arXiv reader

Solving Sequential Linear M fractional Differential Equations with Constants Coefficients

V. Padmapriya and M. Kaliyappan

arXiv
1903.11969v1 · PDF · 27 March 2019
Categories
math.GM
Licence
CC BY 4.0 The site shows the text.

Abstract

Fractional calculus is a powerful and effective tool for modelling nonlinear systems. The M derivative is the generalization of alternative fractional derivative. This M derivative obey the properties of integer calculus. In this paper, we present the method for solving M fractional sequential linear differential equations with constant coefficients for alpha is greater than or equal to 0 and beta is greater than 0. Existence and Uniqueness of the solutions for the nth order sequential linear M fractional differential equations are discussed in detail. We have present illustration for homogeneous and non homogeneous case.

Items 44

Definition

Definition 2.1

066P

Let f:[0,\infty)\rightarrow \mathbb{R} be a function and t>0. Then for 0<\alpha<1, the M-fractional derivative of f of order \alpha is defined as

D^{\alpha,\beta}_{M} f(t)=\lim_{\epsilon \to 0} \frac{⁡f(t\mathbb{E}_{\beta}(\epsilon t^{-\alpha})-f(t)}{\epsilon}

Where \mathbb{E}_{\beta} (.), \beta>0 is Mittag-Leffler function with one parameter.

If f is M-differentiable in some interval (0,a),a>0 and

\lim_{t \to 0^+}D^{\alpha,\beta}_{M} f(t)

⁡ exists, then we have

D^{\alpha,\beta}_{M}f(0)=\lim_{t\to 0^+}D^{\alpha,\beta}_{M} f(t)
Theorem

Theorem 2.1

U4JC

Let 0<\alpha \leq 1,\beta>0, a,b\in\mathbb{R} and f,g be M - differentiable at a point t>0. Then

  1. D^{\alpha,\beta}_{M} (af+bg)(t)=aD^{\alpha,\beta}_{M} f(t)+bD^{\alpha,\beta}_{M} g(t) for all a,b\in \mathbb{R}.

  2. D^{\alpha,\beta}_{M}(f\cdot g) (t)= f(t) D^{\alpha,\beta}_{M}g(t) +g(t) D^{\alpha,\beta}_{M}f(t)

  3. D^{\alpha,\beta}_{M}(\frac{f}{g})(t)=\frac{g(t) D^{\alpha,\beta}_{M} f(t)-f(t) D^{\alpha,\beta}_{M} g(t))}{[g(t)]^{2}}

  4. D^{\alpha,\beta}_{M}(c)=0, where f(t)=c is a constant.

  5. D^{\alpha,\beta}_{M}t^{a}=\frac{a}{\Gamma(\beta+1)} t^{a-\alpha}

  6. Moreover, f is differentiable, then D^{\alpha,\beta}_{M}f(t)=\frac{t^{1-\alpha}}{\Gamma(\beta+1)} \frac{d f(t)}{dt}

Additionally,M-derivatives of certain functions as follows:

  1. D^{\alpha,\beta}_{M}(sin(\frac{1}{\alpha}t^{\alpha}))=\frac{cos(\frac{1}{\alpha}t^{\alpha})}{\Gamma(\beta+1)}

  2. D^{\alpha,\beta}_{M}(cos(\frac{1}{\alpha}t^{\alpha}))=\frac{-sin(\frac{1}{\alpha}t^{\alpha})}{\Gamma(\beta+1)}

  3. D^{\alpha,\beta}_{M}(e^{\frac{t^{\alpha}}{\alpha}})=\frac{e^{\frac{t^{\alpha}}{\alpha}}}{\Gamma(\beta+1)}

Used by Proof
Theorem

Theorem 2.2

RN1R

Let a>0 and t\geq a. Then, the M-integral of order α of a function f is defined by

_{M} I^{\alpha,\beta}_{a}f(t)= \Gamma(\beta+1) \int_{a}^{t} \frac{f(x)}{x^{1-\alpha}}dx
Theorem

Theorem 2.3

BSM4

Let a\geq 0 and 0<\alpha<1. Also let f be a continuous function such that there exists _{M} I^{\alpha,\beta}_{a}f. Then

D^{\alpha,\beta}_{M} (_{M} I^{\alpha,\beta}_{a} f(t))=f(t)
Theorem

Theorem 2.4

DG61

Let f,g:[a,b]\rightarrow \mathbb{R} be two functions such that f,g are differentiable and 0<\alpha<1. Then

\int_{a}^{b}f(x)D^{\alpha,\beta}_{M}g(x)d_{\alpha}x=f(x)g(x)\Bigr|_{a}^{b}-\int_{a}^{b}g(x)D^{\alpha,\beta}_{M}f(x)d_{\alpha}x

where d_{\alpha}x=\frac{\Gamma(\beta+1)}{x^{1-\alpha}}dx

Theorem

Theorem 3.1

7QYM

Let \Gamma(\beta+1)t^{\alpha-1} p(t),\Gamma(\beta+1)t^{\alpha-1}f(t)\in C(a,b) and let y be M-differentiable for 0<\alpha\leq 1 and \beta>0. Then the initial value problem

D^{\alpha,\beta}_{M}y+p(t)y=f(t)
y(t_{0})=y_{0}

has exactly one solution on the interval (a,b) where t_{0}\in(a,b)

Proof of Theorem 3.1

Using property (6) in Theorem 2.1, we have

D^{\alpha,\beta}_{M}y+p(t)y=f(t)
\frac{t^{1-\alpha}}{\Gamma(\beta+1)} y'+p(t)y=f(t)
y'+\Gamma(\beta+1)t^{\alpha-1}p(t)y=\Gamma(\beta+1)t^{\alpha-1}f(t)

The proof is clear from classical linear fundamental theorem existence and uniqueness.

Theorem

Theorem 3.2

I1ML

If \Gamma(\beta+1)t^{\alpha-1}p_{n-1}(t),...,\Gamma(\beta+1)t^{\alpha-1}p_{1}(t),\Gamma(\beta+1)t^{\alpha-1}p_{0}(t),\Gamma(\beta+1)t^{\alpha-1}f(t)\in C(a,b) and y be n times M-differentiable function, then a solution y(t) of the initial value problem

^{n}D^{\alpha,\beta}_{M}y+p_{n-1}(t)^{n-1}D^{\alpha,\beta}_{M}y+...+p_{2}(t)^{2}D^{\alpha,\beta}_{M}y+p_{1}(t)D^{\alpha,\beta}_{M}y+p_{0}(t)y=f(t)
y(t_{0})=y_{0},D^{\alpha,\beta}_{M}y(t_{0})=y_{1},...,^{n-1}D^{\alpha,\beta}_{M}y(t_{0})=y_{n-1}, a<t_{0}<b
Proof of Theorem 3.2

The existence of a local solution is obtained by transform our problem into the first order system of differential equations. So, we introduce new variables

x_{1}=y,x_{2}=D^{\alpha,\beta}_{M}y,x_{3}=^{2}D^{\alpha,\beta}_{M}y,,...,x_{n}=^{n-1}D^{\alpha,\beta}_{M}y

In this, we have

D^{\alpha,\beta}_{M}x_{1}=x_{2}
D^{\alpha,\beta}_{M}x_{2}=x_{3}
\vdots
D^{\alpha,\beta}_{M}x_{n-1}=x_{n}
D^{\alpha,\beta}_{M}x_{n}=-p_{n-1}x_{n}-...-p_{2} x_{3}-p_{1} x_{2}-p_{0} x_{1}+f(t)

The above equations can be written as the following

D^{\alpha,\beta}_{M}\underbrace{\begin{bmatrix} x_{1}\\x_{2}\\\vdots \\x_{n-1}\\x_{n} \end{bmatrix}}_{X(t)}+\underbrace{\begin{bmatrix} 0&-1&0&0&\cdots&0\\0&0&-1&0&\cdots&0\\\vdots&\vdots&\vdots&\vdots&\cdots&\vdots\\0&0&0&0&\cdots&-1\\p_{0}&p_{1}&p_{2}&p_{3}&\cdots&p_{n-1}\end{bmatrix}}_{P(t)} \begin{bmatrix} x_{1}\\x_{2}\\\vdots \\x_{n-1}\\x_{n} \end{bmatrix}=\underbrace{\begin{bmatrix} 0\\0\\\vdots \\0\\f(t) \end{bmatrix}}_{F(t)}
D^{\alpha,\beta}_{M}X(t)+P(t)X(t)=F(t)
X'(t)+\Gamma(\beta+1)t^{\alpha-1}P(t)X(t)=\Gamma(\beta+1)t^{\alpha-1}F(t)

The existence and uniqueness of solution (6)-(7) follows from classical theorems on existence and uniqueness for system equation.

Theorem

Theorem 3.3

Q26Y

If y_{1} and y_{2} are n times M-differentiable functions and c_{1},c_{2} are arbitrary numbers, then L_{\alpha,\beta} is linear.

i.e,\ L_{\alpha,\beta}[c_{1} y_{1}+c_{2} y_{2}]=c_{1} L_{\alpha,\beta}[y_{1} ]+c_{2}L_{\alpha,\beta}[y_{2}]
Used by Proof
Proof of Theorem 3.3

We can easily derived the proof of this theorem by applying same procedure in Theorem-4.3 [r5] to M- derivative.

Theorem

Theorem 3.4

9WFP

If y_{1},y_{2}...y_{n} are the solutions of equation L_{\alpha,\beta}[y]=0 and c_{1},c_{2}...c_{n} are arbitrary constants, then the linear combination y(t)=c_{1} y_{1}+c_{2} y_{2}+...+c_{n}y_{n} is also solution of L_{\alpha,\beta}[y]=0.

Used by Proof
Proof of Theorem 3.4

We can easily derived the proof of this theorem by applying same procedure in Theorem-4.4[r5] to M- derivative.

Definition

Definition 3.1

R0EW

For n functions y_{1},y_{2}...y_{n}, we define the M-Wronskain of these function to be the determinant

W_{\alpha,\beta}(t)=\begin{vmatrix} y_1&y_{2}&\cdots&y_{n}\\ D^{\alpha,\beta}_{M}y_{1}&D^{\alpha,\beta}_{M}y_{2}&\cdots&D^{\alpha,\beta}_{M}y_{n}\\ \vdots&\vdots&\cdots&\vdots\\ ^{n-1}D^{\alpha,\beta}_{M}y_{1}&^{n-1}D^{\alpha,\beta}_{M}y_{2}&\cdots&^{n-1}D^{\alpha,\beta}_{M}y_{n} \end{vmatrix}
Theorem

Theorem 3.5

9B90

Let y_{1},y_{2}...y_{n} be n solutions of L_{\alpha,\beta}[y]=0. If there is a t_{0}\in(a,b) such that W_{\alpha,\beta}(t_{0})\neq0, then {y_{1},y_{2}...y_{n}} is a fundamental set of solutions.

Used by Proof, Proof, Proof
Proof of Theorem 3.5

We need to show that if y(t) is a solution of L_{\alpha,\beta}[y]=0, then we can write y(t) as a linear combination of y_{1},y_{2}...y_{n}.

i.e, y=c_{1}y_{1}+c_{2}y_{2}+...+c_{n}y_{n}

so the problem reduces to finding the constants c_{1},c_{2},...c_{n}.These constants are found by solving the following linear system of n equations

c_{1} y_{1}(t_{0})+c_{2}y_{2}(t_{0})+...+c_{n}y_{n}(t_{0})=y(t_{0})
c_{1}D^{\alpha,\beta}_{M} y_{1}(t_{0})+c_{2}D^{\alpha,\beta}_{M}y_{2}(t_{0})+...+c_{n}D^{\alpha,\beta}_{M}y_{n}(t_{0})=D^{\alpha,\beta}_{M}y(t_{0})
\vdots\\
c_{1}^{n-1}D^{\alpha,\beta}_{M} y_{1}(t_{0})+c_{2}^{n-1}D^{\alpha,\beta}_{M}y_{2}(t_{0})+...+c_{n}^{n-1}D^{\alpha,\beta}_{M}y_{n}(t_{0})=^{n-1}D^{\alpha,\beta}_{M}y(t_{0})

Using Cramer’s rule, we can find

c_{i}=\frac{W_{\alpha,\beta}^{i}(t_{0})}{W_{\alpha,\beta}(t_{0})}, 1\leq i \leq n

Since W_{\alpha,\beta}(t_{0})\neq 0, it follows that c_{1},c_{2},...c_{n} exist.

Theorem

Theorem 3.6

TJTI

Let y_{1},y_{2}...y_{n} be n solutions of L_{\alpha,\beta}[y]=0. Then

  1. W_{\alpha,\beta}(t) satisfies the differential equationD^{\alpha,\beta}_{M}W_{\alpha,\beta}+p_(n-1) W_{\alpha,\beta}=0

  2. If t_{0} is any point in (a,b), then

    W_{\alpha,\beta}(t)=W_{\alpha,\beta}(t_{0})e^{-\Gamma(\beta+1)\int_{t_{0}}^{t}x^{\alpha-1}p_{n-1}(x)dx}

    Further, if W_{\alpha,\beta}(t_{0})\neq0 then W_{\alpha,\beta}(t)\neq0 for all t\in(a,b)

Proof of Theorem 3.6

(1)Let us introduce new variables

x_{1}=y,x_{2}=D^{\alpha,\beta}_{M}y,x_{3}=^{2}D^{\alpha,\beta}_{M}y,...,x_{n}=^{n-1}D^{\alpha,\beta}_{M}y

From this, we have

D^{\alpha,\beta}_{M}x_{1}=x_{2}
D^{\alpha,\beta}_{M}x_{2}=x_{3}
\vdots
D^{\alpha,\beta}_{M}x_{n-1}=x_{n}
D^{\alpha,\beta}_{M}x_{n}=-p_{n-1}x_{n}-...-p_{2} x_{3}-p_{1} x_{2}-p_{0} x_{1}
D^{\alpha,\beta}_{M}\underbrace{\begin{bmatrix} x_{1}\\x_{2}\\\vdots \\x_{n-1}\\x_{n} \end{bmatrix}}_{X(t)}=\underbrace{\begin{bmatrix} 0&-1&0&0&\cdots&0\\0&0&-1&0&\cdots&0\\\vdots&\vdots&\vdots&\vdots&\cdots&\vdots\\0&0&0&0&\cdots&-1\\-p_{0}&-p_{1}&-p_{2}&-p_{3}&\cdots&-p_{n-1}\end{bmatrix}}_{P(t)} \begin{bmatrix} x_{1}\\x_{2}\\\vdots \\x_{n-1}\\x_{n} \end{bmatrix}=\underbrace{\begin{bmatrix} 0\\0\\\vdots \\0\\f(t) \end{bmatrix}}_{F(t)}
D^{\alpha,\beta}_{M}X(t)=P(t)X(t)

We have

D^{\alpha,\beta}_{M}W_{\alpha,\beta}(t)=(a_{11}+a_{22}+...+a_{nn})W_{\alpha,\beta}(t)

In our case

a_{11}+a_{22}+...+a_{nn}=-p_{n-1}(t)

So,

D^{\alpha,\beta}_{M}W_{\alpha,\beta}(t)+p_{n-1}(t)W_{\alpha,\beta}(t)=0

(2)The above differential equation can be solved by the method of integrating factor, we have

W_{\alpha,\beta}(t)=W_{\alpha,\beta}(t_{0})e^{-\Gamma(\beta+1)\int_{t_{0}}^{t}x^{\alpha-1}p_{n-1}(x)dx}

Thus the proof of theorem is completed.

Theorem

Theorem 3.7

L4TJ

If \{y_{1},y_{2}...y_{n}\} is a fundamental set of solutions of L_{\alpha,\beta}[y]=0 where \Gamma(\beta+1)t^{\alpha-1}p_{n-1}(t)...\Gamma(\beta+1)t^{\alpha-1}p_{1}(t),\Gamma(\beta+1)t^{\alpha-1}p_{0}(t)\in C(a,b),then W_{\alpha,\beta}(t)\neq 0 for all t\in (a,b).

Proof of Theorem 3.7

By applying procedure in Theorem-4.8 [5] to M-derivative, we can easily prove this theorem.

Theorem

Theorem 3.8

Q49V

Let \Gamma(\beta+1)t^{\alpha-1}p_{n-1}(t),...,\Gamma(\beta+1)t^{\alpha-1}p_{1}(t), \Gamma(\beta+1)t^{\alpha-1}p_{0}(t)\in C(a,b). The solution set \{y_{1},y_{2},...,y_{n}\} is a fundamental set of solutions to the equation L_{\alpha,\beta}[y]=0 if and only if the functions y_{1},y_{2},...,y_{n} are linearly independent.

Proof of Theorem 3.8

By applying procedure in Theorem-4.9 [r5] to M-derivative, we can easily prove this theorem.

Theorem

Theorem 3.9

GISP

Let y_{1},y_{2},...,y_{n} be a fundamental set of solutions of the equation (1) and y_{p} be any particular solution of the non homogeneous equation (2). Then the general solution of the equation is y=c_{1}y_{1}+c_{2}y_{2}+...+c_{n}y_{n}+y_{p}

Proof of Theorem 3.9

Let L_{\alpha,\beta} be the differential operator and y(t) and y_p(t) be the solutions of the non homogeneous equation L_{\alpha,\beta}[y]=f(t). If we take u(t)=y(t)-y_{p}(t), then by linearity of L_{\alpha,\beta} we have,

L_{\alpha,\beta}[u]=L_{\alpha,\beta}[y(t)-y_{p}(t)]=L_{\alpha,\beta}[y(t)]-L_{\alpha,\beta}[y_p (t)]=f(t)-f(t)=0

Then u(t) is a solution of the homogenous equation L_{\alpha,\beta}[y]=0. Then by Theorem 3.4

u(t)=c_{1}y_{1}(t)+c_{2}y_{2}(t)+...+c_{n}y_{n}(t)

i.e,

y(t)-y_{p}(t)=c_{1}y_{1}(t)+c_{2}y_{2}(t)+...+c_{n}y_{n}(t)

Then

y(t)=c_{1} y_{1}(t)+c_{2} y_{2}(t)+...+c_{n}y_{n}(t)+y_{p}(t)
Lemma

Lemma 4.1

VUPJ

Suppose that L_{\alpha,\beta}[.] is a linear operator with constant coefficients and \alpha\in(0,1] and \beta>0, then for t>0

L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]=P_{n}(r)[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]

Where P_{n}(r)=r^{n}+P_{n-1}r^{n-1}+...+P_{0} and r is a real or complex constant

Proof of Lemma 4.1

M-derivatives of y=e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}} are

D^{\alpha,\beta}_{M}y=re^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}, ^{2}D^{\alpha,\beta}_{M} y=r^2 e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}},...,^{n}D^{\alpha,\beta}_{M} y=r^n e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}

We substitute y=e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}} and Eqn.(10) in L_{\alpha,\beta}[y]

L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]=(^{n}D^{\alpha,\beta}_{M}+p_{n-1}^{n-1}D^{\alpha,\beta}_{M}+...+p_{2}^{2}D^{\alpha,\beta}_{M}+p_{1}D^{\alpha,\beta}_{M}+p_{0})e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}
=(r^{n}+P_{n-1}r^{n-1}+...+P_{0})e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}
L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]=p_{n}(r)e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}

Hence, the proof is completed.

Lemma

Lemma 4.2

0TH8

Let r be a root of the characteristic equation (11), then

\frac{\partial}{\partial r}\{L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]\}=L_{\alpha,\beta} [\frac{\partial}{\partial r}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]

and \frac{\partial^{l}}{\partial r^{l}}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}} = (\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha})^{l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}} where l is integer.

Used by Proof
Proof of Lemma 4.2

From Theorem 3.3 it follows that L_{\alpha,\beta}[.] is linear and also \frac{\partial}{\partial r} is linear by property of classical derivative. Hence

\frac{\partial}{\partial r}\{L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]\}=L_{\alpha,\beta} [\frac{\partial}{\partial r}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]

Additionally, from classical derivative, it follows that

\frac{\partial^{l}}{\partial r^{l}}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}} = (\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha})^{l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}
Lemma

Lemma 4.3

7BYN

If r_{1} is a root of multiplicity of \mu_{1}of the characteristic equation (11), then the functions y_{1,l}(t) , where l=0,1,...,\mu_{l-1} such that

y_{1,l} = (\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha})^{l}e^{\frac{r_{1}\Gamma(\beta+1)}{\alpha}t^{\alpha}}

are solutions of Eq.(8).

Used by Proof, Proof
Proof of Lemma 4.3

Consider L_{\alpha,\beta}[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}]=p_{n}(r)e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}. From Lemma 4.2 and applying classical Leibniz rule it follows that

\Bigg\{L_{\alpha,\beta}\bigg[\frac{\partial^l}{\partial r^l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg\}_{r=r_{1}}=\Bigg\{\frac{\partial^l}{\partial r^l}\Bigg[L_{\alpha,\beta}\bigg[e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg]\Bigg\}_{r=r_{1}}=\Bigg\{\frac{\partial^l}{\partial r^l}\bigg[p_{n}(r)e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg\}_{r=r_{1}}
\Bigg\{L_{\alpha,\beta}\bigg[\frac{\partial^l}{\partial r^l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg\}_{r=r_{1}}=\sum_{j=0}^{l} \begin{pmatrix} l\\j\end{pmatrix}\bigg[\frac{\partial^{l-j}}{\partial r^{l-j}}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]_{r=r_{1}}\frac{\partial^{j}}{\partial r^{j}}\big[P_{n}(r)\big]_{r=r_{1}}

Since \frac{\partial^{j}}{\partial r^{j}}\big[P_{n}(r)\big]_{r=r_{1}}=0 for j=0,1,...,\mu_{1}-1

\Bigg\{L_{\alpha,\beta}\bigg[\frac{\partial^l}{\partial r^l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg\}_{r=r_{1}}=0

From Lemma 4.2

\Bigg\{L_{\alpha,\beta}\bigg[\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{r\Gamma(\beta+1)}{\alpha}t^{\alpha}}\bigg]\Bigg\}_{r=r_{1}}=0
L_{\alpha,\beta}\Big[y_{1,l}(t)\Big]=0

Hence y_{1,l}(t) are solutions of Eq.(8).

Corollary

Corollary 4.1

3GOP

Let r_{j} ,j=1,2,...,k are distinct roots of multiplicity \mu_{j},j=1,2,...,k of the characteristic Eq.(5). Then the following functions

\bigcup_{j=1}^{k}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{r_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Bigg\}_{l=0}^{\mu_{j}-1}
Used by Proof
Proof of Corollary 4.1

Corollary 4.1 follows from Lemma 4.3 and Theorem 3.5.

Lemma

Lemma 4.4

NWO1

If r_{1} and \bar{r_{1}} (r_{1}=a+ib,b\neq0) are complex roots of multiplicity \sigma_{1} of the characteristic equation (11), then for l=0,1,...,\sigma_{1}-1, the functions

y_{1,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)+isin\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]

and

y_{2,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)-isin\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]

are linearly independent solutions of Eq.(8).

Used by Proof
Proof of Lemma 4.4

Since r_{1}=a+ib is a root of multiplicity \sigma_{1} of the characteristic equation (11), From Lemma 4.3 and using Euler’s identity it follows that, the functions

y_{1,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{(a+ib)\Gamma(\beta+1)}{\alpha}t^{\alpha}}

i.e

y_{1,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)+isin\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]

are solutions of the Eq.(8). Similarly, for \bar{r_{1}}=a-ib, the functions

y_{2,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{(a-ib)\Gamma(\beta+1)}{\alpha}t^{\alpha}}

i.e

y_{2,l}(t)=\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)-isin\Big(\frac{b\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]

are solutions of the Eq.(8). Hence proof is completed.

Corollary

Corollary 4.2

GWB3

If \big\{r_{j},\bar{r_{j}}\big\}_{j=1}^{m}, r_{j}=a_{j}+ib_{j}, b_{j}\neq0 distinct 2m roots of multiplicity \big\{\sigma_{j}\big\}_{j=1}^{m} of the characteristic equation (11),then, the functions

\bigcup_{j=1}^{m}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)+isin\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]\Bigg\}_{l=0}^{\sigma_{j}-1}

and

\bigcup_{j=1}^{m}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)-isin\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]\Bigg\}_{l=0}^{\sigma_{j}-1}
Used by Proof
Proof of Corollary 4.2

To prove corollary 4.2, it is sufficient to apply the Lemma 4.4 and Theorem 3.5.

Theorem

Theorem 4.1

95BV

If \big\{r_{j}\big\}_{j=1}^{k} are distinct k roots of multiplicity \big\{\mu_{j}\big\}_{j=1}^{k} and \big\{\lambda_{j},\bar{\lambda_{j}}\big\}_{j=1}^{m}, \lambda_{j}=a_{j}+ib_{j},b_{j}\neq 0 are distinct 2m roots of multiplicity \big\{\sigma_{j}\big\}_{j=1}^{m} of the characteristic equation (11) such that \sum_{j=1}^{k}\mu_{j}+2\sum_{j=1}^{m}\sigma_{j}=n, then the functions

\bigcup_{j=1}^{k}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{r_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Bigg\}_{l=0}^{\mu_{j}-1}
\bigcup_{j=1}^{m}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)+isin\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]\Bigg\}_{l=0}^{\sigma_{j}-1}

and

\bigcup_{j=1}^{m}\Bigg\{\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)^{l}e^{\frac{a_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)-isin\Big(\frac{b_{j}\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]\Bigg\}_{l=0}^{\sigma_{j}-1}

are the fundamental set of solutions of the equation (8).

Proof of Theorem 4.1

The proof of the Theorem 4.1 is follows from Corollary 4.1, Corollary 4.2 and Theorem 3.5.

Example

Example 4.1

ZWX4
^{2}D^{\alpha,\beta}_{M}y+4D^{\alpha,\beta}_{M}y+3y=0

The characteristic equation of (12) is

r^{2}+4r+3=0

Therefore, the roots are r=-3 and r=-1
Hence, the general solution is

y(t)=c_{1}e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+c_{2} e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}
Example

Example 4.2

7X52
^{2}D^{\alpha,\beta}_{M}y-4D^{\alpha,\beta}_{M}y+4y=0

The characteristic equation of (13) is

r^{2}-4r+4=0

The roots are r_{1,2}=2
Hence, the general solution is

y(t)=\Bigg(c_{1}+c_{2}\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Bigg)e^{\frac{2\Gamma(\beta+1)}{\alpha}t^{\alpha}}
Example

Example 4.3

PV69
^{2}D^{\alpha,\beta}_{M}y+4D^{\alpha,\beta}_{M}y+5y=0

The characteristic equation of (14) is

r^{2}+4r+5=0

The roots are r_{1}=-2+i and r_{2}=-2-i
Hence, the general solution is

y(t)=e^{\frac{-2\Gamma(\beta+1)}{\alpha}t^{\alpha}}\Big[cos\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)+isin\Big(\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}\Big)\Big]
Theorem

Theorem 5.1

9WOZ

If u(t) is a solution of homogeneous case of the equation (15) such that

u(t)=\sum_{i=1}^{n}c_{i} y_{i}(t)

then particular solution of the equation (15) is

v(t)=\sum_{i=1}^{n}c_{i}(t)y_{i}(t)

Where c_{1}(t),c_{2}(t),...,c_{n}(t) provide following system of equations

\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)y_{i}(t)=0
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)=0
\vdots
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-2}D^{\alpha,\beta}_{M}y_{i}(t)=0
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)=f(t)
Proof of Theorem 5.1

The solution of the equation (15) is in the form

v(t)=\sum_{i=1}^{n}c_{i}(t)y_{i}(t)

The M-derivative of v(t) for \alpha \in (0,1] and \beta>0 will be

D^{\alpha,\beta}_{M}v(t)=\sum_{i=1}^{n}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)+ \sum_{i=1}^{n}D^{\alpha,\beta}_{M} c_{i}(t)y_{i}(t)

Applying the first condition \sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)y_{i}(t)=0, we obtain

D^{\alpha,\beta}_{M}v(t)=\sum_{i=1}^{n}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)

If we calculate the M-derivative of D^{\alpha,\beta}_{M}v(t) for \alpha\in(0,1] and \beta>0, then we get

^{2}D^{\alpha,\beta}_{M}v(t)=\sum_{i=1}^{n}c_{i}(t)^{2}D^{\alpha,\beta}_{M}y_{i}(t)+ \sum_{i=1}^{n}D^{\alpha,\beta}_{M} c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)

Apply second condition \sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)=0, we obtain

^{2}D^{\alpha,\beta}_{M}v(t)=\sum_{i=1}^{n}c_{i}(t)^{2}D^{\alpha,\beta}_{M}y_{i}(t)

By continuing in this way, we get

^{n-1}D^{\alpha,\beta}_{M}v(t)=\sum_{i=1}^{n}c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+ \sum_{i=1}^{n}D^{\alpha,\beta}_{M} c_{i}(t)^{n-2}D^{\alpha,\beta}_{M}y_{i}(t)

We substitute v(t),^{2}D^{\alpha,\beta}_{M}v(t),..., ^{n}D^{\alpha,\beta}_{M}v(t) in the equation (15), we have

\begin{gathered} \sum_{i=1}^{n}c_{i}(t)^{n}D^{\alpha,\beta}_{M}y_{i}(t)+\sum_{i=1}^{n}D^{\alpha,\beta}_{M} c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+p_{n-1}\sum_{i=1}^{n} c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+...+ \\p_{2}\sum_{i=1}^{n}c_{i}(t)^{2}D^{\alpha,\beta}_{M}y_{i}(t)+p_{1}\sum_{i=1}^{n}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)+p_{0}\sum_{i=1}^{n} c_{i}(t)y_{i}(t)=f(t) \end{gathered}
\begin{gathered} \sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+\sum_{i=1}^{n} c_{i}(t)\big[^{n}D^{\alpha,\beta}_{M}y_{i}(t)+p_{n-1}^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+...+\\p_{1}D^{\alpha,\beta}_{M}y_{i}(t)+p_{0}y_{i}(t)\big]=f(t) \end{gathered}

Since y_{1}(t),y_{2}(t),...,y_{n}(t) are solutions of homogeneous case of equation (8), then

\sum_{i=1}^{n} c_{i}(t)\big[^{n}D^{\alpha,\beta}_{M}y_{i}(t)+p_{n-1}^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)+...+\\p_{1}D^{\alpha,\beta}_{M}y_{i}(t)+p_{0}y_{i}(t)\big]=0

We obtain n^{th} condition as

\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)=f(t)

Hence we obtain the following system

\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)y_{i}(t)=0
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)D^{\alpha,\beta}_{M}y_{i}(t)=0
\vdots
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-2}D^{\alpha,\beta}_{M}y_{i}(t)=0
\sum_{i=1}^{n}D^{\alpha,\beta}_{M}c_{i}(t)^{n-1}D^{\alpha,\beta}_{M}y_{i}(t)=f(t)

Solving the above system (17) provides D^{\alpha,\beta}_{M}c_{i}(t), i=1,2,...,n. Therefore we can write the particular solution of equation (15) as v(t)=\sum_{i=1}^{n}c_{i}(t)y_{i}(t)

Example

Example 5.1

X5N7
^{2}D^{\alpha,\beta}_{M}y+4D^{\alpha,\beta}_{M}y+3y=f(t)

(a) Let f(t)=e^{2t^{\alpha}}. For v(t)=c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}, the system of equations are built by the conditions as following

D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=0
-3D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}-D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=e^{2t^{\alpha}}

Solving the above system of equations and using M-integral we obtain c_{1}(t)=\frac{-\Gamma(\beta+1)}{4\alpha+6\Gamma(\beta+1)} e^{\frac{2\alpha+3\Gamma(\beta+1)}{\alpha}t^{\alpha}},c_{2}(t)=\frac{\Gamma(\beta+1)}{4\alpha+2\Gamma(\beta+1)} e^{\frac{2\alpha+\Gamma(\beta+1)}{\alpha}t^{\alpha}}. Then particular solution v(t) is

v(t)=\frac{\Gamma(\beta+1)^{2}}{4\alpha^{2}+8\alpha\Gamma(\beta+1)+3\Gamma(\beta+1)^{2}} e^{2t^{\alpha}}

(b) Let f(t)=2t^{2\alpha}+t^{\alpha}-3. The system of equations for this case is

D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=0
-3D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}-D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=2t^{2\alpha}+t^{\alpha}-3

Solve this system of equations, we have

\begin{gathered} c_{1}(t)=\frac{-1}{3}t^{2\alpha}e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+\Big(\frac{4\alpha-3\Gamma(\beta+1)}{18\Gamma(\beta+1)}\Big) t^{\alpha} e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}}\\ +\Bigg(\frac{-4\alpha^{2}+3\alpha\Gamma(\beta+1)+27\Gamma(\beta+1)^{2}}{54\Gamma(\beta+1)^{2}}\Bigg)e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}} \end{gathered}
\begin{gathered} c_{2}(t)=t^{2\alpha}e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}}+\Big(\frac{\Gamma(\beta+1)-4\alpha}{2\Gamma(\beta+1)}\Big) t^{\alpha} e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}}\\ +\Bigg(\frac{4\alpha^{2}-\alpha\Gamma(\beta+1)-3\Gamma(\beta+1)^{2}}{2\Gamma(\beta+1)^{2}}\Bigg)e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}} \end{gathered}

Hence, particular solution v(t) is obtained by

v(t)=\frac{2}{3}t^{2\alpha}+\Big(\frac{3\Gamma(\beta+1)-16\alpha}{9\Gamma(\beta+1)}\Big) t^{\alpha} e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}}\\ +\Bigg(\frac{52\alpha^{2}-12\alpha\Gamma(\beta+1)-27\Gamma(\beta+1)^{2}}{27\Gamma(\beta+1)^{2}}\Bigg)e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}}

(c) Let f(t)=sin2t^{\alpha}. The system of equations for this case is

D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=0
-3D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}-D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=sin2t^{\alpha}

Solve this system of equations, we have

c_{1}(t)=\frac{-3\Gamma(\beta+1)^{2}}{8\alpha^{2}+18\Gamma(\beta+1)^2} e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}}sin2t^{\alpha} +\frac{\alpha\Gamma(\beta+1)}{4\alpha^{2}+9\Gamma(\beta+1)^{2}}e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}}cos2t^{\alpha}
c_{2}(t)=\frac{\Gamma(\beta+1)^{2}}{8\alpha^{2}+2\Gamma(\beta+1)^2} e^{\frac{\Gamma(\beta+1)}{\alpha}t^{\alpha}}sin2t^{\alpha}\\ -\frac{\alpha\Gamma(\beta+1)}{4\alpha^{2}+\Gamma(\beta+1)^{2}}e^{\frac{3\Gamma(\beta+1)}{\alpha}t^{\alpha}}cos2t^{\alpha}

Hence, particular solution v(t) is obtained by

\begin{gathered} v(t)=\frac{-4\alpha^{2}\Gamma(\beta+1)^{2}+3\Gamma(\beta+1)^{4}}{16\alpha^{4}+40\alpha^{2}\Gamma(\beta+1)^{2}+9\Gamma(\beta+1)^4}sin2t^{\alpha}\\ -\frac{8\alpha\Gamma(\beta+1)^{3}}{16\alpha^{4}+40\alpha^{2}\Gamma(\beta+1)^{2}+9\Gamma(\beta+1)^4}cos2t^{\alpha} \end{gathered}

(d) Let f(t)=e^{2t^{\alpha}}t^{\alpha}. The system of equations for this case is

D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=0
-3D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}-D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=e^{2t^{\alpha}}t^{\alpha}

Solve this system of equations, we have

c_{1}(t)=\frac{-\Gamma(\beta+1)}{4\alpha+6\Gamma(\beta+1)}t^{\alpha}e^{\frac{2\alpha+3\Gamma(\beta+1)}{\alpha}t^{\alpha}} +\frac{\alpha\Gamma(\beta+1)}{2\big(2\alpha+3\Gamma(\beta+1)\big)^{2}}e^{\frac{2\alpha+3\Gamma(\beta+1)}{\alpha}t^{\alpha}}
c_{2}(t)=\frac{\Gamma(\beta+1)}{4\alpha+2\Gamma(\beta+1)}t^{\alpha}e^{\frac{2\alpha+\Gamma(\beta+1)}{\alpha}t^{\alpha}} -\frac{\alpha\Gamma(\beta+1)}{2\big(2\alpha+\Gamma(\beta+1)\big)^{2}}e^{\frac{2\alpha+\Gamma(\beta+1)}{\alpha}t^{\alpha}}

Hence, particular solution v(t) is obtained by

v(t)=\frac{\Gamma(\beta+1)^{2}}{4\alpha^{2}+8\alpha\Gamma(\beta+1)+3\Gamma(\beta+1)^2}t^{\alpha}e^{2t^{\alpha}}-\frac{4\alpha^{2}\Gamma(\beta+1)^{2}+4\alpha\Gamma(\beta+1)^{3}}{\big(4\alpha^{2}+8\alpha\Gamma(\beta+1)+3\Gamma(\beta+1)^{2}\big)^{2}}e^{2t^{\alpha}}

(e) Let f(t)=e^{-4t^{\alpha}}. Take \alpha\neq \frac{3}{4} and \alpha\neq \frac{1}{4}, the system of equations for this case is

D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}+D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=0
-3D^{\alpha,\beta}_{M}c_{1}(t) e^{\frac{-3\Gamma(\beta+1)}{\alpha}t^{\alpha}}-D^{\alpha,\beta}_{M}c_{2}(t) e^{\frac{-\Gamma(\beta+1)}{\alpha}t^{\alpha}}=e^{-4t^{\alpha}}

Solve this system of equations, we have

c_{1}(t)=\frac{\Gamma(\beta+1)}{8\alpha-6\Gamma(\beta+1)}e^{\frac{3\Gamma(\beta+1)-4\alpha}{\alpha}t^{\alpha}}
c_{2}(t)=\frac{\Gamma(\beta+1)}{2\Gamma(\beta+1)-8\alpha}e^{\frac{\Gamma(\beta+1)-4\alpha}{\alpha}}t^{\alpha}

Hence, we obtain particular solution v(t) as following:

v(t)=\frac{\Gamma(\beta+1)^{2}}{16\alpha^{2}-16\alpha\Gamma(\beta+1)+3\Gamma(\beta+1)^{2}}e^{-4t^{\alpha}}

Take \alpha=\frac{3}{4} and v(t)=c_{1}(t)e^{-4\Gamma(\beta+1)t^{\frac{3}{4}}}+c_{2}(t) e^{\frac{-4\Gamma(\beta+1)}{3}t^{\frac{3}{4}}}
The system of equations is

D^{\frac{3}{4},\beta}_{M}c_{1}(t) e^{-4\Gamma(\beta+1)t^{\frac{3}{4}}}+D^{\frac{3}{4},\beta}_{M}c_{2}(t) e^{\frac{-4\Gamma(\beta+1)}{3}t^{\frac{3}{4}}}=0
-3D^{\frac{3}{4},\beta}_{M}c_{1}(t) e^{-4\Gamma(\beta+1)t^{\frac{3}{4}}}-D^{\frac{3}{4},\beta}_{M}c_{2}(t) e^{\frac{-4\Gamma(\beta+1)}{3}t^{\frac{3}{4}}}=e^{-4t^{\frac{3}{4}}}

We solve the above equation, c_{1}(t)=-\frac{\Gamma(\beta+1)}{-6+6\Gamma(\beta+1)} e^{\big(-4+4\Gamma(\beta+1)\big)t^{\frac{3}{4}}} and c_{2}(t)=\frac{\Gamma(\beta+1)}{-6+2\Gamma(\beta+1)}e^{\big(-12+4\Gamma(\beta+1)\big)t^{\frac{3}{4}}} is obtained.
The particular solution is v(t)=\frac{\Gamma(\beta+1)^{2}}{9-12\Gamma(\beta+1)+3\Gamma(\beta+1)^{2}}e^{-4t^{\frac{3}{4}}}

Take \alpha=\frac{1}{4} and v(t)=c_{1}(t)e^{-12\Gamma(\beta+1)t^{\frac{1}{4}}}+c_{2}(t) e^{-4\Gamma(\beta+1)t^{\frac{1}{4}}}
The system of equations is

D^{\frac{1}{4},\beta}_{M}c_{1}(t) e^{-12\Gamma(\beta+1)t^{\frac{1}{4}}}+D^{\frac{1}{4},\beta}_{M}c_{2}(t) e^{-4\Gamma(\beta+1)t^{\frac{1}{4}}}=0
-3D^{\frac{1}{4},\beta}_{M}c_{1}(t) e^{-12\Gamma(\beta+1)t^{\frac{1}{4}}}-D^{\frac{1}{4},\beta}_{M}c_{2}(t) e^{-4\Gamma(\beta+1)t^{\frac{1}{4}}}=e^{-4t^{\frac{1}{4}}}

We solve the above equation,c_{1}(t)=-\frac{\Gamma(\beta+1)}{-2+6\Gamma(\beta+1)} e^{\big(-4+12\Gamma(\beta+1)\big)t^{\frac{1}{4}}} and c_{2}(t)=\frac{\Gamma(\beta+1)}{-2+2\Gamma(\beta+1)}e^{\big(-4+4\Gamma(\beta+1)\big)t^{\frac{1}{4}}} is obtained.
The particular solution is v(t)=\frac{4\Gamma(\beta+1)^{2}}{4-16\Gamma(\beta+1)+12\Gamma(\beta+1)^{2}}e^{-4t^{\frac{1}{4}}}

Source and licence

The text on this page is from arXiv:1903.11969v1 by V. Padmapriya and M. Kaliyappan, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/1903/1903.11969.jsonl.