Chap 4 老師講義

2016年9月19日 下午 06:32

微分方程式
Chapter 4
管傑雄
台大電機系
Chapter 4 Differential Equations of Higher
Order
本章將以求解線性的高階微分方程為主。其中的解法首先介紹降階方式(降至一階
可解為主)，其次是分別介紹高階Homogeneous Equation的求解(yc)及Particular
Solution(yp)的找法，最後的General Solution為 y = yc + yp，也就是將前面的一階線
性微分方程的求解方法擴大應用，過程如下表。值得注意的是，integration factor
的求解方法對高階方程式變得過於複雜而不適用。

How to find yc? How to find yp?

(1) Equ's with constant coeff's (1) Undetermined Coeff's
(2) Euler-Cauchy Equ's (a) Superposition Approach
(3) Reduction of Order (b) Annihilator Approach
(2) Variation of Parameters

Ch4 第 1 頁
4.1 Preliminary Theory: Linear Equations
4.1.1 Initial-Value and Boundary-Value Problem

Initial-Value Problem
Solve:

Subject to: y(x0) = y0, y'(x0) = y1, ... , yn-1(x0) = yn-1.

The values at x = x0 are called initial conditions.

[Exitence of a Unique Solution]

Let an(x), an-1(x), ... , a1(x), a0(x), and g(x) be continuous on an interval I and let an(x)  0
for every x in this interval. If x = x0 is any point in this interval, then a solution y(x) of
the initial-value problem exists on the interval and is unique.
此定理的證明在課本中是省略的，事實上由微分方程式相對應的家族曲線，即所謂的
通解，就可以看出微分方程的參數唯一解區及其相對應的solution interval。此interval即
為此處所指的interval。在後續中將利用代數幾何的聯立方程組的相關概念證明an(x)  0
的 interval就是solution interval。

Example: x2y" - 2xy' + 2y = 6, the general solution (family curves) is y = c1x + c2x2 + 3.
(1) Unique solution
In either (-. 0) or (0, ) but not both, for any initial conditions, y(x0) = y0 and y'(x0) = y1, we
are always able to find a unique solutions c1 and c2 since

in that range.
For example, y(1) = 5 and y’(1) = 3, the unique solution is y = x + x2 + 3 on (0, ).
(2) 無限多解

If x0 happens to be zero, e. g.y(0) = 3 and y'(0) = A(Arbitrary), on (-,), you may find the
function y = cx2 + Ax + 3 is the solution for any choice of the parameter c.

(3) 無解
But for y(0)  3, there is no solution at all since the determinant is zero and we cannot find
the solutions c1 and c2.
(4) 每個 solution interval 均必須經必須給定一組initial conditions才能唯一決定一個全區
解，否則仍是多解的。
In the case of y(1) = 5 and y’(1) = 3, we use the unique solution to find y(0) = 3 and y’(0) =
1. If we extend the solution range to (-,), the general solution becomes

Hence, the solution number is infinite.

Boundary-Value Problem (BVP)

Solve:

Ch4 第 2 頁
 Solve:

Subject to: y(a) = y0, y(b) = y1 (Boundary Conditions)

Note 1: Other pairs of boundary conditions
y'(a) = y0, y(b) = y1
y(a) = y0, y'(b) = y1
y'(a) = y0, y'(b) = y1
Note 2: General boundary conditions
1y(a) + 1y'(a) = 1
2y(b) + 2y'(b) = 2

Example: A boundary-value problem may have many, one, or no solutions. For example,
y" + 16y = 0 has a general solution of y = c1cos4t + c2sin4t:
(a) with the boundary conditions of y(0) = 0, y(/2) = 1 has no solutions;
(b) with the boundary conditions of y(0) = 0, y(/8) = 0 has one solution; and
(c) with the boundary conditions of y(0) = 0, y(/2) = 0 has infinite solutions.

4.1.2 Homogeneous Equations

[Homogeneous Equation]

[Nonhomogeneous Equation]

where g(x) is not identically zero.

Note: To avoid needless repetition, the following points are assumed in the remainder
1. the coefficients ai(x), i = 0, 1, ...... , n, are continuuous
2. the right-hand member g(x) is continuous
3. an(x)  0 for every x in the interval

Differential Operator: dy/dx = Dy

[nth-order Differential Operator]: L = an(x)Dn + an-1(x)Dn-1 + .....+ a1(x)D + a0(x)
[Linear Operator]: if L{f(x) + g(x)} = L(f(x)) + L(g(x)) or L{f(x)} = L(f(x)) and
L{f(x) + g(x)} = L(f(x)) + L(g(x))
Differential Equation: L(y) = 0 or L(y) = g(x)

[Superposition Principle -- Homogeneous Equation]

Let y1, y2, ... , yk be solutions of the homogeneous linear nth-order differential equation
on an interval I. Then the linear combination

where the ci, i = 1, 2, ... , k, are arbitrary constants, is also a solution on the interval.

[Superposition Principle for the input – Nonhomogeneous case]

Let

, ... ,

be k particular solutions of the nonhomogeneous linear nth-order differential equation on an

Ch4 第 3 頁
 be k particular solutions of the nonhomogeneous linear nth-order differential equation on an
interval I corresponding in turn to k distinct functions g1, g2, ... , gk. Then

is a particular solution of

[Corollaries]
(A) A constant multiple y = c1y1(x) of a solution y1(x) of a homogeneous linear differential
equation is also a solution.
(B) A homogeneous linear differential equation always possesses the trivial solution y =
0.(線性空間的零向量)

[functions and solutions]

ㄧ個函數 f(x) 的定義域 function interval 是依照函數本身的型式決定的。但 f(x)
若視為 solution，則其相對的 solution interval 是決定於那一個 DE。不同 DE 會
有不同的 solution interval。一般而言，對同一個 f(x) 之 function interval 是包含
solution interval。

Example: f(x) = x
Function interval: - < x < 
Solution interval for y = c1x + c2ex : either - < x < 1 or 1 < x < 

Function interval 是可包含 Solution interval。我們必須先定義相容性，亦即在各自定

義的 interval 內，一函數(解)可由其它函數(解)來代表。然後由相容性的否定來看獨立
性的意義。

※相容性及獨立性
[Definition--Linear Dependence for functions and solutions]
A set of functions, f1(x), f2(x), ... , fn(x), is said to be linear dependent on an interval I
(which may not be the same for functions and solutions) if and only if there exist
constants, c1, c2, ... , cn, not all zero, such that

for every x in the interval. 有不全為零的 ci 值通解。

[相容性的另類解釋]
1. 換言之，在整個interval內，存在一個函數可用其它函數來代表：

，其中的ck  0
2. 若定義家族曲線

，則 fi(x) 的相容性代表參數並非全是 effective。

[Linear Independence for functions]
將原來相容性的定義取否定，其中有兩個地方可分別取否定。
1. Condition A, ~ not all zero將原相容性的非全為零取否定，即沒有不全為零的 ci 值通
解。

Ch4 第 4 頁
解。
A set of functions, f1(x), f2(x), ... , fn(x), is said to be linear independent on an interval
I(which may not be the same for functions and solutions) if it is not linear dependent on
the interval. In other words,
if there exists one set of the constants which such that

for every x in the interval are c1 = c2 = c3 = ... = cn = 0.

2. Condition B, ~ every將原來相容性的每一個取否定，此情形即說明x interval可至少

被分割成兩個以上的區段，相鄰區段間有 Singular Point 分割。至少有一個區段或
Singular Point 有一組不全為零的c值，但沒有整個 interval 均適用的不全為零的通
解。
If there exists one set of constants, c1, c2, ... , cn, not all zero, which such that

for only some x but not every x in the interval, then the n functions are linearly independ

[Linear Independence for Solutions]

Solution interval 是不可包括任何 Singular Point，故上述兩條件只有Condition A適
用於 Solution 的情形。

[Singular Point 的定義]

由線性家族曲線

推得 nth-order 線性微分方程式時，我們定義 n 個參數唯一解區為

相對地，W = 0 為 Singular Point 的所在處。在此已相當於假設該 n 個函數從零次至

(n – 1) 次導函數均是連續的。針對這倜假設不存在的 x 值而言，也會使得 W(x) 為
Uncertain，意即不連續或趨向正負無窮大。這些點當然不在唯一解區，也應該歸類於
分割唯一解區的 Singular Point。換言之，Singular Point 的定義應擴大為 W = 0 或
Uncertain 的 x 值。

[c值的通解、多重組解及唯一解]
基於上述的定義，我們可採用通解及c是否為多重組解的觀念來進行判別函數或
解的獨立性判斷。所謂通解是指適用於整個函數或解interval；若僅適用部份
interval或單獨點則不謂之通解。對於c是否有多重組解，我們有

1st derivative

2nd derivative

.......
(n-1)th derivative

for every x on the interval I (除了使得 W 為 Uncertain 的 Singular Point 以外),

c有多重組解

Ch4 第 5 頁
 c僅有全為零的唯一解

W0
對於如下的statements
There exist constants, c1, c2, ... , cn, not all zero, such that

for every x in the interval.(c有非全為零的通解)c有多重組解

W=0
注意Þ是不可反過來，因為在每一點有多重組解並非意謂蓍有通解，如c1x + c2|x| = 0。

[圖解相容性及獨立性]
根據相容性及獨立性的定義、Singular Point 的定義和 W 值判斷 c 值的多重解及
唯一解，我們可歸納成如下的圖解。
1. function interval 是可包涵 Singular Point 而 Solution Interval 則否，故 function interval 是
可大於 solution interval。
2. Independent function 是根據 Singular Point 來進行分類。其中 Condition A 是不包涵
Singular Point，而 Condition B 是包涵 W = 0 及 W Uncertainty 等兩類 Singular Point。
3. 由於 W = 0 的 Singular Point 已意味著在該點已有不全為零的 c 值解，故在其分割的區
段上，可以是全為零或不全為零的 c 值，即符合 Condition B 的要求。
4. 相對地，在 W Uncertainty 的 Singular Point 的情形，在分割區段上則需至少有一不全
為零的 C 值出現，才能符合 Condition B 的要求。
5. 不同區段或點若有不全為零的 c 值，它們個別是不同的組合。

Ch4 第 6 頁
 [其它注意事項]
1. Since the solution interval is a subset of the function interval for the same form of the
solution and function, the above statement indicates that dependent functions must be
the dependent solutions. 函數的相容性條件比解函數的嚴格，相對地，獨立性條件將變
得較寬大，這在底下會說明。
2. The dependence is much related with the (function or solution) interval I.
3. Another equivalent linear independence if and only if condition
The only constants for which

for every x in the interval are c1 = c2 = c3 = ... = cn = 0 (只有全為零的 ci 值通解)。used in

the textbook)
4. The independent solutions must be the independent functions and some dependent
solutions may be independent functions. 此 independent function 就是 singular
solution。Not true for that the independent functions must be the independent
solutions(Condition B).
5.

Ch4 第 7 頁
 6.

[Example]
Linear dependence functions:
sin2x and sinx cosx
cos2x, sin2x, sec2x, tan2x (since cos2x + sin2x - sec2x + tan2x = 0)

Linear independence functions:

Example 1：y = x and y = |x| 的相容性及獨立性並找出在

的線性微分方程及其通解。
Sokution
(1) 相容性：dependent solutions on (-, 0) or

.
(2) 獨立性：若考慮c1x + c2|x| = 0的c解而言，在(-, 0)內的解為(c1, c2) = (0, 0)或(c, c)；但
在(0, )內的解則為(c1, c2) = (0, 0)或(c, -c)；在原點時則為隨意。因此，就

而言，我們有三組不全為零的c解；分別在(-, 0)、(0, )及原點符合Condition B，或者

是通解只有 (0, 0)，因此x and |x|互為獨立。
(3) 線微分方程式的通解為 y = cx on (-, 0) or

亦即 xy’ – y = 0；而且 y = |x|也是這個微分方程式的一 singular solution。

(4) DE 的通解：考慮xy’ – y = 0的逯解在

，若符合y(0)是連續的，則其通解為

其中i為常數。若符合y(0)及y’(0)均是連續的，則其通解為

，c is arbitrary。由本例得知，在singular point的邊界條件影響結果甚鉅，如何求知該條
件是一重要課題。

Ch4 第 8 頁
 Example 2：考慮 y = x 與 y1 的相容性及獨立性並找出在

的線性微分方程及其通解。其中的

Solution
(1) 相容性：independent solutions on (-, 0) but dependent solution on

(2) 獨立性：就 y = x 與 y1 的 c1x + c2y1 = 0 而言，在(-, 0)內的解為(c1, c2) = (0, 0)；但在(0,

)內的解則為(c1, c2) = (0, 0)或(c, -c)；在原點時則為隨意。因此，就

而言符合Condition B，因此x and y1互為獨立。.

(3) 線性 DE：由於x及x2為兩個獨立解，所以相對應的通解為y = c1x + c2x2 on (-, 0) 或 (0,
)；而微分方程為x2y" - 2xy' + 2y = 0而y1也是一解。
DE 通解：考慮在

的通解，若符合y(0)是連續的，則其通解為

(4)

其中i為常數。

[Wronskian Theorem]
[W 值的定義]
Suppose f1(x), f2(x), ... , fn(x), possesses at least n-1 derivatives. The determinant
W(f1(x), f2(x), ... ,fn(x)) =

is defined as the Wronskian of functions.

[Independence and Dependence For functions]

[Corollary I]
If f1(x), f2(x), ... , fn(x), possesses at least n-1 derivatives and are linearly dependent on
I, then
W(f1(x), f2(x), ... ,fn(x)) = 0
for every x in the interval.
Proof:
Since they are linear dependent, we have

true for every x on the interval I and the ci is not all zeros. The derivatives of the above zero
function are

1st derivative

2nd derivative

.......
(n-1)th derivative

Ch4 第 9 頁
This becomes a set of linear equations with variables of ci. Because the solution is not unique
(not all zeros, and all zeros), we have
W(f1(x), f2(x), ... ,fn(x)) =

=0
for every x on the interval I.

[Corollary II]
因為 AB is equivalent to ~B~A，故由 Corollary I 得到
Condition A
If W(f1(x), f2(x), ... ,fn(x))  0 for every x in the interval,
Or
Condition B
If W(f1(x), f2(x), ... ,fn(x)) = 0 for some x instead of every x in the interval,
then the functions are linear independent on the interval.
此結果與前面 Independent function 定義一致。課本所描述的定理雖沒有錯誤，但其
否定的範圍不夠完整。

Example 1(Corollary I): f1(x) = sin2x and f2(x) = 1 - cos2x are linearly dependent on

.
proof: We find
W(f1(x), f2(x)) =

= 0 for every x

Example 2(Corollary IIA): f1(x) =

and f2(x) =

, m1  m2.
Proof: We find
W(f1(x), f2(x)) =

for every real value of x. Hence they are independent.

Example 3(Corollary IIB): f1(x) =

and f2(x) =

.
Proof: We find
W(f1(x), f2(x)) =

Ch4 第 10 頁
if x  0. They are independent for all real numbers even though W = 0 for x = 0.

Example 4: It is not true that if W(f1(x), f2(x), ... ,fn(x)) = 0 for every x in the interval,
then f1(x), f2(x), ... , fn(x) are linearly dependent on I. The example is f1(x) = x3 and f2(x) =
| x3 |. We do have W( f1(x), f2(x)) = 0 for every x on (-, ) but they are independent
functions.

Note: It should also be noted especially that they cannot be solutions of a second-order
linear differential equation although they are independent functions. The reason is that
they are composed of dependent solutions and a singular point x = 0.(請參考前面的圖,
在independent function是可能有W=0情形)

[Independence and Dependence For solutions]

[Corollary III]
Linearly Independent Solutions
Let y1, y2, ... , yn be n solutions of the homogeneous linear nth-order differential
equation on an interval I. Then the set of solutions is linearly independent on I if and
only if (

)
W(y1, y2, ... , yn)  0
for every x in the interval.
仔細的證明提供參考
Proof: The necessity (

) i.e., to find the homogeneous differential equation

Consider y(x) and their derivatives
y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x)
y’(x) = c1y1'(x) + c2y2'(x) + ... + cnyn'(x)
........
y(n-1)(x) = c1y1(n-1)(x) + c2y2(n-1)(x) + ... + cnyn(n-1)(x)
Since W(y1, y2, ... , yn)  0 for every x in the interval, we can find the unique solution ci and
substitute them into the nth derivative of y(x). Finally we can derive the homogeneous
equation.

The sufficiency(

) is proved as followings.
"If y1, y2, ... , yn are linearly independent, W(y1, y2, ... , yn)  0 for every x in I."
That is we are going to prove that if W(y1, y2, ... , yn) = 0 for some x0, then the solutions are
linearly dependent.
If W(y1, y2, ... , yn) = 0 for some x0, then we consider the system of the equations for ci
c1y1(x0) + c2y2(x0) + ... + cnyn(x0) = 0
c1y1'(x0) + c2y2'(x0) + ... + cnyn'(x0) = 0
........
c1y1(n-1)(x0) + c2y2(n-1)(x0) + ... + cnyn(n-1)(x0) = 0
Since W = 0, there exist solutions ci, i = 1, ... , n which are not all zero. If we define a
function y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x), then the above conditions indicate that initial

Ch4 第 11 頁
 function y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x), then the above conditions indicate that initial
conditions are y(x0) = 0, y'(x0) = 0, ... , y(n-1)(x0) = 0. In addition, due to superposition principle,
y(x) is also a solution of the homogeneous nth-order linear differential equation. We know
that y(x) = 0 is the unique solution which satisfies the initial conditions. This indicates that
y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x) = 0 for every x in I where. ci, i = 1, ... , n are not all zero.
This implies that yi, i = 1, ... , n, are not independent..

[Choice from n functions]

If we choose k ones, say y1, y2, ... , yk, from the n functions where k < n, the Wronskian
W(y1, y2, ... , yk) may be 0 for some point and their associated linear homogeneous kth-
order differential equation may have a different solution interval I.

[線性空間及線性方程式的關係]
線性空間可分割成多個次區段，而且每個次區段均可找到一組獨立解函數；使得
該次區段的解為該組獨立解函數的線性組合，亦即該次區段可用一Homogeneous
微分方程式來代表。兩相次區段的解函數間必須符合一個線性的邊界條件，這可
由微分方程求到，正如同電磁學所提到的邊界條件。每一次區段的微分方程式是
可以不相同的；但若相同的話，分割該兩個相鄰次區段者必為某singular points
(Condition B)。甚至整個線性空間是可用一個微分方程式來代表，但有數個
singular points將其分割成數個次區段。這意味著該微分方程式的解是包括family
of solutions及Singular solutions，而這些singular solutions也可做線性組合展開該
空間，使其變得更大更複雜。

[n-parameter Family Curves  nth-order linear homogeneous equation]

1. 線性空間的區段分割概念
Let y1, y2, ... , yn be n solutions of the homogeneous linear nth-order differential
equation. Then the set of solutions is linearly independent on I if and only if
W(y1, y2, ... , yn)  0
for every x in the interval I.
如何找 Solution Interval I？即先找到W(y1, y2, ... , yn)  0或不連續(包括有限及無
限，其中前者在nth derivative也會趨向無窮大)的點 xi(i = 1, 2, 3, …)，這些點就是
分割區段的singular points；而在相鄰兩singular points間或正負無窮遠與singular
point間的次區段，即可是interval I，該次區段由左至右依次定義為 I j (j = 1, 2, 3,
…)。
2. 線性方程式的找法
由於在interval I中，W(y1, y2, ... , yn)  0 for every point，故有
y=

qst derivative y’ =

.......
(n-1)th derivative y(n-1) =

由上面連立方程組可解得

(這個式子必須對每一個x均成立)
代入y(n) =

Ch4 第 12 頁
 得到

其中的W0(x)即為前面定義的an(x)。所以singular points由兩者看出是一致的(請特別注
意如何找singular points)。
3. 整個區段的通解

其中的ci,j (i = 1, 2, …, n)是 Ij 範園內的參數且ci,j與ci,j+1必須符合在x = xj的邊界條

件。

Example: Consider three functions, f0(x) = 1, f1(x) = x, and f2(x) = x2. They are linearly
independent since

for every real x

and their associated differential equation is y"' = 0.
If we choose two functions, say f1(x) = x, and f2(x) = x2, the Wronskian

under the condition x  0. They are also the two independent solutions of the following
differential equation
x2y" -2xy' + 2y = 0.
Since the equation is 2nd-order and we have x  0. Therefore, their Wronskian is absolutely
nonzero in (-, 0) or (0, ). Why not include zero? If we consider the solution interval would
be (-, ), the IVP may have infinite solutions or no solution. For example, if the initial
condition is y(0) = 0 and y’(0) = 0. Since y(x) = c1x + c2x2, the initial condition gives c1 = 0
and c2 is arbitrary. Therefore, the solution number is infinite. On another hand, if the initial
condition is y(1) = 1 and y’(1) = 2, then c1 = 0 and c2 = 1 for the unique solution in the range
(0, But the solution also has y(0) = 0 and y’(0) = 0. We can not expect the solution have
to follow y(x) = x2 in the range (-, ). The solution should be

.
Hence the solution number is also infinite.
If the equation becomes
x2y" -2xy' + 2y = x,
then the general solution is y(x) = c1x + c2x2 - xln

. That is, we can not find any solution which is true in the range (-, ).
Therefore, x = 0 is a singular point. If it is included in the solution range, then the
solution will not become unique.

[nth-order Linear Equation  n-parameter family curves] & [Solution method for
linear differential equations]
由於任何線性空間可分割成多次區段的獨立解函數。故線性微分方程式的求解即是在
solution interval內求出IVP的唯一解問題。

Ch4 第 13 頁
對於nonhomogeneous case，即如下所述
Solve:

Subject to: y(x0) = y0, y'(x0) = y1, ... , yn-1(x0) = yn-1.

Logic Procedure to find the solutions of the nonhomogeneous linear differential

equations
1. Existence and Uniqueness of the solutions for the nonhomogeneous /homogeneous
linear differential equations
2. Existence of a fundamental set for homogeneous linear differential equation
3. A general solution of the homogeneous linear differential equation can be expressed
by a linear combination of the fundamental set.
4. A general solution of the nonhomogeneous linear differential equation can be
expressed by a linear combination of the fundamental set plus a particular solution.

[整體簡昜說明]
由an(x) 或 W0(x) 找出唯一解區。由於 W0 並不含有與 y 及其 derivative 相關的變
數，意味初始條件有關 y 及其導函數是可以任意給定的。因此可指定如下的 n 組
initial condition
y(t) = 0, y'(t) = 0, ... , y(k-1)(t) = 0, y(k)(t) = 1, y(k+1)(t) = 0, ... , y(n-1)(t) = 0
此虛的 k = 0, 1, ... ,(n-1)，而 t 為唯一解區的一點。如此相對應可得到 n 個解。這 n 個
解是獨立的，可形成一組fundamental set，此乃因為 W(t) = 1，即可推演到 W(x)  0 for
any x in the solution interval。而 Homogenous Equation 的通解為
yc = c1y1(x) + c2y2(x) + ... + cnyn(x)
因為只要給定任意的 initial condition，由於W(x)  0所以 c 值可唯一決定。同理
nonhomogeneous equation 的通解為
y = c1y1(x) + c2y2(x) + ... + cnyn(x) + yp(x)
此處的 yp(x) 是個任意的 particular solution，這是因為 y(x) - yp(x) 是 Homogenous
Equation 的 Solution。

[進一步說明]
&Logic Procedure 1 is supposed to have been proved. (由家族曲線的微分方程參數唯一
解區得到，即由an(x) 或 W0(x) 找出 Singular Point 及唯一解區。在唯一解區每一點均
不為零)另外，由於 W0 並不含有與 y 及其 derivative 相關的變數，意味初始條件有關
y 及其導函數是可以任意給定的。

&Logic Procedure 2(獨立解集合)

Fundamental set of solutions
Any set y1, y2, ... , yn of n linearly independent solutions of the homogeneous linear nth-
order differential equation on an interval I is said to be a fundamental set of solutions on
the interval.

Existence of a fundamental set

There exists a fundamental set of the solutions for the homogeneous linear nth-order
differential equation on an interval I.
Proof: We can define n solutions among which the kth one yk(x) is defined as
y(t) = 0, y'(t) = 0, ... , y(k-1)(t) = 0, y(k)(t) = 1, y(k+1)(t) = 0, ... , y(n-1)(t) = 0
where k = 0, 1, ... ,(n-1). From the existence and uniqueness of the linear differential
equation, we know that we get exactly n solutions. Now we are going to confirm the
independence of the n solutions. Suppose they are linearly dependent. We can find n
Ch4 第 14 頁
independence of the n solutions. Suppose they are linearly dependent. We can find n
constants, Ci, i= 0, ... , (n-1), of not all zero such that
Y = C0y0(x) + C1y1(x) + ... + Cn-1yn-1(x) = 0 for every x in I.
At x = t, from Y(t), we find C0 = 0, and from Y(k)(t), we find Ck = 0. Hence, all Ck = 0, k =
0, ..., (n-1). This contradicts to the assumption of linear dependence. Therefore, the n
solutions are linearly independent.

Note: The fundamental set is not unique.

&Logic Procedure 3
Let y1, y2, ... , yn be a fundamental set of solutions of the homogeneous linear nth-order
differential equation on an interval I. Then for any solution Y(x) of the equation on I,
constants C1, C2, ... , Cn, can be found so that
Y = C1y1(x) + C2y2(x) + ... + Cnyn(x)
Proof: Suppose Y(x) satisfy the initial conditions of Y(t) = k0, Y'(t) = k1, ... , Y(n-1)(t) = kn-1,
where t is a point of the interval. We consider a function G(x) which is defined as
G(x) = C1y1(x) + C2y2(x) + ... + Cnyn(x)
and also satisfies the initial conditions. That is
C1y1(t) + C2y2(t) + ... + Cnyn(t) = k0
C1y1'(t) + C2y2'(t) + ... + Cnyn'(t) = k1
..........
C1y1(n-1)(t) + C2y2(n-1)(t) + ... + Cnyn(n-1)(t) = kn-1
Since W(y1, y2, ... , yn)  0 at x = t, we can find a unique set of solutions of Ci, i = 1, ... , n.
Therefore, G(x) is also unique. In addition, Due to the existence and uniqueness of the linear
differential equation, we conclude Y(x) = G(x).

General solution -- Homogeneous Equation

Let y1, y2, ... , yn be a fundamental set of solutions of the homogeneous linear nth-order
differential equation on an interval I. The general solution of the equation on the
interval is defined to be
y = c1y1(x) + c2y2(x) + ... + cnyn(x)
where ci, i = 1, .. , n, are arbitrary constants.

Example y1 = ex, y2 = e2x, and y3 = e3x, satisfy the third order equation

Since

for every x on

, we conclude that the general solution is

y = c1ex + c2e2x +c3e3x

4.1.3 Nonhomogeneous Equations

&Logic Procedure 4
Let y1, y2, ... , yk be solutions of the homogeneous linear nth-order differential equation
on an interval I and let yp be any particular solution of the corresponding
nonhomogeneous equation on the same interval. Then
y = c1y1(x) + c2y2(x) + ... + ckyk(x) + yp(x)
is also a solution of the nonhomogeneous equation on the interval for any constant ci.

Ch4 第 15 頁
 Let yp be a given solution of the nonhomogeneous linear nth-order differential equation
on an interval I and let y1, y2, ... , yn be a fundamental set of solutions of the associated
homogeneous equation on the interval. Then for any solution Y(x) of the
nonhomogeneous equation on I, constants C1, C2, C3, ... , Cn can be found so that
Y = C1y1(x) + C2y2(x) + ... + Cnyn(x) + yp(x)
Proof: If we define a function u(x) = Y(x) -yp(x), then u(x) follows the homogeneous
differential equation. For any solution u(x), C1, C2, C3, ... , Cn can be found so that
u = C1y1(x) + C2y2(x) + ... + Cnyn(x)
Therefore, Y = C1y1(x) + C2y2(x) + ... + Cnyn(x) + yp(x).

General solution -- Nonhomogeneous Equation

Let yp be a given solution of the nonhomogeneous linear nth-order differential equation
on an interval I and let

denote the general solution of the associated homogeneous equation on the interval. The
general solution of the nonhomogeneous equation on the interval is defined to be

= yc(x) + yp(x)
where yc(x) is called the complementary function for the nonhomogeneous equation.

Example:

is a particular solution of

.
The complementary function of the equation is
yc = c1ex + c2e2x + c3e3x
Hence, the general solution is

[Superposition Principle -- Nonhomogeneous Equation]

Let

, ... ,

be k particular solutions of the nonhomogeneous linear nth-order differential equation on an

interval I corresponding in turn to k distinct functions g1, g2, ... , gk. Then

is a particular solution of

Note: If

, then yp is a solution of

Ch4 第 16 頁
Remark:

is said to be nth-order linear system. The function g(t) is called the input function, forcing
function, or excitation function. A solution y(t) is called the output or response of the
system. The n functions, y(t), y'(t), .... , yn-1(t) are called the state variables of the system.

4.2 Reduction of Order

Take a second-order linear equation as example
y"(x) + P(x) y'(x) + Q(x) y(x) = 0
Let us suppose that y1(x) is a known solution of the above equation. We define y(x) = u(x)
y1(x) and substitute it into the equation to get a differential equation of u:
y"(x) + Py' + Qy = u[y1" + Py1' + Qy1] + y1u" + (2y1' + Py1)u' = 0
That is,
y1u" + (2y1' + Py1)u' = 0
If we let w = u', then
y1w' + (2y1' + Py1)w = 0
is a first-order differential equation. This method is called reduction of order. The variables
are separable and we find

.
Choose c1 = 1 and c2 = 0, and the second solution of y is

.
Now, y1(x) and y2(x) are linearly independent since

Ch4 第 17 頁
is not zero on any interval on which y1(x) is not zero.

Example: y1 = x2 is a solution of
x2y" - 3xy' + 4y = 0.
Find the general solution on the interval (0, ).
Solution: Since

,
the second solution is

The general solution is

y = c1x2 + c2x2lnx

Note: Similarly, we may reduce a third-order equation into a second-order one with the
same method.

[昇階的方式]
已知某 Homogeneous linear equation Ly = 0，若欲增加一線性組合解 y = cf(x)，亦
即假設原解若為 y0 =

，則新解 y = y0 + cf(x)依照 Ly = 0 的找方程式方法為

y - cf(x) =

1st derivative y’ – cf’(x) =

.......
(n-1)th derivative y(n-1) – cf(n-1)(x) =

解得 ci( i = 1,…, n)代入y(n) – cf(n)(x) =

得到
L(y – cf(x)) = 0
這是因為 y 與 cf(x) 是處於相同的狀況。故有

再對 x 進行微分得到
(Ly)’L(f(x))-Ly[L(f(x))]’ = 0

這即是所求 (n+1)th-order linear DE。對於 Nonhomogeneous Equation，同理可推得

Ly = cL[f(x)] + g(x)
解得 c 再對 x 微分得到
(Ly – g(x))’L(f(x)) - (Ly – g(x))[L(f(x))]’ = 0

Ch4 第 18 頁
 (Ly – g(x))’L(f(x)) - (Ly – g(x))[L(f(x))]’ = 0

Example: 分別求解兩個一階方程式 y’ – y = 0 及 xy’ – y = 0，然後求出相對於此兩獧立

解線性組合的二階方程式。接著考慮該二階方程式與 xy’ – y = 0 所形成的聯立方程
組，請求出相對應的解。
Solution
利用變數分離的方式可求出 y’ – y = 0 及 xy’ – y = 0 的解分別為 y = cex 及 y = cx。
利用昇階公式於 y’ – y = 0 可得

整理後為

考慮聯立方程組

先昇階第二式

第一式減去第二式得到

即 xy’ – y = 0 故解為 y = cx。

4.3 Homogeneous Linear Equations with Constant

Coefficients
解區段是(-, ¥)的範例。

where ai, i = 0, ... , n, are constants.

Take the second-order one as example:
ay" + by' + cy = 0
Use y = emx as a trial solution, we can find that m is given by
am2 + bm + c = 0
This equation is called the auxiliary equation or characteristic equation of the differential
equation.

Case I Distinct Real Roots

The auxiliary equation has unequal roots, m1 and m2. We find two solutions,
y1 = em x and y2 = em x
1 2

and they are linearly independent. Hence, the general solution is

Case II Repeated Real Roots

The auxiliary equation has the same roots, i.e., m1 = m2 = - b/2a. Since y1 = exp(- bx/2a),
the second solution can be given by

Ch4 第 19 頁
 the second solution can be given by

Note: 將 y = xexp(-bx/(2a)) 代入 ay” + by’ + cy = 0 可得到 b2 – 4ac = 0。

The general solution is

Case III Conjugate Complex Roots

Two complex solutions:  + j and  - j. The two linearly independent solutions can
be given by

and

or

and

Note:

Higher-Order Equations

with the auxiliary or characteristic equation of

anmn + an-1mn-1 + ... + a1m + a0 = 0
If all the roots are distinct, then the general solution is

When m1 is a root of multiplicity k of an nth-degree auxiliary equation (that is, k roots are
equal to m1), the it can be shown that the general solution must contain the linear combination

Example 1: Solve 3y''' + 5y" +10y' - 4y = 0

Solution: The auxiliary equation is
3m3 + 5m2 + 10m - 4 = 0
3m3 - m2 + 6m2 - 2m + 12m - 4 = 0
(3m - 1)(m2 + 2m + 4) = 0

, or

Hence, the general solution is

Ch4 第 20 頁
Example 2: Solve

Solution: The auxiliary equation is

m4 + 2m2 + 1 = 0
with mutiplicity of 2..
Hence, the general solution is
y = c1cosx + c2sinx + c3xcosx + c4xsinx

4.4 Undetermined Coefficients -- Superposition Approach

To obtain the general solution of a nonhomogeneous differential equation with constant
coefficients we must do two things:
1. to find the complementary function yc
2. to find any particular function yp of the nonhomogeneous equation.

The method of undetermined coefficients presented in this section is limited to

nonhomogeneous linear equations that have constant coefficients, and where g(x) is a
constant, polynomial function, exponential function, sine, cosine, or finite sums and
products of these functions. That is, g(x) is a linear combination of the functions of the
type
k(constant), xn, xnex, xnexcosx, and xnexsinx.
These functions has the remarkable property that derivatives of their sums and products
are again sums and products of constants, polynomials, exponentials, sines, and cosines.
Therefore, it seems reasonable to assume that yp has the same form as g(x).

Case I No function in the assumed particular solution duplicates a function in the

complementary function.

Example 1: Solve y" + 4y' - 2y = 2x2 - 3x + 6

Solution: The auxiliary equation is
m2 + 4m - 2 = 0

Hence, the complemmentary function is

For the particular solution, the trial forrm is

yp = Ax2 + Bx + C
Then,
yp" + 4yp' - 2yp = -2Ax2 + (8A-2B)x + (2A + 4B -2C) = 2x2 - 3x + 6
This leads to A = -1, B = -5/2, and C = -9. The general solution is

Ch4 第 21 頁
Example 2: y''' + y" = excosx
Solution: The auxiliary equation is
m3 + m2 = 0
and the roots are m = 0, m = -1. The complementary solution is
yc = c1 + c2x +c3e-x.
The trial particular function is
yp = ex(Acosx + Bsinx)
Then,
yp''' + yp" = (-2A + 4B)excosx + (-4A-2B)exsinx
and consequently,
-2A + 4B = 1
-4A - 2B = 0
This gives A = -1/10, B = 1/5. The general solution is

Example 3: Solve y" - 2y' - 3y = 4x - 5 + 6xe2x.

Solution: y" - 2y' - 3y = 0 has the solutions of yc = c1e-x + c2e3x. The trial particular solution
form is
yp = Ax + B + Cxe2x + Ee2x
Substituting it into the original equation gives
A = -4/3, B = 23/9, C = -2, E = -4/3
So the general solution is
yc = c1e-x + c2e3x - 4x/3 + 23/9 - 2xe2x - 4e2x/3

Case II A function in the assumed particular solution duplicates a function in the

complementary function.

Example 1: Solve y" - 2y' + y = ex

Solution: The complementary function is
yc = c1ex + c2xex
Suppose we set the particular function as yp = Aex. Then as it is substituted into the equation,
the left side must be zero while the right side is still ex since ex is a complementary function.
In order to avoid this trouble, we may try the following particular solution form
yp = Ax2ex
and we get
2Aex = ex, and so A = 1/2.
Therefore, the general solution is

Ch4 第 22 頁
.

Example 2: Solve y" + y = 4x + 10sinx

Solution: The complementary function is
yc = c1cosx + c2sinx
Now, the trial particular solution is
yp = Ax + B + Cxcosx + Dxsinx
Substituting the solution form into the equation gives
Ax + B - 2Csinx + 2Dcosx = 4x + 10sinx
and so
A=4
B=0
C = -5
D=0
The general solution is
y = c1cosx + c2sinx + 4x - 5xcosx

4.5 Undetermined Coefficients -- Annihilator Approach

Define

,
and an nth-order linear differential operator can be written as
L = anDn + an-1Dn-1 + ... + a1D + a0.
We can regard D as a number and proceed any mathematical operations. For example,
D2 + 5D + 6 = (D + 2)(D + 3) = (D + 3)(D + 2)
This can be proved by (D2 + 5D + 6)y = [(D + 2)(D + 3)]y and also illustrates a general
property:
Factors of a linear differential operator with constant coefficients commute.

Annihilator Operator If L is a linear differential operator with constant coefficients and y =

f(x) is a sufficiently differentiable function such that
L(y) = 0
the L is said to be an annihilator of y. For example, the differential operator Dn annihilates
each of the functions 1, x, x2, ... , xn-1.

The functions that are annihilated by an nth-order linear differential operators L are
simply those functions that can be obtained from the general solution of the
homogeneous differential equation L(y) = 0.
The differential operator (D - )n annihilates each of the functions, ex, xex, x2ex, ... ,
xn-1ex.

Example: Find the functions which are annihilated by [D2 - 2D + (2 + 2)]n.
Solution: The roots of the auxiliary solution are

with mutiplicity of n.
The functions annihilated by the operator are
excosx, xexcosx, x2excosx, ... , xn-1excosx,
exsinx, xexsinx, x2exsinx, ... , xn-1exsinx.

Ch4 第 23 頁
 Suppose L1 and L2 are linear differential operators with constant coefficients such that
L1 annihilates y1(x) and L2 annihilates y2(x) but L1(y2)  0 and L2(y1)  0. Then the
product of the differential operators L1L2 annihilates the sum c1y1(x) + c2y2(x).
The main reason for this property is that L1 and L2 are commute and linear.
L1L2(c1y1(x) + c2y2(x)) = c1L2L1y1(x) + c2L1L2y2(x) = 0

Example: Find a differential operator that annihilates e-3x + xex.

Solution: D + 3 annihilates e-3x, and (D - 1)2 annihilates xex. The differential operator (D + 3)
(D - 1)2 annihilates their sum.

An alternative approach to undetermined coefficients

A differential equation of
L(y) = g(x)
can be solved by the following steps.
1. Find the complementary function yc(x) of L(y) = 0.
2. Find the annihilator operator, L1 of g(x).
3. Find the complementary function gc(x) of L1L.
4. The additional terms of gc(x) except those terms in yc(x) are the trial particular function of
L(y) = g(x).

Example 1: Solve y" + y = xcosx - cosx

Solution: L = D2 + 1 and the complementary function is
yc = c1cosx + c2sinx
The annihilator operator of xcosx - cosx is L1 = (D2 + 1)2, and the complementar function of
L1L is
gc = c1cosx + c2sinx + c3xcosx + c4xsinx + c5x2cosx + c6x2sinx.
The trial particular function of y is
yp = Axcosx + Bxsinx + Cx2cosx + Ex2sinx
and consequently,
yp" + yp = 4Excosx - 4Cxsinx + (2B + 2C)cosx + (-2A + 2E)sinx.
This gives E = 1/4, C = 0, B = -1/2, and A = 1/4. The general solution is

Example 2: Determine the form of a particular function for

y''' - 4y" + 4y' = 5x2 - 6x + 4x2e2x + 3e5x
Solution: L = D3 - 4D2 + 4D = D(D - 2)2 and the complementary function is
yc = c1 + c2e2x + c3xe2x.
The annihilator operator, L1 of g(x) is D3(D - 2)3(D - 5). The product of L1 and L is
L1L = D4(D - 2)5(D - 5),
and the corresponding complementary function is
gc = c1 + c2x + c3x2 + c4x3 + c5e2x + c6xe2x + c7x2e2x + c8x3e2x + c9x4e2x + c10e5x
The trial particular function is
yp = Ax + Bx2 + Cx3 + Ex2e2x + Fx3e2x + Gx4e2x + He5x.

4.6 Varaition of Parameters

We consider a nth-order linear differential equation:

.
The complementary function of the equation is
yc(x) = c1y1(x) + c2y2(x) + ... + cnyn(x)

Ch4 第 24 頁
 yc(x) = c1y1(x) + c2y2(x) + ... + cnyn(x)
where yi(x) are the solution of the associated homogeneous solution and ci are arbitrary
parameters. Now we are trying to find a particular solution from these complementary
functions. Since ci are arbitrary parameters, we may replace them with n dependent functions
of x, u1(x), u2(x), ... , un(x) and get
yp(x) = u1y1 + u2y2 + ... + unyn.
y1, y2, ... , yn are known functions and now, we have n dependent variables and one
differential equation. This indicates that we can set another n-1 more equations of ui to solve
the problem. We consider the following situations:
yp' = u1y1' + u2y2' + ... + unyn' + (y1u1' + y2u2' + ... + ynun')
and we set
y1u1' + y2u2' + ... + ynun' = 0.
Then,
yp' = u1y1' + u2y2' + ... + unyn'.
The second derivative is
yp" = u1y1" + u2y2" + ... + unyn" + (y1'u1' + y2'u2' + ... + yn'un')
and we set
y1'u1' + y2'u2' + ... + yn'un' = 0
Then
yp" = u1y1" + u2y2" + ... + unyn"
Similarly, the (n-1)th derivative is
yp(n-1) = u1y1(n-1) + u2y2(n-1) + ... + unyn(n-1)
and we set n-1 differential equations:
y1u1' + y2u2' + ... + ynun' = 0
y1'u1' + y2'u2' + ... + yn'un' = 0
...
y1(n-2)u1' + y2(n-2)u2' + ... + yn(n-2)un' = 0.
Now, we consider the nth derivative of yp:
yp(n) = u1y1(n) + u2y2(n) + ... + unyn(n) + (y1(n-1)u1' + y2(n-1)u2' + ... + yn(n-1)un')
and we have to substitute it and the low order derivatives into the nonhomogeneous equation.
We find that
y1(n-1)u1' + y2(n-1)u2' + ... + yn(n-1)un' = f(x)
and conclude a set of equations
y1u1' + y2u2' + ... + ynun' = 0
y1'u1' + y2'u2' + ... + yn'un' = 0
...
y1(n-2)u1' + y2(n-2)u2' + ... + yn(n-2)un' = 0
y1(n-1)u1' + y2(n-1)u2' + ... + yn(n-1)un' = f(x).
We can use Cramer's rule to find ui':

where W is the Wronskian of y1, y2, ... , yn, and Wi is

where the ith column is replaced by 0, 0, ... , 0, f(x). From ui'(x), we can get ui without an
integrating constant and find the particular solution.

Remark:
The advantages of the variation of parameters over the method of undetermined
coefficients are that
1. it will always yield a particular solution yp(x)
2. it is not limited to a function f(x), which is a combination of polynomials, sines, cosines,
and exponentials, .

Ch4 第 25 頁
and exponentials, .
3. it is applicable to differential equations with variable coefficients.

Example 1: Solve 4y" + 36y = csc3x

Solution: Put it into a standard form

.
The particular solution is assumed to be

and the set of the linear equations is

cos3x u1' + sin3x u2' = 0
-3sin3x u1' + 3cos3x u2' = csc3x/4
Therefore,

and

and we solve these two equations to get

and

.
The particular solution is

and the general solution is

Example 2: Solve y" - y = 1/x.

Solution: The particular function is assumed to have the form of
yp = exu1 + e-xu2
and the set of linear equtions of u1' and u2' is
exu1' + e-xu2' = 0
exu1' - e-xu2' = 1/x
Therefore,

and

and

and the general solution is

Ch4 第 26 頁
.
We can integrate on any interval x0  t  x not containing the origin.

4.7 Cauchy-Euler Equation

解區段是either (-, 0) or (0, )虳範例。

Another name for this equation is equidimensional equation.

In general, for the equations with variable coefficients, we always expect an infinite
series solution. But for this case, we can find a simple close form with fundamental
functions. Since the equation is linear, if we find n linearly independent complementary
functions and one particular function, then we can derive the family of the general
solution. Especially, we will try the xm as a complementary function, which is similar to
emx for the equation with constant coefficients.
For the sake of discussion, we shall confine our attention to solving the homogeneous
second-order equation.

Also, we can solve the nonhomogeneous equation

by variation of parameters once we have determined the complementary function yc(x).

Note: The coefficient of d2y/dx2 is zero at x = 0. Hence, in order to guarantee that the solution
of Cauchy-Euler equation is unique, we shall confine our attention to finding the general
solution on the interval (0, ). Solutions on the interval (-, 0) can be obtained by
substituting t = -x in the differential equation.

Method of Solution
Substituting y = xm into the homogeneous equation, we find that
am(m - 1)xm + bmxm + cxm = 0
Thus, the auxiliary equation can be simplified as
am(m - 1) + bm + c = 0 or am2 + (b -a)m + c = 0

Case I: Distinct Real Roots m1 and m2 are the two different real roots of the auxiliary
equation. Then the general solution is
y = c1xm + c2xm
1 2

Example Solve

Solution: The auxiliary equation is

m(m - 1) - 2m - 4 = 0
i.e., m = -1 or 4
The general solution is
y = c1x-1 + c2x4

Case II: Repeated Real Roots If m1 = m2 = -(b - a)/2a then we can use the solution to
construct a second solution. First we rewrite the Cauchy-Euler equation in the form

Ch4 第 27 頁
and then the second solution is given as

.
The general solution is
y = c1xm + c2xm lnx
1 1

Example: Solve

The auxiliary equation is

4m(m - 1) + 8m + 1 = 0
and the roots are m = -1/2. the general solution is

For higher-order equations, if m1 is a root of multiplicity k, then it can be shown that

, ... ,

are k linearly independent solutions.

Case III: Conjugate Complex Roots If m1 and m2 are complex conjugates

m1 =  + j, m2 =  - j
where  and  > 0 are real, then

The two independent solution can be chosen as

y1 = xcos(lnx), and y2 = xsin(lnx)
The general solution is
y = c1xcos(lnx) + c2xsin(lnx)

Example: Solve

.
Solution: The auxiliary equation is
m(m - 1) + 3m + 3 = 0
and the roots are

. The general solution is

Ch4 第 28 頁
Higher-order case
Example: Solve

Solution: The auxiliary equation is

m(m - 1)(m - 2) + 5m(m - 1) + 7m +8 = 0
The roots are m1 = -2, m2 = 2i, m3 = -2i. The general solution is
y = c1x-2 + c2cos(2lnx) + c3sin(2lnx)

Nonhomogeneous case
Example: Solve x2y" - 3xy' +3y = 2x4ex
Solution: The auxiliary equation is
m(m - 1) - 3m + 3 = 0
and the roots are 1 and 3. The complementary function is
yc = c1x + c2 x3
Using the variation of parameters, we can find the particular function. Now

and

Hence,
u1 = -x2ex + 2xex - 2ex and u2 = ex
The particular function is
yp = u1y1 + u2y2
= 2x2ex - 2xex
Finally we have
y = c1x + c2 x3 + 2x2ex - 2xex

Example: Find a particular solution of the nonhomogeneous equation

x3y(3) + x2y" - 6xy' + 6y = 30x
Solution: Writing it in the standard form gives

The complementary function is

yc = c1x + c2x3 + c3x-2
The variation of parameter leads to

and

Ch4 第 29 頁
Integrating these derivatives, we get
u1 = 5 lnx,

Thus our particular solution (for x > 0) is

Alternative Method of Solution

Any Cauchy-Euler equation can be reduced to an equation with constant coefficients by
means of the substitution x = et.

Example: Solve

.
Solution: Let x = et and we find

The given equation can be rewritten as

For this equation, the auxiliary equation is

m2 - 2m + 1 = 0
and the roots are m = 1 with multiplicity 2. The complementary function is
yc = c1e2t + c2te2t
and the particular function is 2 + t. We get the general solution is
y = c1e2t + c2te2t + 2 + t.
Writing it as a function of x gives
y = c1x + c2xlnx + 2 + lnx

4.8 Systems of Linear Equations

Ch4 第 30 頁
 Simultaneous ordinary differential equations involve two or more equations containing
derivatives of two or more unknown functions of a single independent variable.

Example: x' - 3x + y' + z' = 5

x' - y' + 2z' = t2
x + y' - 6z' = t - 1

Systematic Elimination to get the solution of a system

Example 1: Solve x' - 4x + y" = t2
x' + x + y' = 0
Solution: Write the simultaneous differential equations in operator notation:
(D - 4)x + D2y = t2
(D + 1)x + Dy = 0
Then, by eliminating x we get
[(D + 1)D2 - (D - 4)D]y = (D + 1)t2
(D3 + 4D)y = 2t + t2
The complementary function of y is
yc = c1 + c2cos2t + c3sin2t.
The trial particular function of y is
yp = At3 + Bt2 + Ct
and hence,
yp''' + 4yp' = 12At2 + 8Bt + 6A + 4C = t2 + 2t.
We conclude that A = 1/12, B = 1/4, C = -1/8. The general solution of y is
y = c1 + c2cos2t + c3sin2t + t3/12 + t2/4 -t/8
Eliminating y from the simultaneous equations, we get
[(D-4) - D(D + 1)]x = t2
(D2 + 4)x = -t2
The general solution of x is
x = c4cos2t + c5sin2t -t2/4 + 1/8
Bringing the solutions of x and y back into the second equation, we find
(c5 - 2c4 - 2c2)sin2t + (2c5 + c4 + 2c3)cos2t = 0
so that
c5 - 2c4 - 2c2 = 0 and 2c5 + c4 + 2c3 = 0
This leads to

and

Finally, a solution of the simultaneous differential equations is

Use of Determinants
Suppose
L1x + L2y = g1(t)
L3x + L4y = g2(t)
where L1, ... , L4 are linear differential operators with constant coefficients. The differential
equation of x only is

Ch4 第 31 頁
 Similiarly, for y, we have

If

and is a differential operator of order n, then

1. The system can be decoupled into two nth-order differential equations in x and y.
2. The characteristic equation and hence the complementary function of each of these
differential equations are the same.
3. Since x and y both contain n constants, there is a total of 2n constants appearing.
4. The total number of independent constants in the solution of the system is n.
If

then the system may have a solution containing any number of independent constants or may
have no solution at all.

Example 2: Given the system

Dx + Dz = t2
2x + D2y = et
-2Dx - 2y + (D+1)z = 0
find the differential equation for the variable y.
Solution:

The final result is

D(3D3 + D2 -4)y = 4et - 2t2 - 4t

4.9 Nonlinear Equations

Reducible Second-Order Equations:
(1) y is missing: F(x, y', y") = 0; Let p = y', and then F(x, p, p') = 0
(2) x is missing: F(y, y', y") = 0 or y" = f(y,y'); Let p = y' and then F(y, p, pdp/dy) = 0. Such kind
of equations is called autonomous.

A general form of a second order differential equation is

F(x, y, y', y"). = 0
If either the dependent variable y or the independent variable x is missing from a second
order equation, then it is easily reduced to a first order equation by a change of variables.
If the dependent variable y is missing, i.e.,
F(x, y', y") = 0,
then we make the substitution,
p = y', and

Ch4 第 32 頁
,
and the transformed equation becomes
F(x, p, p') = 0.
If the independent variable x is missing, i.e.,
F(y, y', y") = 0,
then we make the substitution,
p = y', and

and assume that p is a function of y. The transformed equation becomes

.
If we can find its general solution p = p(y, c1), then we can the general solution of the original
second order equation by integrating:

.
Example 1: Solve the equation xy" + 2y' = 6
Solution: The substitution p = y' gives

, that is,

The solution given by integration factor method is

Another integration produces the solution

Example 2: Solve the equation yy" = (y')2 under the assumption that y and y' are known to be
positive.
Solution: The substitution p = y' yields

and the solution is

p = c1y
Thus,

Another form of the solution may be given by

y = AeBx
where A = exp(-c2) and B = c1.

Use of Taylor Series:

Example: Solve y" = x + y - y2, subject to y(0) = -1, y'(0) = 1.
Solution: y"' = (x + y - y2)' = 1 + y' -2yy' = 4 at x = 0
Similarly, y(4)(0) = (1 + y' - 2yy')' = y" - 2yy" - 2(y')2 = -8
y(5)(0) = (y" - 2yy" - 2(y')2)' = y'" - 2yy"' - 6y'y" = 24
Therefore,
y(x) = -1 + x - x2 +2x3/3 - x4/3 + x5/5 + ....

Ch4 第 33 頁
 y(x) = -1 + x - x2 +2x3/3 - x4/3 + x5/5 + ....

Use of an ODE solver

The solutions found by the Taylor series and the ODE solver are consistent on (-1, 1).
The oscillatory behavior for x > 1 seems to suggest that it will oscillate as x approaches
infinity. For the initial condition, the graph generated by the ODE solver on a larger
interval suggest the answer is yes. But this single example does not answer the basic
question of whether all solutions of the differential equation y" = x + y - y2 are
oscillatory in nature.

Ch4 第 34 頁
2016.10.19 （三）ok
2016年10月19日 下午 04:36

Ch4 第 35 頁
Higher order LDE解法簡介
2016年10月19日 下午 04:37

nth-order Linear equation

-> singular condition & solution interval
-> Homogeneous equation, find yc
-> Linear dep. & indep.
-> Find yp
-> expand solution interval

考慮 n階線性微分方程式

須注意，線性微分方程式的係數 必須不含y以及其微分項。
首先由微分方程式樣貌判斷唯一解析解區間以及singular points
將原方程式整理成normal form

唯一解析解區要求 對於所有變數的任何階次偏微分均存在，使之不滿
足的條件稱作singular conditions。以上條件式對於任何微分方程式均適用，但對於n
階線性微分方程式而言，可以簡化Singular conditions 為 與 非對於所有
變數的任何階次偏微分均存在。並且須注意到線性微分方程式係數 僅與x有關，
因此singular conditions 分割的邊界是以x=const 做分割，而非線性微分方程式可能出
現與x-y均有關的分割邊界，簡言之，線性微分方程式的區間分割僅與x有關，而非線
性微分方程式可能是x-y平面的一條曲線，或者是更高維空間的圖形。

在每個唯一解析解區之內，各別求出在該區間的解。求解方式為先令
(homogeneous)，求解出 ，此為 n-parameter family of solution，對於線性微分
方程式n-parameter family of solution 的集合可以縮小為
，其中 為線性獨立的解， 為這些線性獨立的解之線性
組合。

Note : 給出微分方程式general form ，從中藉由代數方法抽取出數

個normal form (在chap 1 已介紹過)。n階normal form 微分方
程的解會有n個自由度，也就是會得出n-parameter family of solution ，這是因為需
要給出n個初始條件 才可唯一確定 。上述是對於
任何normal form 均成立，但並沒有規範n-parameter family of solution的形式，因
此parameter C 可以是非線性的。而n階線性微分方程已經對方程式形式有所規範，因
Ch4 第 36 頁
此parameter C 可以是非線性的。而n階線性微分方程已經對方程式形式有所規範，因
此會縮小n-parameter family of solution的集合，也就是parameter C 必須是線性組
合。

再解決non-homogeneous的情況，只要找出一個particular solution 滿足
且與 線性獨立即可，
該微分方程的解即為 。特別注意，這個解僅
在該唯一解析解區成立，若解區間要擴大而跨過singular point 則需要另外討論。

本書將方程式的解分成 solution 與 function兩種，兩者的差異是solution的區間為唯

一解析解區(不含singular points)，而function的區間中可以包含singular points。當
在唯一解析解區之內完成求解後，得到的解會被限制在該唯一解析解區間，若要求擴
大解區間(區間可以包含數個唯一解析解區以及分割各個唯一解析解區的singular
point)，則會讓解的集合增大(?)，也就是說在唯一解析解區中的解(稱作solution)可以
寫為 ，但如果解區間包含singular point (稱作
function)則可以不是這樣的形式。

Ch4 第 37 頁
Superposition principle
2016年10月19日 下午 04:37

1. For homogeneous DE

Consider n-th order linear DE

(1) Let be the solution of the DE

代入微分方程，滿足

，c為常數
因此c 也是微分方程的解

(2)Let and be the solutions of the DE

與 各自滿足微分方程，將兩式相加可得

因此 亦為微分方程的解

由(1)(2)可以推得，如果 為微分方程的解，則 亦為微

分方程的解。此為Superposition principle for homogeneous DE。

2. For Nonhomogeneous DE
Consider n-th order linear DE

先考慮q個微分方程

設 分別為該q個微分方程的解，意即滿足

將上式相加可得

由上式可看出如果g(x)可以分解成多個函數相加，則求解各別方程式得出的解，再相
加則為原非齊性微分方程的解。此為Superposition principle for nonhomogeneous DE。
Ch4 第 38 頁
加則為原非齊性微分方程的解。此為Superposition principle for nonhomogeneous DE。

Ch4 第 39 頁
Existence & Uniqueness
2016年10月19日 下午 04:37

一般課本所給的定理

[Existence and Uniqueness Theorem for initial value problem]

Consider n-th order linear DE

With initial value

If the coefficients are continuous on an interval I and

for every x in this interval, then the DE have a unique solution y(x) on I.

須注意此定理並非if and only if

[Unique and Analytic solution Theorem for initial value problem]

本書給出另一個定理

Consider n-th order linear DE

With initial value

If and only if the coefficients have any order derivatives on

an interval I and for every x in this interval, then the DE have a unique analytic
solution y(x) on I.

本定理為if and only if，若要求出interval 的邊界，取此定理的否定即可找出singular

condition，該singular condition 與微分方程的交集即為unique analytic solution interval 的
邊界，也就是本書所定義的 singular points。

Ch4 第 40 頁
Linear Dependence & Independence
2016年10月19日 下午 04:58

[Definition -- Linear Dependence ]

Consider a set of piecewise continuous functions .
It is said to be linear dependent on an interval I if and only if there exist constants
not all zero, such that for every x in the
interval I.

[Definition -- Linear Independence ]

If and only if a set of piecewise continuous functions
is not linear dependence, then it is linear independence.

首先定義Linear Dependence，而該定義的否定為Linear Independence。

可以將該定義的形式改變，更清楚看出linear dependence 的定義。

假設 (必有另一個 c值不為零)，可將定義式改寫成

意即 可以由其他的函數之線性組合來表示。

對於piece wise continuous 的函數通常會分成多個子區間來討論c值，而定義要求同一組

c值對於每個x在區間 I 均要成立，意即要可以求出 c的通解。因此linear dependence 的兩
個條件為交集 : " 有不全為 0之解" and " "。Linear independence 為上述條件的否
定 : " 僅有全為 0之解" or "for some (不能適用所有子區間)"。

Example:

討論兩函數的相依性

Ch4 第 41 頁
2016.10.21(五)
2016年10月30日 下午 07:41

Ch4 第 42 頁
Wronskian Theorem
2016年10月30日 下午 07:44

對於解析函數(可進行無窮多次微分)可以有較有系統的方式來討論相依性

Consider analytic function .

將上式微分1~(n-1)次可得聯立方程組

如果 有不全為 0的解(多解)，則

定義為Wronskian

(未完)

[圖解相容性及獨立性]
根據相容性及獨立性的定義、Singular Point 的定義和 W 值判斷 c 值的多重解及
唯一解，我們可歸納成如下的圖解。

Ch4 第 43 頁
 唯一解，我們可歸納成如下的圖解。
1. function interval 是可包涵 Singular Point 而 Solution Interval 則否，故 function interval 是
可大於 solution interval。
2. Independent function 是根據 Singular Point 來進行分類。其中 Condition A 是不包涵
Singular Point，而 Condition B 是包涵 W = 0 及 W Uncertainty 等兩類 Singular Point。
3. 由於 W = 0 的 Singular Point 已意味著在該點已有不全為零的 c 值解，故在其分割的區
段上，可以是全為零或不全為零的 c 值，即符合 Condition B 的要求。
4. 相對地，在 W Uncertainty 的 Singular Point 的情形，在分割區段上則需至少有一不全
為零的 C 值出現，才能符合 Condition B 的要求。
5. 不同區段或點若有不全為零的 c 值，它們個別是不同的組合。

[其它注意事項]
6. Since the solution interval is a subset of the function interval for the same form of the
solution and function, the above statement indicates that dependent functions must be
the dependent solutions. 函數的相容性條件比解函數的嚴格，相對地，獨立性條件將變
得較寬大，這在底下會說明。
7. The dependence is much related with the (function or solution) interval I.
8. Another equivalent linear independence if and only if condition
The only constants for which

for every x in the interval are c1 = c2 = c3 = ... = cn = 0 (只有全為零的 ci 值通解)。used in

the textbook)
9. The independent solutions must be the independent functions and some dependent
solutions may be independent functions. 此 independent function 就是 singular

Ch4 第 44 頁
 solutions may be independent functions. 此 independent function 就是 singular
solution。Not true for that the independent functions must be the independent
solutions(Condition B).
10.

6.

[Example]
Linear dependence functions:
sin2x and sinx cosx
cos2x, sin2x, sec2x, tan2x (since cos2x + sin2x - sec2x + tan2x = 0)

Linear independence functions:

Example 1：y = x and y = |x| 的相容性及獨立性並找出在

的線性微分方程及其通解。
Sokution
(1) 相容性：dependent solutions on (-, 0) or

.
(2) 獨立性：若考慮c1x + c2|x| = 0的c解而言，在(-, 0)內的解為(c1, c2) = (0, 0)或(c, c)；但
在(0, )內的解則為(c1, c2) = (0, 0)或(c, -c)；在原點時則為隨意。因此，就

而言，我們有三組不全為零的c解；分別在(-, 0)、(0, )及原點符合Condition B，或者

是通解只有 (0, 0)，因此x and |x|互為獨立。
(3) 線微分方程式的通解為 y = cx on (-, 0) or

亦即 xy’ – y = 0；而且 y = |x|也是這個微分方程式的一 singular solution。

(4) DE 的通解：考慮xy’ – y = 0的逯解在

，若符合y(0)是連續的，則其通解為

其中i為常數。若符合y(0)及y’(0)均是連續的，則其通解為

Ch4 第 45 頁
 其中i為常數。若符合y(0)及y’(0)均是連續的，則其通解為

，c is arbitrary。由本例得知，在singular point的邊界條件影響結果甚鉅，如何求知該條
件是一重要課題。

Example 2：考慮 y = x 與 y1 的相容性及獨立性並找出在

的線性微分方程及其通解。其中的

Solution
(4) 相容性：independent solutions on (-, 0) but dependent solution on

(5) 獨立性：就 y = x 與 y1 的 c1x + c2y1 = 0 而言，在(-, 0)內的解為(c1, c2) = (0, 0)；但在(0,

)內的解則為(c1, c2) = (0, 0)或(c, -c)；在原點時則為隨意。因此，就

而言符合Condition B，因此x and y1互為獨立。.

(6) 線性 DE：由於x及x2為兩個獨立解，所以相對應的通解為y = c1x + c2x2 on (-, 0) 或 (0,
)；而微分方程為x2y" - 2xy' + 2y = 0而y1也是一解。
DE 通解：考慮在

的通解，若符合y(0)是連續的，則其通解為

其中i為常數。

[Wronskian Theorem]
[W 值的定義]
Suppose f1(x), f2(x), ... , fn(x), possesses at least n-1 derivatives. The determinant
W(f1(x), f2(x), ... ,fn(x)) =

is defined as the Wronskian of functions.

[Independence and Dependence For functions]

[Corollary I]
If f1(x), f2(x), ... , fn(x), possesses at least n-1 derivatives and are linearly dependent on
I, then
W(f1(x), f2(x), ... ,fn(x)) = 0
for every x in the interval.
Proof:
Since they are linear dependent, we have

true for every x on the interval I and the ci is not all zeros. The derivatives of the above zero
function are

Ch4 第 46 頁
1st derivative

2nd derivative

.......
(n-1)th derivative

This becomes a set of linear equations with variables of ci. Because the solution is not unique
(not all zeros, and all zeros), we have
W(f1(x), f2(x), ... ,fn(x)) =

=0
for every x on the interval I.

[Corollary II]
因為 AB is equivalent to ~B~A，故由 Corollary I 得到
Condition A
If W(f1(x), f2(x), ... ,fn(x))  0 for every x in the interval,
Or
Condition B
If W(f1(x), f2(x), ... ,fn(x)) = 0 for some x instead of every x in the interval,
then the functions are linear independent on the interval.
此結果與前面 Independent function 定義一致。課本所描述的定理雖沒有錯誤，但其
否定的範圍不夠完整。

Example 1(Corollary I): f1(x) = sin2x and f2(x) = 1 - cos2x are linearly dependent on

.
proof: We find
W(f1(x), f2(x)) =

= 0 for every x

Example 2(Corollary IIA): f1(x) =

and f2(x) =

, m1  m2.
Proof: We find
W(f1(x), f2(x)) =

for every real value of x. Hence they are independent.

Example 3(Corollary IIB): f1(x) =

Ch4 第 47 頁
and f2(x) =

.
Proof: We find
W(f1(x), f2(x)) =

if x  0. They are independent for all real numbers even though W = 0 for x = 0.

Example 4: It is not true that if W(f1(x), f2(x), ... ,fn(x)) = 0 for every x in the interval,
then f1(x), f2(x), ... , fn(x) are linearly dependent on I. The example is f1(x) = x3 and f2(x) =
| x3 |. We do have W( f1(x), f2(x)) = 0 for every x on (-, ) but they are independent
functions.

Note: It should also be noted especially that they cannot be solutions of a second-order
linear differential equation although they are independent functions. The reason is that
they are composed of dependent solutions and a singular point x = 0.(請參考前面的圖,
在independent function是可能有W=0情形)

[Independence and Dependence For solutions]

[Corollary III]
Linearly Independent Solutions
Let y1, y2, ... , yn be n solutions of the homogeneous linear nth-order differential
equation on an interval I. Then the set of solutions is linearly independent on I if and
only if (

)
W(y1, y2, ... , yn)  0
for every x in the interval.
仔細的證明提供參考
Proof: The necessity (

) i.e., to find the homogeneous differential equation

Consider y(x) and their derivatives
y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x)
y’(x) = c1y1'(x) + c2y2'(x) + ... + cnyn'(x)
........
y(n-1)(x) = c1y1(n-1)(x) + c2y2(n-1)(x) + ... + cnyn(n-1)(x)
Since W(y1, y2, ... , yn)  0 for every x in the interval, we can find the unique solution ci and
substitute them into the nth derivative of y(x). Finally we can derive the homogeneous
equation.

The sufficiency(

) is proved as followings.
"If y1, y2, ... , yn are linearly independent, W(y1, y2, ... , yn)  0 for every x in I."
That is we are going to prove that if W(y1, y2, ... , yn) = 0 for some x0, then the solutions are
linearly dependent.
If W(y1, y2, ... , yn) = 0 for some x0, then we consider the system of the equations for ci

Ch4 第 48 頁
If W(y1, y2, ... , yn) = 0 for some x0, then we consider the system of the equations for ci
c1y1(x0) + c2y2(x0) + ... + cnyn(x0) = 0
c1y1'(x0) + c2y2'(x0) + ... + cnyn'(x0) = 0
........
c1y1(n-1)(x0) + c2y2(n-1)(x0) + ... + cnyn(n-1)(x0) = 0
Since W = 0, there exist solutions ci, i = 1, ... , n which are not all zero. If we define a
function y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x), then the above conditions indicate that initial
conditions are y(x0) = 0, y'(x0) = 0, ... , y(n-1)(x0) = 0. In addition, due to superposition principle,
y(x) is also a solution of the homogeneous nth-order linear differential equation. We know
that y(x) = 0 is the unique solution which satisfies the initial conditions. This indicates that
y(x) = c1y1(x) + c2y2(x) + ... + cnyn(x) = 0 for every x in I where. ci, i = 1, ... , n are not all zero.
This implies that yi, i = 1, ... , n, are not independent..

[Choice from n functions]

If we choose k ones, say y1, y2, ... , yk, from the n functions where k < n, the Wronskian
W(y1, y2, ... , yk) may be 0 for some point and their associated linear homogeneous kth-
order differential equation may have a different solution interval I.

Ch4 第 49 頁
Solution interval -> function interval
2016年10月30日 下午 07:48

考慮微分方程

其解為 、x
以Wronskian thm判斷singular point

當x=1，Wronskian=0, 表示在x=1處 、x不滿足線性獨立的條

件，在該點為singular point

可以得出兩個唯一解析解區 與
在左邊區間
右邊區間
在各自唯一解析解區的解稱做solution，僅在各自的解區間中解
才會成立，若要將區間擴大，求出整個實數軸的通解（稱做
function)，就要處理singular point接點的問題。

至於在x=1處要如何接起來，這就要考慮接點的連續性。
在第七章會介紹DiracDelta function，不連續點的微分就會出現
DiracDelta function，在第七章會仔細介紹。在此先給出結果

在整個實數軸的通解可以寫成

其中x與 為analytic function，任意階微分都連續

而 從二階微分開始不連續

若乘上

可以有在x=1處不連續的解

（再確認）

Ch4 第 50 頁
（再確認）

Ch4 第 51 頁
2016.10.26(三) ok
2016年10月30日 下午 07:41

Ch4 第 52 頁
由解求出微分方程
2016年10月30日 下午 07:52

設在domain 內無singular points

令 為analytic function

Let
求出描述 y(x)的n-th order linear DE?

將上式做(n-1)次微分後即可得出

Let

(將原本Wronskian 第 i 行換成(y-yp)的各階微分項)
須注意上式最高階出現 項，且y的各階微分項為線性組合

代入n階微分式

以上為n-th order linear DE 的normal form

上式可以更加簡化
定義

Ch4 第 53 頁
將原本Wronskian 第 i 列換成 的 n階微分，須注意上式僅與 x有關，不含
。須注意 與 不同，前者為已知的解，後者為用來求出滿足這
些解之微分方程的各階應變數微分。
經由複雜的行列式運算，可得

此化簡最重要的意義是將 從藏在 之內提到外面。

定義線性微分算符

(nonhomogeneous)
若 (homogeneous)

Example
求出描述兩解之二階齊性微分方程

Ch4 第 54 頁
2016.10.28(五)
2016年10月30日 下午 07:41

Ch4 第 55 頁
2016.11.01(三) ok
2016年11月2日 下午 02:54

Ch4 第 56 頁
求解LDE方法 yc & yp
2016年11月2日 下午 02:56

Homogeneous

Nonhomogeneous

其中， 指某個特定係數

指某個特定係數，與待定係數 不同。至於可以將 拆解
成數個函數的線性組合分別求出particular solution的理由請見
superposition principle for nonhomogeneous LDE

上式的 、 分別為 y與 g的消滅運算子(Annihilator operator)。

Ch4 第 57 頁
LDE Yc 解法
2016年11月2日 下午 02:57

考慮自由度為1

若為constant coefficient，

考慮
由上式可知，如果要消滅 項，應使用消滅運算子

經過 的運算後 仍維持一樣的形式，因此若要再消滅 也是選用一樣

形式的消滅運算子

可以看出兩個運算子是可以交換的，也就是

此為常係數(constant coefficient)特有的特性，並非對於任意消滅運算子都成立。

普遍來說，考慮 ， 均不相同( 均線性獨立)

由上面討論可知

將消滅運算子 乘開可得

其中 為常數，所對應的微分方程為

為n-th order homogeneous linear DE with constant coefficient

意即給出 ， 均不相同，可找到n-th order

homogeneous linear DE

現在考慮敘述的反面，若給出n-th order homogeneous linear DE with constant

coefficient

( 為常數)
是否可以得到其解為 ?
將微分方程式寫為operator的形式

依代數基本定理，可以將運算子分解成
(注意 : 此處要求 為常數，若 為x的函數則微分運算子亦須對 作用)
令 代入

解得
以下討論解的種類 ，相異根(再細分成相異實根與共軛複數根)、重根。

1.若 均相異，則所對應的解 線性獨立，因此已找到n階線性其

性方程式所有的解。
若重根，解只有一種形式?
2.重根
假設 重根兩次。可以確定 為方程式之一解，且與其他的解線性獨立。

Ch4 第 58 頁
假設 重根兩次。可以確定 為方程式之一解，且與其他的解線性獨立。
而重根的另一個解可由以下公式得出 (詳見。。。)

由此可見具有其他函數形式滿足此微分方程。

普遍而言，若 重根m次，其解為

須注意，x的冪次必須為0,1,2,…,(m-1) 依序出現，若不按此次序則非其解。
以下舉例說明

注意， 無法消滅

非constant coefficient operator

3.complex conjugate
為了使係數為實數，其形式為
、

為相異根的子集合， 、 即為方程式的解。其有方法轉換為實數解。
若 、 為微分方程式的解且線性獨立，令

且 、 為線性獨立，則 、 亦為微分方程的解。

使用Euler's formula :

用此方法可以將複數解轉為實數，可以證明 與 線性獨立，因此 與 為微分

方程式的解。

Ch4 第 59 頁
Ch4 第 60 頁
Nonhomogeneous LDE
2016年11月2日 下午 02:55

1.
2. 求解

Ch4 第 61 頁
Undetermined coef
2016年11月2日 下午 02:58

1.
2. 求解

依照 形式來決定 該如何假設

1.

假設為

2.

假設為

3.

假設為

4.1

4.2

假設為

Ch4 第 62 頁
Annihilate operator
2016年11月2日 下午 03:03

Annihilate operator 方法的概念是乘上能夠消滅掉 的消滅運算子（會讓集合變

大），但新的方程式 相較於原方程式 可能更好計算。 求出
之解後找出在原方程式解集合內的解，並與原方程式的齊性解線性獨立之解
即為特解。可以依此方法找出特解。

Example 1: Solve y" + y = xcosx - cosx

Solution: L = D2 + 1 and the complementary function is
yc = c1cosx + c2sinx
The annihilator operator of xcosx - cosx is L1 = (D2 + 1)2, and the complementar function of
L1L is
gc = c1cosx + c2sinx + c3xcosx + c4xsinx + c5x2cosx + c6x2sinx.
去除掉與complementary function線性相依的部分
The trial particular function of y is
yp = Axcosx + Bxsinx + Cx2cosx + Ex2sinx
帶回原微分方程，將解集合縮小到原方程式的解集合
and consequently,
yp" + yp = 4Excosx - 4Cxsinx + (2B + 2C)cosx + (-2A + 2E)sinx.
This gives E = 1/4, C = 0, B = -1/2, and A = 1/4. The general solution is
可以求出特解

Ch4 第 63 頁
可以求出特解

Example 2: Determine the form of a particular function for

y''' - 4y" + 4y' = 5x2 - 6x + 4x2e2x + 3e5x
Solution: L = D3 - 4D2 + 4D = D(D - 2)2 and the complementary function is
yc = c1 + c2e2x + c3xe2x.
The annihilator operator, L1 of g(x) is D3(D - 2)3(D - 5). The product of L1 and L is
L1L = D4(D - 2)5(D - 5),
and the corresponding complementary function is
gc = c1 + c2x + c3x2 + c4x3 + c5e2x + c6xe2x + c7x2e2x + c8x3e2x + c9x4e2x + c10e5x

去除掉與complementary function線性相依的部分
The trial particular function is
yp = Ax + Bx2 + Cx3 + Ex2e2x + Fx3e2x + Gx4e2x + He5x.
帶回原微分方程，將解集合縮小到原方程式的解集合

Ch4 第 64 頁
2016.11.04(五) ok
2016年12月6日 下午 08:39

Ch4 第 65 頁
Cauchy-Euler equation
2016年12月6日 下午 08:39

從solution 看equation會看到很多組合；若從equation 看solution只是猜答案

1.從equation 看solution

Cauchy-Euler equation

注意到每一項均為equal dimension
猜解為 代入原方程式，另 驗證齊性解

可以得知x=0為singular point，只要找出m值符合
之 即為方程式的解

2.從solution看equation

*比較

Case 1: 相異實根
考慮

Ch4 第 66 頁
注意 的位置可以互換，只是先消滅哪項的差別而已

比較

觀察

可以看出，若令 、 ，則兩者會相同( 因為考慮3個獨立解，微分算符

作用3次，x降三次方，因而出現)

Case 2: 重根
按照上述觀察結果繼續驗證重根的情況。

考慮

按 轉換重根之解

按Cauchy-Euler equation 的annihilate operator寫法找出重根之微分算符為

驗證 是否可以解出三重根

Ch4 第 67 頁
其解為 （三重根）
因此，令 、 ，重根的情形兩者也會相同

Case 3:Complex conjugate

按Cauchy-Euler equation 的annihilate operator寫法找出重根之微分算符為

驗證 是否可以解出

其解為
因此，令 、 ，complex conjugate的情形兩者也會相同

Ch4 第 68 頁
找出x<0的解
2016年12月6日 下午 08:42

Cauchy-Euler equation 在x=0為singular point，現在要考慮將解區間跨過singular point延

展到整個實數軸。
將x=0代入Cauchy-Euler equation 可以解出y=0，也就是說（x,y)=(0,0)為多解的singular
point，而y軸上的其他點為無解的singular points。

Cauchy-Euler equation 可能的解 或者是

在x趨近於0處，可能會趨向零或者正負無窮大，若要求在x=0處連續，則要求函數值趨
近於0

考慮二階Cauchy-Euler equation

其中x為dummy variable，用其他變數代入都可使方程式成立
令 代入

若令
則 與 之解 相同

也就是說要找出整個實數軸的解，要滿足

因此，若在x>0處求出
那麼將解擴展到整個實數軸則為
（再確認）

Ch4 第 69 頁
2016年12月6日 下午 09:03

Example

Hint:
(1)Guess a solution
(2)According to the guessing solution, find the relative first order differential
operator or differential equation
(3)To find the other second-order differential operator or differential
equation.

Sol:

帶入，為其解。令
令

左式=
比較係數

原式 ->

令
……(Cauchy-Euler equation)

…..(1st linear DE)

Ch4 第 70 頁
2016.11.09 (三)ok
2016年12月6日 下午 08:42

Ch4 第 71 頁
Variation of parameters
2016年12月6日 下午 08:42

考慮微分方程

設 已求到。
要求 滿足
設
可以發現上式共有n個未知 ，但我們只要求其中一個解即可，因此可以另
外假設(n-1)條限制方程式，最後希望可以求出唯一解，而非多解或是解為空集合。
將上式微分

令 ，此為第一條設定的限制方程式

在將上式微分

令 ，此為第二條設定的限制方程式

重複此步驟
可得第(n-1)條設定的方程式為

而第n條方程式為
用以上 結果可以將上式整理成

Ch4 第 72 頁
解出 ，代回 求出

Ch4 第 73 頁
2016.11.11 (五) ok
2016年12月6日 下午 08:43

Ch4 第 74 頁
 Nonlinear DE
2016年12月6日 下午 08:44

4.9 Nonlinear Equations

[有 constraint 的線性系統]
Reducible Second-Order Equations: 對於無法再因式分解的非線性系統，我們可以依照前述求解參數方式或擴大參數個數而變成線性
(1) y is missing: F(x, y', y") = 0; Let p = y', and then F(x, p, p') = 0 系統，但某些參數間彼此有 constraint 限制。其基本做法如下
(2) x is missing: F(y, y', y") = 0 or y" = f(y,y'); Let p = y' and then F(y, p, pdp/dy) = 0. Such kind of equations is 1. 升降階的基本做法
called autonomous. 方程式升階並維持原有集合大小

A general form of a second order differential equation is

F(x, y, y', y"). = 0
If either the dependent variable y or the independent variable x is missing from a second order equation, then it 方程式降階並維持原有集合大小
is easily reduced to a first order equation by a change of variables.
If the dependent variable y is missing, i.e.,
F(x, y', y") = 0,
then we make the substitution,
p = y', and
或無解
兩個同階的normal form聯立方程組相減可得到一個較低階(大都是低一階)的方程式。如果較
,
and the transformed equation becomes 低階者可求出解，則須反代入高階者去再度驗證，因為有可能是無解，反代不成就是代表無
F(x, p, p') = 0. 解。兩同階方程式相減可仍是同階或較低階。其中後者是最小集合，如此的唯一性可保證是
If the independent variable x is missing, i.e., 降階者是兩者之交集，但仍須排除交集是空集合的情形。利用上述升降階的方式可進行聯立
F(y, y', y") = 0, 方程組的簡化。
then we make the substitution,
p = y', and

and assume that p is a function of y. The transformed equation becomes

.
If we can find its general solution p = p(y, c1), then we can the general solution of the original second order
equation by integrating:

. 2. 假設某 n0-parameter family curves 是非線性系統，其中有 n0 項一次方參數，稱之為舊參數項；另外

有高次方(>1)項，它是某些一次方參數的組合，稱之為組合項，假設共有 n1 項。例如

前兩項為舊參數，後兩項為組合項。
Example 1: Solve the equation xy" + 2y' = 6
Solution: The substitution p = y' gives 3. 可依照前述求解參數方式而得微分方程。例如

, that is,
可解得參數共4組解，代入二階導函數後可得四次二階微分方程式

The solution given by integration factor method is

4. 對每一組合項指定一個一次方新參數，而使系統成為 n0+n1 個參數的線性系統，但另外有 n1 項舊
參數與新參數間的 constraint。例如
Another integration produces the solution

Example 2: Solve the equation yy" = (y')2 under the assumption that y and y' are known to be positive.
Solution: The substitution p = y' yields 1. 由於線性系統有 n0+n1 個參數，我們可依序微分直到 (n0+n1-1)th-order 導函數出現(此動作依昇階方
式事實可到任意階導函數)，而將所有參數用線性系統求解。然後代入 n1 個 constraint，即可得到
n1 個 (n0+n1-1)th-order DE。此 n1 個 DE 解的交集就是我們前述的 n0th-order DE 的解集合。例如
and the solution is
p = c1y
Thus,

Another form of the solution may be given by

y = AeBx
where A = exp(-c2) and B = c1. 四倜參數六個方程式，可解得參數及 2 DE
解得參數為

代入constraints而得

2. 利用每兩個 (n0+n1-1)th-order DE 可降階簡化成一個 n0+n1-2)th-order DE；故 n1 個 (n0+n1-1)th-order

DE 可針對共同一個(或相鄰) DE 降階簡化成 n1-1 個 (n0+n1-2)th-order DE。如此降階程序依序重覆
n1-1 次即可得到一個 n0th-order DE。

Ch4 第 75 頁
 此結果應與先前得到的相同。至於降階的程序是先視最高階導函數為未知數而求得每一個
constraint 的 normal form DE，然後兩方程式相減即是得到降階的結果。此降階過程即相等於
求解聯立微分方程組，因為針對最低階方程式，相對於家族曲線參數最少，應是較容易求
得。事實上，降階的過程即相當於代數求解過程中的消去未知數。由於說明例最後兩個
constraint 是二次式，我們就以二元二次聯立方程組來說明：

由於參數可解得是導函數的線性組合，而且最高階是第三階。因此，上式的 x 是用來代表第三階
導數，而 y 是代表其它較低階的導函數；為求符號簡便，故用一個來代表多數。故消去三階導函
數即相等於消去未知數 x。上式在代數中是代表雙曲線、橢圓或拋物線的交點，一般而言應有四組
解。如何消去 x？首先視 x 為未知數而其它均為已知數，而兩式分別求得兩根：

求根得到

這相當於有四組解

亦即為四組三階 normal form DE，相減後即為四組二階 DE。由於彼此是“或〞的關係，故可用相

乘得到

展開並以同一個二次方程的兩根和及兩根積來組合，即可得到

配合兩兩個二次方程式的型式，即可求出 y 所遵守的方程式，也是較低階(二階)的微分方程式。

7. 任何非線性的家族曲線均可改成線性家族曲線加上參數constraints的型式。因此非線方程式是可
包含在線性方程之內。

Example: y = c2e2x+2ce-2x, Please find the associated DE.

Solution:該 Family curves 可做是 y = c1e2x+c2e-2x 但受限於

的限制。故由前例的結果得到

代入受限條件得到

得刲

由前例得知此方程式的 solution interval 是整個 x 軸。又由於參數 c 是實數，故由其家族曲線方程

式得到
4e-4x-4 e2x (-y)  0 即 y  -e-8x
這即微分方程一個參數唯一解區，其相對應的微分方程有兩個一階的 normal form DE，但相乘整理後
為如上的 general form DE。
從求解過程可以看出：原組成單元是以原來的型式(或組合)出現在方程式的係數。此外，該方程式
對 y 而言是非線性的，此跟非線性組合是一致的。
該方程式相對於 one-parameter family curves，故階數應是一階的。方程式中如 (y’)2 及 yy’ 項的階
應指定為 1，意指他們無含有比一階更高階的項，其方程式解的參數僅有一個。
在本例中，我們是用 y (y = c2e2x+2ce-2x)及 y’兩項與 c1 及 c2 的關式來求解後兩者。相對地，我們也
可用如 y’及 y”(事實上此時的 y 可為 y = c2e2x + 2ce-2x + c3 )兩項與 c1 及 c2 的關式來求解後兩者。最
後代入兩者的受限條件會發現：兩方程式的階數是不相同的。事實上，後者的方程式是包含前者
的，因為前者獨立參數僅有一個(c)，後者郤有兩個(c 及 c3)。這說明在求解過程中，階數必須保持
最小，所得的微分方程式才能真正等於 family curves；否則變成包含 family curves。

從前逑兩 family curves 例子是可以看出：線性方程式

是包含非線性方程式

的所有解，這可由兩者 family curves 的定義得到。 從 family curves 來解釋常態微分方程式：方程式的

階數是相對

Ch4 第 76 頁