = \cos^2 t + \sin^2 t = 1. $$

Then

$$ u_1' = \frac{-y_2 f}{W} = -\sin t\,\sec t = -\tan t, \qquad u_2' = \frac{y_1 f}{W} = \cos t\,\sec t = 1. $$

Integrate: $u_1 = \ln|\cos t|$ and $u_2 = t$. Therefore

$$ y_p = \cos t\,\ln|\cos t| + t\,\sin t, $$ $$ y = C_1\cos t + C_2\sin t + \cos t\,\ln|\cos t| + t\,\sin t. $$
Worked Example 4.2: standard form with a non-unit leading coefficient例题 4.2:最高次系数非1时的标准化

Solve $4y'' - 4y' + y = e^{t/2}\sqrt{1 - t^2}$ on $(-1, 1)$. The forcing has no finite trial form, so use variation of parameters, but first put the equation in standard form.

Divide by the leading coefficient $4$:

$$ y'' - y' + \tfrac14 y = \tfrac14 e^{t/2}\sqrt{1 - t^2}, \qquad f = \tfrac14 e^{t/2}\sqrt{1 - t^2}. $$

The characteristic equation $4r^2 - 4r + 1 = (2r - 1)^2 = 0$ has the double root $r = \tfrac12$, so $y_1 = e^{t/2}$ and $y_2 = t\,e^{t/2}$. Compute the Wronskian:

$$ W = y_1 y_2' - y_2 y_1' = e^{t/2}\big(e^{t/2} + \tfrac12 t e^{t/2}\big) - t e^{t/2}\cdot\tfrac12 e^{t/2} = e^{t}. $$

Then

$$ u_1' = \frac{-y_2 f}{W} = \frac{-t e^{t/2}\cdot \tfrac14 e^{t/2}\sqrt{1-t^2}}{e^{t}} = -\tfrac14 t\sqrt{1 - t^2}, $$ $$ u_2' = \frac{y_1 f}{W} = \frac{e^{t/2}\cdot \tfrac14 e^{t/2}\sqrt{1-t^2}}{e^{t}} = \tfrac14\sqrt{1 - t^2}. $$

Integrate: $u_1 = \tfrac{1}{12}(1 - t^2)^{3/2}$ (using $\int t\sqrt{1-t^2}\,dt = -\tfrac13(1-t^2)^{3/2}$), and $u_2 = \tfrac14\!\left(\tfrac{t}{2}\sqrt{1-t^2} + \tfrac12\arcsin t\right)$. Then $y_p = u_1 y_1 + u_2 y_2$. The point is structural: had we forgotten to divide by $4$ and used $f = e^{t/2}\sqrt{1-t^2}$, every coefficient would have been four times too large.

Common error.常见错误。

The single most common mistake in variation of parameters is using the raw forcing $g$ in place of $f = g/a$ when the leading coefficient $a e 1$. The formulas $u_1' = -y_2 f / W$ and $u_2' = y_1 f / W$ require the equation in standard form with unit leading coefficient; reading $f$ from the un-divided equation makes the particular solution $a$ times too large.

参数变易法中最常见的单一错误,是当最高次系数 $a e 1$ 时,直接使用原始强迫项 $g$ 代替 $f = g/a$。在读取 $f$、$p$、$q$ 之前,始终先将 $y''$ 的系数化为1。

In variation of parameters, the determinant $W = y_1 y_2' - y_2 y_1'$ is called the:在参数变易法中,行列式 $W = y_1 y_2' - y_2 y_1'$ 称为:
4.1
Jacobian雅可比行列式
Laplacian拉普拉斯算子
Wronskian朗斯基行列式
Hessian海森矩阵

Correct. $W$ is the Wronskian of $y_1$ and $y_2$; it is nonzero precisely when the solutions are independent.

正确。$W$ 是 $y_1$ 和 $y_2$ 的朗斯基行列式;它非零当且仅当两个解线性无关。

The determinant of solutions and their first derivatives is the Wronskian $W$.

解及其一阶导数构成的行列式即为朗斯基行列式 $W$。

Before applying the formulas $u_1' = -y_2 f / W$ and $u_2' = y_1 f / W$, the equation must be written so that:在应用公式 $u_1' = -y_2 f / W$ 和 $u_2' = y_1 f / W$ 之前,方程必须满足:
4.2
the coefficient of $y''$ is $1$$y''$ 的系数为 $1$
the forcing is a polynomial强迫项为多项式
the Wronskian equals $1$朗斯基行列式等于 $1$
$y_1 = y_2$

Correct. The formula requires standard form $y'' + p y' + q y = f$, so divide by the leading coefficient first; then $f = g/a$.

正确。公式要求标准形式 $y'' + p y' + q y = f$,须先除以最高次系数;此时 $f = g/a$。

The derivation assumes a unit leading coefficient. Divide through by $a$ so the equation reads $y'' + p y' + q y = f$.

推导过程假设最高次系数为1。将方程除以 $a$,使其化为 $y'' + p y' + q y = f$。

Comparing the Methods

Both methods produce a valid particular solution, and adding it to $y_h$ gives the same general solution. The practical question is which method to reach for, since one is fast but narrow and the other is general but more laborious.

两种方法均能给出有效的特解,将其与 $y_h$ 相加得到相同的通解。实际问题是选用哪种方法,因为一种快速但适用范围窄,另一种通用但计算量较大。

Key idea.核心思想。

Use undetermined coefficients when $g(t)$ is a polynomial, exponential, sine or cosine, or a product of these: the algebra is short. Use variation of parameters for every other continuous forcing, or when you already know $y_1, y_2$ and want a guaranteed formula.

当 $g(t)$ 为多项式、指数函数、正弦或余弦或其乘积时,使用待定系数法,代数运算简短。对于其他所有连续强迫项,以及当你已知 $y_1, y_2$ 并需要通用公式时,使用参数变易法。

Decision summary方法决策汇总
$$ \begin{array}{l|l|l} \text{aspect} & \text{undetermined coefficients} & \text{variation of parameters} \\ \hline \text{forcing } g & \text{polynomial, } e^{\alpha t}, \cos, \sin, \text{products} & \text{any continuous } g \\ \text{effort} & \text{algebra, match coefficients} & \text{two integrals} \\ \text{resonance} & \text{handled by } t^{s} \text{ rule} & \text{handled automatically} \end{array} $$

Note that variation of parameters needs no special case for resonance: if $g$ happens to coincide with a homogeneous solution, the integrals simply produce an extra $t$ factor automatically. Undetermined coefficients trades that generality for speed, at the cost of the explicit modification rule from Section 3.

注意,参数变易法无需对共振作特殊处理:若 $g$ 恰好与某个齐次解重合,积分过程将自动产生额外的 $t$ 因子。待定系数法以第3节的显式修正规则为代价,换取更快的计算速度。

Worked Example 5.1: same problem, both methods例题 5.1:同一问题,两种方法

Solve $y'' - y = e^{2t}$ two ways. Homogeneous: $r^2 - 1 = 0$, so $y_1 = e^{t}$, $y_2 = e^{-t}$, and $y_h = C_1 e^{t} + C_2 e^{-t}$.

Undetermined coefficients. Since $2$ is not a root, try $y_p = A e^{2t}$. Then $y_p'' - y_p = (4A - A)e^{2t} = 3A e^{2t} = e^{2t}$, so $A = \tfrac13$ and $y_p = \tfrac13 e^{2t}$.

Variation of parameters. The Wronskian is $W = e^{t}(-e^{-t}) - e^{-t}(e^{t}) = -2$, and $f = e^{2t}$.

$$ u_1' = \frac{-e^{-t} e^{2t}}{-2} = \tfrac12 e^{t}, \qquad u_2' = \frac{e^{t} e^{2t}}{-2} = -\tfrac12 e^{3t}. $$

Integrate: $u_1 = \tfrac12 e^{t}$, $u_2 = -\tfrac16 e^{3t}$. Then

$$ y_p = e^{t}\cdot\tfrac12 e^{t} + e^{-t}\cdot\big(-\tfrac16 e^{3t}\big) = \tfrac12 e^{2t} - \tfrac16 e^{2t} = \tfrac13 e^{2t}. $$

Both methods give the same $y_p = \tfrac13 e^{2t}$, as they must.

Worked Example 5.2: a resonant case, both methods例题 5.2:共振情形,两种方法

Solve $y'' - y = e^{t}$. Homogeneous: $r^2 - 1 = 0$, so $y_1 = e^{t}$, $y_2 = e^{-t}$, $y_h = C_1 e^{t} + C_2 e^{-t}$. The forcing $e^{t}$ matches the simple root $r = 1$: this is resonant.

Undetermined coefficients. The modification rule (Section 3) forces $y_p = A t e^{t}$. Then $y_p' = A(1+t)e^{t}$, $y_p'' = A(2+t)e^{t}$, so $y_p'' - y_p = A(2 + t - t)e^{t} = 2A e^{t} = e^{t}$, giving $A = \tfrac12$ and $y_p = \tfrac12 t e^{t}$.

Variation of parameters. No special case is needed. The Wronskian is $W = e^{t}(-e^{-t}) - e^{-t}(e^{t}) = -2$, and $f = e^{t}$.

$$ u_1' = \frac{-e^{-t} e^{t}}{-2} = \tfrac12, \qquad u_2' = \frac{e^{t} e^{t}}{-2} = -\tfrac12 e^{2t}. $$

Integrate: $u_1 = \tfrac12 t$ and $u_2 = -\tfrac14 e^{2t}$. Then

$$ y_p = e^{t}\cdot\tfrac12 t + e^{-t}\cdot\big(-\tfrac14 e^{2t}\big) = \tfrac12 t e^{t} - \tfrac14 e^{t}. $$

The extra $-\tfrac14 e^{t}$ is a homogeneous solution, so it merges into $C_1 e^{t}$. After absorbing it, the particular solution is again $\tfrac12 t e^{t}$. Notice the factor of $t$ appeared automatically from $\int u_1' = \int \tfrac12\,dt$, with no modification rule invoked.

Common error.常见错误。

When checking that two methods agree, students sometimes conclude the answers "disagree" because variation of parameters produced an extra term like $- frac14 e^{t}$ above. A particular solution is unique only up to a homogeneous solution, so you should never demand the two $y_p$ formulas match term for term.

在验证两种方法结果一致时,同学们有时会因参数变易法产生了额外项而得出答案"不一致"的结论。特解只在差一个齐次解的意义下唯一。

Which forcing requires variation of parameters because undetermined coefficients does not apply?
5.1
$g(t) = t^2 e^{t}$
$g(t) = \tan t$
$g(t) = 3\cos 5t$
$g(t) = 7$
Correct. $\tan t$ is not a polynomial, exponential, or sinusoid, so it has no finite trial form; use variation of parameters.
The other three forcings have standard trial forms. Only $\tan t$ falls outside the undetermined-coefficients families.

Applications

The canonical application is the driven mechanical or electrical oscillator. A mass on a spring with damping and an external force obeys exactly the nonhomogeneous equation studied here, and the same equation governs a series RLC circuit driven by a voltage source.

典型应用是受迫机械或电气振荡器。带阻尼和外力的弹簧质量系统恰好满足本节研究的非齐次方程,同一方程也描述由电压源驱动的串联RLC电路。

Key idea.核心思想。

A forced damped oscillator $m y'' + c y' + k y = F(t)$ has a solution $y = y_h + y_p$. The complementary part $y_h$ decays to zero when there is damping (the transient); the particular part $y_p$ persists at the forcing frequency (the steady state). Once the transient dies, only $y_p$ remains.

受迫阻尼振荡器 $m y'' + c y' + k y = F(t)$ 的解为 $y = y_h + y_p$。有阻尼时,齐次解 $y_h$ 趋向零(瞬态),特解 $y_p$ 持续存在(稳态)。瞬态消散后,只剩 $y_p$。

Forced spring-mass-damper and the RLC circuit受迫弹簧质量阻尼系统与RLC电路
$$ m y'' + c y' + k y = F(t), \qquad L q'' + R q' + \tfrac{1}{C} q = E(t). $$

For sinusoidal forcing $F(t) = F_0\cos\omega t$, the steady-state response is a sinusoid of the same frequency but shifted in phase and scaled in amplitude. The amplitude as a function of drive frequency $\omega$ peaks near the natural frequency, which is the practical face of resonance from Section 3.

对于正弦型强迫项 $F(t) = F_0\cos\omega t$,稳态响应是同频率但相位偏移、振幅缩放的正弦函数。振幅关于驱动频率 $\omega$ 的函数在固有频率附近达到峰值,这正是第3节共振现象的实际体现。

Steady-state amplitude for $m y'' + c y' + k y = F_0\cos\omega t$$m y'' + c y' + k y = F_0\cos\omega t$ 的稳态振幅
$$ A(\omega) = \frac{F_0}{\sqrt{(k - m\omega^2)^2 + (c\omega)^2}}. $$
Worked Example 6.1: transient and steady state例题 6.1:瞬态与稳态

Solve $y'' + 2y' + 2y = 5\cos t$ and identify the transient and steady-state parts.

Homogeneous: $r^2 + 2r + 2 = 0$ gives $r = -1 \pm i$, so $y_h = e^{-t}(C_1\cos t + C_2\sin t)$. This decays to zero as $t \to \infty$: it is the transient.

For the steady state, try $y_p = A\cos t + B\sin t$. Then $y_p'' = -A\cos t - B\sin t$ and $y_p' = -A\sin t + B\cos t$, so

$$ y_p'' + 2y_p' + 2y_p = (A + 2B)\cos t + (B - 2A)\sin t = 5\cos t. $$

Matching: $A + 2B = 5$ and $B - 2A = 0$, so $B = 2A$ and $A + 4A = 5$, giving $A = 1$, $B = 2$. Thus the steady state is $y_p = \cos t + 2\sin t$, and

$$ y = \underbrace{e^{-t}(C_1\cos t + C_2\sin t)}_{\text{transient}} + \underbrace{\cos t + 2\sin t}_{\text{steady state}}. $$
Worked Example 6.2: an RLC circuit例题 6.2:RLC电路

A series circuit with $L = 1$, $R = 3$, $1/C = 2$ is driven by $E(t) = 10\sin t$. The charge obeys $q'' + 3q' + 2q = 10\sin t$. Find the steady-state charge.

Homogeneous roots: $r^2 + 3r + 2 = (r+1)(r+2) = 0$, both negative, so $q_h$ is transient. For steady state try $q_p = A\cos t + B\sin t$.

$$ q_p'' + 3q_p' + 2q_p = (A + 3B)\cos t + (B - 3A)\sin t = 10\sin t. $$

Match: $A + 3B = 0$ and $B - 3A = 10$. From the first, $A = -3B$; substituting, $B - 3(-3B) = 10B = 10$, so $B = 1$ and $A = -3$. The steady-state charge is

$$ q_p = -3\cos t + \sin t. $$
Worked Example 6.3: steady-state amplitude and phase例题 6.3:稳态振幅与相位

For the steady state $y_p = \cos t + 2\sin t$ found in Worked Example 6.1, express it as a single shifted cosine $R\cos(t - \phi)$.

The amplitude and phase come from $R = \sqrt{A^2 + B^2}$ and $\tan\phi = B/A$ for $y_p = A\cos t + B\sin t = R\cos(t - \phi)$:

$$ R = \sqrt{1^2 + 2^2} = \sqrt{5}, \qquad \tan\phi = \frac{2}{1} = 2,\;\; \phi = \arctan 2 \approx 1.107\ \text{rad}. $$

So the steady-state oscillation has amplitude $\sqrt5 \approx 2.236$ and lags the drive by about $63.4^\circ$. Cross-check against the amplitude formula with $m = 1$, $c = 2$, $k = 2$, $F_0 = 5$, $\omega = 1$:

$$ A(\omega) = \frac{F_0}{\sqrt{(k - m\omega^2)^2 + (c\omega)^2}} = \frac{5}{\sqrt{(2 - 1)^2 + 2^2}} = \frac{5}{\sqrt5} = \sqrt5. $$

The amplitude formula and the direct coefficient computation agree, which is a reliable way to catch arithmetic slips.

Common error.常见错误。

When a question asks only for the steady-state response, students sometimes solve the full initial value problem and report $y_h + y_p$ with the constants. With damping ($c > 0$) the transient $y_h$ decays to zero, so the steady state is $y_p$ alone. Conversely, if the system is undamped ($c = 0$), there is no transient decay and no true steady state.

当题目只要求稳态响应时,同学们有时会求解完整的初值问题并给出含常数的 $y_h + y_p$。有阻尼($c > 0$)时,瞬态 $y_h$ 趋向零,因此稳态仅为 $y_p$。反之,若系统无阻尼($c = 0$),则没有瞬态衰减,也没有真正的稳态。

In a damped driven oscillator $m y'' + c y' + k y = F(t)$ with $c > 0$, the complementary solution $y_h$ is called the:在阻尼受迫振荡器 $m y'' + c y' + k y = F(t)$($c > 0$)中,齐次解 $y_h$ 称为:
6.1
steady state稳态
resonance共振
forcing强迫项
transient瞬态

Correct. With damping, $y_h$ decays to zero over time, so it is the transient response.

正确。有阻尼时,$y_h$ 随时间趋向零,因此它是瞬态响应。

Damping makes the homogeneous part die out: it is the transient. The persisting $y_p$ is the steady state.

阻尼使得齐次部分衰减至零,它是瞬态。持续存在的 $y_p$ 是稳态。

Going Deeper

Two ideas extend the methods of this unit. The first is a unifying view through the Wronskian and the Green's function; the second is the resonance amplitude as a function of frequency, which explains why resonance is sharp when damping is small.

两个思想延伸了本单元的方法。第一个是通过朗斯基行列式和格林函数的统一视角;第二个是振幅关于频率的函数,解释了为何阻尼小时共振峰值会很尖锐。

Key idea.核心思想。

Variation of parameters can be packaged as an integral against a single kernel, the Green's function, which depends only on the homogeneous solutions. The particular solution becomes a definite integral of the forcing against this kernel, presenting the response as a superposition of impulse responses.

参数变易法可以整合为对单一核函数(即格林函数)的积分形式,该核函数仅依赖于齐次解。特解化为强迫项关于此核函数的定积分,将响应呈现为一系列冲激响应的叠加。

Particular solution as a Green's-function integral以格林函数积分表示的特解
$$ y_p(t) = \int_{t_0}^{t} G(t, \tau)\,f(\tau)\,d\tau, \qquad G(t,\tau) = \frac{y_1(\tau) y_2(t) - y_2(\tau) y_1(t)}{W(\tau)}. $$

This is precisely the variation-of-parameters answer with the integration carried inside a single kernel. Each value $f( au)$ acts like a small impulse; $G(t, au)$ is the system's response at time $t$ to a unit impulse delivered at time $ au$. The total response is the accumulated sum of these impulse responses.

这正是参数变易法的结果,只是将积分整合进单一核函数内。每个值 $f( au)$ 如同一个小冲激,$G(t, au)$ 是系统在时刻 $ au$ 受到单位冲激后在时刻 $t$ 的响应。总响应是这些冲激响应的累积之和。

Going deeper: resonance amplitude and its peak深入探讨:共振振幅及其峰值

Take $m y'' + c y' + k y = F_0\cos\omega t$ with steady-state amplitude

$$ A(\omega) = \frac{F_0}{\sqrt{(k - m\omega^2)^2 + (c\omega)^2}}. $$

To find the resonant frequency, minimize the denominator $D(\omega) = (k - m\omega^2)^2 + c^2\omega^2$. Differentiate with respect to $\omega$ and set to zero:

$$ D'(\omega) = 2(k - m\omega^2)(-2m\omega) + 2c^2\omega = 2\omega\big[-2m(k - m\omega^2) + c^2\big] = 0. $$

Discarding $\omega = 0$, the bracket gives

$$ \omega_{\text{res}}^2 = \frac{k}{m} - \frac{c^2}{2m^2}. $$

When damping $c$ is small, $\omega_{\text{res}}^2 \approx k/m$, the square of the natural frequency, and the denominator nearly vanishes, so $A(\omega)$ has a tall sharp peak. As $c$ grows the peak shifts left and flattens, and once $c^2 \ge 2mk$ no interior peak survives: heavily damped systems do not resonate.

Worked Example 7.1: building the Green's function例题 7.1:构造格林函数

Construct $G(t, \tau)$ for $y'' + y = f(t)$.

Homogeneous solutions are $y_1 = \cos t$, $y_2 = \sin t$, with Wronskian $W = 1$. Substitute into the kernel:

$$ G(t, \tau) = \frac{\cos\tau\,\sin t - \sin\tau\,\cos t}{1} = \sin(t - \tau). $$

Therefore the particular solution to $y'' + y = f$ is

$$ y_p(t) = \int_{t_0}^{t} \sin(t - \tau)\,f(\tau)\,d\tau, $$

a single clean integral that solves the equation for any continuous $f$.

Worked Example 7.2: Green's function applied to a step forcing例题 7.2:格林函数应用于阶跃强迫项

Use the kernel $G(t,\tau) = \sin(t - \tau)$ from Worked Example 7.1 to solve $y'' + y = 1$ with $y(0) = 0$, $y'(0) = 0$, taking $t_0 = 0$.

The Green's-function form gives the solution that already satisfies the zero initial data:

$$ y(t) = \int_0^t \sin(t - \tau)\cdot 1\,d\tau. $$

Substitute $s = t - \tau$, so $ds = -d\tau$ and the limits run from $s = t$ down to $s = 0$:

$$ y(t) = \int_0^t \sin s\,ds = \big[-\cos s\big]_0^t = 1 - \cos t. $$

Check directly: $y = 1 - \cos t$ gives $y'' = \cos t$, so $y'' + y = \cos t + 1 - \cos t = 1$, and $y(0) = 0$, $y'(0) = \sin 0 = 0$. The Green's function delivers the full solution of the initial value problem in one integral, with the homogeneous part already tuned to the zero initial conditions.

Worked Example 7.3: a sharp resonance peak例题 7.3:尖锐共振峰

For $y'' + 0.1\,y' + y = \cos\omega t$ (so $m = 1$, $c = 0.1$, $k = 1$, $F_0 = 1$), estimate the peak steady-state amplitude and the frequency where it occurs.

From the resonance result, $\omega_{\text{res}}^2 = \dfrac{k}{m} - \dfrac{c^2}{2m^2} = 1 - \dfrac{0.01}{2} = 0.995$, so $\omega_{\text{res}} \approx 0.9975$. The peak amplitude is

$$ A(\omega_{\text{res}}) = \frac{F_0}{\sqrt{(k - m\omega_{\text{res}}^2)^2 + (c\,\omega_{\text{res}})^2}} = \frac{1}{\sqrt{(0.005)^2 + (0.0998)^2}} \approx \frac{1}{0.0999} \approx 10.0. $$

With light damping the peak amplitude is roughly $F_0/(c\,\omega_{\text{res}}) \approx 1/(0.1\cdot 1) = 10$, ten times the static deflection $F_0/k = 1$. This factor, the quality factor $Q$, measures how sharply the system resonates: small $c$ produces a tall narrow peak.

Common error.常见错误。

A frequent mistake is to identify the resonant frequency of a damped system with the undamped natural frequency $\sqrt{k/m}$. The true amplitude peak sits at $\omega_{ ext{res}}^2 = k/m - c^2/(2m^2)$. Only in the zero-damping limit do the two frequencies nearly coincide.

一个常见错误是将阻尼系统的共振频率等同于无阻尼固有频率 $\sqrt{k/m}$。真正的振幅峰值位于 $\omega_{ ext{res}}^2 = k/m - c^2/(2m^2)$。

The Green's function $G(t, au)$ for a second-order linear equation represents the system's response to:二阶线性方程的格林函数 $G(t, au)$ 表示系统对以下输入的响应:
7.1
a constant forcing恒定强迫项
a unit impulse at time $\tau$
the homogeneous initial data齐次初始数据
a sinusoidal drive正弦驱动

Correct. $G(t, au)$ is the impulse response: the output at time $t$ from a unit impulse delivered at $ au$.

正确。$G(t, au)$ 是冲激响应:时刻 $ au$ 施加单位冲激后在时刻 $t$ 的输出。

The Green's function is the impulse response; integrating it against $f$ superposes the responses to each small impulse.

格林函数是冲激响应;将其与 $f$ 做积分,将每个小冲激的响应叠加起来。

Flashcards

0 / 12 flipped0 / 12 已翻
General solution structure通解结构

$y = y_h + y_p$: complementary solution plus any particular solution. The constants live in $y_h$.

$y = y_h + y_p$:齐次解加任意特解。任意常数存在于 $y_h$ 中。

Complementary solution $y_h$齐次解 $y_h$

General solution of $a y'' + b y' + c y = 0$, found from the characteristic roots.

$a y'' + b y' + c y = 0$ 的通解,由特征根求得。

Particular solution $y_p$特解 $y_p$

Any single function with $a y_p'' + b y_p' + c y_p = g(t)$. No arbitrary constants.

满足 $a y_p'' + b y_p' + c y_p = g(t)$ 的任意单一函数,不含任意常数。

Undetermined coefficients: when
Use when $g$ is a polynomial, $e^{\alpha t}$, $\cos\beta t$, $\sin\beta t$, or a product of these.
Trial for $e^{lpha t}$ (no resonance)$e^{lpha t}$ 的试解(无共振)

$y_p = A e^{lpha t}$. For $\coseta t$ or $\sineta t$ use $A\coseta t + B\sineta t$ (both terms).

$y_p = A e^{lpha t}$。对于 $\coseta t$ 或 $\sineta t$,用 $A\coseta t + B\sineta t$(两项都要)。

Superposition of forcings强迫项的叠加

If $g = g_1 + g_2$, solve each separately and add: $y_p = y_1 + y_2$.

若 $g = g_1 + g_2$,分别求解再相加:$y_p = y_1 + y_2$。

Resonance / modification rule共振与修正规则

If a trial term lies in $y_h$, multiply the trial by $t^{s}$, where $s$ is the root's multiplicity ($1$ or $2$).

若试解中某项已在 $y_h$ 中,将整个试解乘以 $t^{s}$,其中 $s$ 为该根的重数($1$ 或 $2$)。

Pure resonance amplitude纯共振振幅

$y'' + \omega^2 y = \cos\omega t$ gives $y_p = \dfrac{t}{2\omega}\sin\omega t$: amplitude grows linearly in $t$.

$y'' + \omega^2 y = \cos\omega t$ 给出 $y_p = \dfrac{t}{2\omega}\sin\omega t$:振幅随 $t$ 线性增长。

Variation of parameters formulas参数变易法公式

In standard form $y'' + p y' + q y = f$: $u_1' = -y_2 f / W$, $u_2' = y_1 f / W$, $W = y_1 y_2' - y_2 y_1'$.

在标准形式 $y'' + p y' + q y = f$ 下:$u_1' = -y_2 f / W$,$u_2' = y_1 f / W$,$W = y_1 y_2' - y_2 y_1'$。

Wronskian $W$朗斯基行列式 $W$

$W = y_1 y_2' - y_2 y_1'$. Nonzero exactly when $y_1, y_2$ are independent, so the system is solvable.

$W = y_1 y_2' - y_2 y_1'$。当且仅当 $y_1, y_2$ 线性无关时非零,方程组可解。

Transient vs steady state瞬态与稳态

With damping, $y_h o 0$ (transient) and $y_p$ persists (steady state).

有阻尼时,$y_h o 0$(瞬态),$y_p$ 持续存在(稳态)。

Green's function for $y'' + y = f$$y'' + y = f$ 的格林函数

$G(t, au) = \sin(t - au)$, so $y_p = \int_{t_0}^{t}\sin(t- au) f( au)\,d au$.

$G(t, au) = \sin(t - au)$,故 $y_p = \int_{t_0}^{t}\sin(t- au) f( au)\,d au$。

Unit Quiz

The general solution of $a y'' + b y' + c y = g(t)$ is:$a y'' + b y' + c y = g(t)$ 的通解是:
Q.1
$y_p - y_h$
$y_h$ only
$y_h + y_p$
$y_p$ only

Correct. The general solution is the complementary solution plus a particular solution.

正确。通解等于齐次解加上一个特解。

By linearity the answer is $y_h + y_p$, the homogeneous family shifted by one particular solution.

由线性性,答案为 $y_h + y_p$,即齐次解族经一个特解平移的结果。

For $y'' + y = 3e^{2t}$ (with $\pm i$ the homogeneous roots), a particular solution is:对于 $y'' + y = 3e^{2t}$(齐次方程的根为 $\pm i$),一个特解是:
Q.2
$\tfrac35 e^{2t}$
$3 e^{2t}$
$\tfrac13 e^{2t}$
$t e^{2t}$

Correct. Try $A e^{2t}$: $(4A + A)e^{2t} = 5A e^{2t} = 3e^{2t}$, so $A = frac35$.

正确。令 $y_p = A e^{2t}$:$(4A + A)e^{2t} = 5A e^{2t} = 3e^{2t}$,故 $A = frac35$。

Substitute $A e^{2t}$: $y_p'' + y_p = 5A e^{2t} = 3 e^{2t}$, giving $A = frac35$.

代入 $A e^{2t}$:$y_p'' + y_p = 5A e^{2t} = 3 e^{2t}$,得 $A = frac35$。

For $y'' + 4y = \sin 2t$ with $y_h = C_1\cos 2t + C_2\sin 2t$, the correct trial $y_p$ is:对于 $y'' + 4y = \sin 2t$,其中 $y_h = C_1\cos 2t + C_2\sin 2t$,正确的试解 $y_p$ 是:
Q.3
$A\cos 2t + B\sin 2t$
$A\sin 2t$
$A e^{2t}$
$t(A\cos 2t + B\sin 2t)$

Correct. The forcing matches a homogeneous solution, so apply the modification rule and multiply by $t$.

强迫项与某个齐次解吻合,故应用修正规则乘以 $t$。

Since $\sin 2t$ already appears in $y_h$, the plain trial fails. Multiply by $t$: $t(A\cos 2t + B\sin 2t)$.

由于 $\sin 2t$ 已在 $y_h$ 中,普通试解失效。乘以 $t$。

For variation of parameters on $y'' + p y' + q y = f$ with solutions $y_1, y_2$, we have $u_1' =$对参数变易法($y'' + p y' + q y = f$,解为 $y_1, y_2$),有 $u_1' =$
Q.4
$y_1 f / W$
$-y_2 f / W$
$y_2 f / W$
$-y_1 f / W$

Correct. Cramer's rule gives $u_1' = -y_2 f / W$ and $u_2' = y_1 f / W$.

正确。克莱默法则给出 $u_1' = -y_2 f / W$ 和 $u_2' = y_1 f / W$。

Solving the $2 imes 2$ system yields $u_1' = -y_2 f / W$.

求解 $2 imes 2$ 方程组得 $u_1' = -y_2 f / W$。

Which forcing $g(t)$ cannot be handled by undetermined coefficients?哪种强迫项 $g(t)$ 不能用待定系数法处理?
Q.5
$t^3$
$e^{t}\cos t$
$\ln t$
$4\sin 3t$

Correct. $\ln t$ is outside the polynomial, exponential, and sinusoidal families, so use variation of parameters.

正确。$\ln t$ 不属于多项式、指数函数或正弦/余弦函数族。

The other three have standard trial forms. Only $\ln t$ requires variation of parameters.

其他三种有标准试解形式,只有 $\ln t$ 需要参数变易法。

In a damped driven system, after the transient decays the remaining response is the:在阻尼受迫系统中,瞬态衰减后剩余的响应为:
Q.6
steady state $y_p$稳态 $y_p$
complementary solution $y_h$齐次解 $y_h$
Wronskian朗斯基行列式
characteristic equation特征方程

Correct. The particular solution persists as the steady-state response once $y_h$ has died out.

正确。$y_h$ 衰减后,特解作为稳态响应持续存在。

The transient $y_h$ decays; what remains is the steady-state particular solution $y_p$.

瞬态 $y_h$ 衰减,剩余的是稳态特解 $y_p$。

Readiness Checklist

Tap each item you can do without notes. 0 / 8 mastered

点击你能在不看笔记的情况下完成的每一项。