Over the past 20 years, there has been a surge of diabetes cases in Iraq. Blood tests administered in the absence of professional medical judgment have allowed for the early detection of diabetes, which will fasten disease detection and lower medical costs. This work focuses on the use of a Long-Short Term Memory (LSTM) neural network for diabetes classification in Iraq. Some medical tests and body features were used as classification features. The most relevant features were selected using the Binary Dragon Fly Algorithm (BDA) Binary version of the selection method because the features either selected or not. To reduce the number of features that are used in prediction, features without effects will be eliminated. This effects the classification accuracy, which is very important in both the computation time of the method and the cost of medical test that the individual will take during annual check ups.This work found out that among 11 features, only five features are most relevant to the disease. These features provide a classification accuracy up to 98% among three classes: diabetic, non diabetic and pre-diabetic.
Citation: Zaineb M. Alhakeem, Heba Hakim, Ola A. Hasan, Asif Ali Laghari, Awais Khan Jumani, Mohammed Nabil Jasm. Prediction of diabetic patients in Iraq using binary dragonfly algorithm with long-short term memory neural network[J]. AIMS Electronics and Electrical Engineering, 2023, 7(3): 217-230. doi: 10.3934/electreng.2023013
[1] | Salim A. Messaoudi, Mohammad M. Al-Gharabli, Adel M. Al-Mahdi, Mohammed A. Al-Osta . A coupled system of Laplacian and bi-Laplacian equations with nonlinear dampings and source terms of variable-exponents nonlinearities: Existence, uniqueness, blow-up and a large-time asymptotic behavior. AIMS Mathematics, 2023, 8(4): 7933-7966. doi: 10.3934/math.2023400 |
[2] | Adel M. Al-Mahdi, Mohammad M. Al-Gharabli, Maher Nour, Mostafa Zahri . Stabilization of a viscoelastic wave equation with boundary damping and variable exponents: Theoretical and numerical study. AIMS Mathematics, 2022, 7(8): 15370-15401. doi: 10.3934/math.2022842 |
[3] | Xiongmei Fan, Sen Ming, Wei Han, Zikun Liang . Lifespan estimate of solution to the semilinear wave equation with damping term and mass term. AIMS Mathematics, 2023, 8(8): 17860-17889. doi: 10.3934/math.2023910 |
[4] | Mohammad Kafini, Jamilu Hashim Hassan, Mohammad M. Al-Gharabli . Decay result in a problem of a nonlinearly damped wave equation with variable exponent. AIMS Mathematics, 2022, 7(2): 3067-3082. doi: 10.3934/math.2022170 |
[5] | Adel M. Al-Mahdi . The coupling system of Kirchhoff and Euler-Bernoulli plates with logarithmic source terms: Strong damping versus weak damping of variable-exponent type. AIMS Mathematics, 2023, 8(11): 27439-27459. doi: 10.3934/math.20231404 |
[6] | Noufe H. Aljahdaly . Study tsunamis through approximate solution of damped geophysical Korteweg-de Vries equation. AIMS Mathematics, 2024, 9(5): 10926-10934. doi: 10.3934/math.2024534 |
[7] | Mohammad Kafini, Mohammad M. Al-Gharabli, Adel M. Al-Mahdi . Existence and stability results of nonlinear swelling equations with logarithmic source terms. AIMS Mathematics, 2024, 9(5): 12825-12851. doi: 10.3934/math.2024627 |
[8] | Zhiqiang Li . The finite time blow-up for Caputo-Hadamard fractional diffusion equation involving nonlinear memory. AIMS Mathematics, 2022, 7(7): 12913-12934. doi: 10.3934/math.2022715 |
[9] | Muhammad Shakeel, Amna Mumtaz, Abdul Manan, Marouan Kouki, Nehad Ali Shah . Soliton solutions of the nonlinear dynamics in the Boussinesq equation with bifurcation analysis and chaos. AIMS Mathematics, 2025, 10(5): 10626-10649. doi: 10.3934/math.2025484 |
[10] | Khaled Kefi, Nasser S. Albalawi . Three weak solutions for degenerate weighted quasilinear elliptic equations with indefinite weights and variable exponents. AIMS Mathematics, 2025, 10(2): 4492-4503. doi: 10.3934/math.2025207 |
Over the past 20 years, there has been a surge of diabetes cases in Iraq. Blood tests administered in the absence of professional medical judgment have allowed for the early detection of diabetes, which will fasten disease detection and lower medical costs. This work focuses on the use of a Long-Short Term Memory (LSTM) neural network for diabetes classification in Iraq. Some medical tests and body features were used as classification features. The most relevant features were selected using the Binary Dragon Fly Algorithm (BDA) Binary version of the selection method because the features either selected or not. To reduce the number of features that are used in prediction, features without effects will be eliminated. This effects the classification accuracy, which is very important in both the computation time of the method and the cost of medical test that the individual will take during annual check ups.This work found out that among 11 features, only five features are most relevant to the disease. These features provide a classification accuracy up to 98% among three classes: diabetic, non diabetic and pre-diabetic.
We consider the Cauchy problem for the nonlinear damped wave equation with linear damping
utt−Δu+ut=|u|p+|∇u|q+w(x),t>0,x∈RN, | (1.1) |
where N≥1, p,q>1, w∈L1loc(RN), w≥0 and w≢0. Namely, we are concerned with the existence and nonexistence of global weak solutions to (1.1). We mention below some motivations for studying problems of type (1.1).
Consider the semilinear damped wave equation
utt−Δu+ut=|u|p,t>0,x∈RN | (1.2) |
under the initial conditions
(u(0,x),ut(0,x))=(u0(x),u1(x)),x∈RN. | (1.3) |
In [12], a Fujita-type result was obtained for the problem (1.2) and (1.3). Namely, it was shown that,
(ⅰ) if 1<p<1+2N and ∫RNuj(x)dx>0, j=0,1, then the problem (1.2) and (1.3) admits no global solution;
(ⅱ) if 1+2N<p<N for N≥3, and 1+2N<p<∞ for N∈{1,2}, then the problem (1.2) and (1.3) admits a unique global solution for suitable initial values.
The proof of part (ⅰ) makes use of the fundamental solution of the operator (∂tt−Δ+∂t)k. In [15], it was shown that the exponent 1+2N belongs to the blow-up case (ⅰ). Notice that 1+2N is also the Fujita critical exponent for the semilinear heat equation ut−Δu=|u|p (see [3]). In [6], the authors considered the problem
utt+(−1)m|x|αΔmu+ut=f(t,x)|u|p+w(t,x),t>0,x∈RN | (1.4) |
under the initial conditions (1.3), where m is a positive natural number, α≥0, and f(t,x)≥0 is a given function satisfying a certain condition. Using the test function method (see e.g. [11]), sufficient conditions for the nonexistence of a global weak solution to the problem (1.4) and (1.3) are obtained. Notice that in [6], the influence of the inhomogeneous term w(t,x) on the critical behavior of the problem (1.4) and (1.3) was not investigated. Recently, in [5], the authors investigated the inhomogeneous problem
utt−Δu+ut=|u|p+w(x),t>0,x∈RN, | (1.5) |
where w∈L1loc(RN) and w≢0. It was shown that in the case N≥3, the critical exponent for the problem (1.5) jumps from 1+2N (the critical exponent for the problem (1.2)) to the bigger exponent 1+2N−2. Notice that a similar phenomenon was observed for the inhomogeneous semilnear heat equation ut−Δu=|u|p+w(x) (see [14]). For other results related to the blow-up of solutions to damped wave equations, see, for example [1,2,4,7,8,9,13] and the references therein.
In this paper, motivated by the above mentioned works, our aim is to study the effect of the gradient term |∇u|q on the critical behavior of problem (1.5). We first obtain the Fujita critical exponent for the problem (1.1). Next, we determine its second critical exponent in the sense of Lee and Ni [10].
Before stating the main results, let us provide the notion of solutions to problem (1.1). Let Q=(0,∞)×RN. We denote by C2c(Q) the space of C2 real valued functions compactly supported in Q.
Definition 1.1. We say that u=u(t,x) is a global weak solution to (1.1), if
(u,∇u)∈Lploc(Q)×Lqloc(Q) |
and
∫Q(|u|p+|∇u|q)φdxdt+∫Qw(x)φdxdt=∫Quφttdxdt−∫QuΔφdxdt−∫Quφtdxdt, | (1.6) |
for all φ∈C2c(Q).
The first main result is the following Fujita-type theorem.
Theorem 1.1. Let N≥1, p,q>1, w∈L1loc(RN), w≥0 and w≢0.
(i) If N∈{1,2}, for all p,q>1, problem (1.1) admits no global weak solution.
(ii) Let N≥3. If
1<p<1+2N−2or1<q<1+1N−1, |
then problem (1.1) admits no global weak solution.
(iii) Let N≥3. If
p>1+2N−2andq>1+1N−1, |
then problem (1.1) admits global solutions for some w>0.
We mention below some remarks related to Theorem 1.1. For N≥3, let
pc(N,q)={1+2N−2ifq>1+1N−1,∞ifq<1+1N−1. |
From Theorem 1.1, if 1<p<pc(N,q), then problem (1.1) admits no global weak solution, while if p>pc(N,q), global weak solutions exist for some w>0. This shows that pc(N,q) is the Fujita critical exponent for the problem (1.1). Notice that in the case N≥3 and q>1+1N−1, pc(N,q) is also the critical exponent for utt−Δu=|u|p+w(x) (see [5]). This shows that in the range q>1+1N−1, the gradient term |∇u|q has no influence on the critical behavior of utt−Δu=|u|p+w(x).
It is interesting to note that the gradient term |∇u|q induces an interesting phenomenon of discontinuity of the Fujita critical exponent pc(N,q) jumping from 1+2N−2 to ∞ as q reaches the value 1+1N−1 from above.
Next, for N≥3 and σ<N, we introduce the sets
I+σ={w∈C(RN):w≥0,|x|−σ=O(w(x)) as |x|→∞} |
and
I−σ={w∈C(RN):w>0,w(x)=O(|x|−σ) as |x|→∞}. |
The next result provides the second critical exponent for the problem (1.1) in the sense of Lee and Ni [10].
Theorem 1.2. Let N≥3, p>1+2N−2 and q>1+1N−1.
(i) If
σ<max{2pp−1,qq−1} |
and w∈I+σ, then problem (1.1) admits no global weak solution.
(ii) If
σ≥max{2pp−1,qq−1}, |
then problem (1.1) admits global solutions for some w∈I−σ.
From Theorem 1.2, if N≥3, p>1+2N−2 and q>1+1N−1, then problem (1.1) admits a second critical exponent, namely
σ∗=max{2pp−1,qq−1}. |
The rest of the paper is organized as follows. In Section 2, some preliminary estimates are provided. Section 3 is devoted to the study of the Fujita critical exponent for the problem (1.1). Namely, Theorem 1.1 is proved. In Section 4, we study the second critical exponent for the problem (1.1) in the sense of Lee and Ni. Namely, we prove Theorem 1.2.
Consider two cut-off functions f,g∈C∞([0,∞)) satisfying
f≥0,f≢0,supp(f)⊂(0,1) |
and
0≤g≤1,g(s)={1if0≤s≤1,0ifs≥2. |
For T>0, we introduce the function
ξT(t,x)=f(tT)ℓg(|x|2T2ρ)ℓ=F(t)G(x),(t,x)∈Q, | (2.1) |
where ℓ≥2 and ρ>0 are constants to be chosen. It can be easily seen that
ξT≥0andξT∈C2c(Q). |
Throughout this paper, the letter C denotes various positive constants depending only on known quantities.
Lemma 2.1. Let κ>1 and ℓ>2κκ−1. Then
∫Q|ξT|−1κ−1|ΔξT|κκ−1dxdt≤CT−2ρκκ−1+Nρ+1. | (2.2) |
Proof. By (2.1), one has
∫Q|ξT|−1κ−1|ΔξT|κκ−1dxdt=(∫∞0F(t)dt)(∫RN|G|−1κ−1|ΔG|κκ−1dx). | (2.3) |
Using the properties of the cut-off function f, one obtains
∫∞0F(t)dt=∫∞0f(tT)ℓdt=∫T0f(tT)ℓdt=T∫10f(s)ℓds, |
which shows that
∫∞0F(t)dt=CT. | (2.4) |
Next, using the properties of the cut-off function g, one obtains
∫RN|G|−1κ−1|ΔG|κκ−1dx=∫Tρ<|x|<√2Tρ|G|−1κ−1|ΔG|κκ−1dx | (2.5) |
On the other hand, for Tρ<|x|<√2Tρ, an elementary calculation yields
ΔG(x)=Δ[g(r2T2ρ)ℓ],r=|x|=(d2dr2+N−1rddr)g(r2T2ρ)ℓ=2ℓT−2ρg(r2T2ρ)ℓ−2θ(x), |
where
θ(x)=Ng(r2T2ρ)g′(r2T2ρ)+2(ℓ−1)T−2ρr2g(r2T2ρ)g′(r2T2ρ)2+2T−2ρr2g(r2T2ρ)g″(r2T2ρ). |
Notice that since Tρ<r<√2Tρ, one deduces that
|θ(x)|≤C. |
Hence, it holds that
|ΔG(x)|≤CT−2ρg(|x|2T2ρ)ℓ−2,Tρ<|x|<√2Tρ |
and
|G(x)|−1κ−1|ΔG(x)|κκ−1≤CT−2ρκκ−1g(|x|2T2ρ)ℓ−2κκ−1,Tρ<|x|<√2Tρ. |
Therefore, by (2.5) and using the change of variable x=Tρy, one obtains
∫RN|G|−1κ−1|ΔG|κκ−1dx≤CT−2ρκκ−1∫Tρ<|x|<√2Tρg(|x|2T2ρ)ℓ−2κκ−1dx=CT−2ρκκ−1+Nρ∫1<|y|<√2g(|y|2)ℓ−2κκ−1dy, |
which yields (notice that ℓ>2κκ−1)
∫RN|G|−1κ−1|ΔG|κκ−1dx≤CT−2ρκκ−1+Nρ. | (2.6) |
Finally, using (2.3), (2.4) and (2.6), (2.2) follows.
Lemma 2.2. Let κ>1 and ℓ>κκ−1. Then
∫Q|ξT|−1κ−1|∇ξT|κκ−1dxdt≤CT−ρκκ−1+Nρ+1. | (2.7) |
Proof. By (2.1) and the properties of the cut-off function g, one has
∫Q|ξT|−1κ−1|∇ξT|κκ−1dxdt=(∫∞0F(t)dt)(∫Tρ<|x|<√2Tρ|G|−1κ−1|∇G|κκ−1dx). | (2.8) |
On the other hand, for Tρ<|x|<√2Tρ, an elementary calculation shows that
|∇G(x)|=2ℓT−2ρ|x||g′(|x|2T2ρ)|g(|x|2T2ρ)ℓ−1≤2√2ℓT−ρ|g′(|x|2T2ρ)|g(|x|2T2ρ)ℓ−1≤CT−ρg(|x|2T2ρ)ℓ−1, |
which yields
|G|−1κ−1|∇G|κκ−1≤CT−ρκκ−1g(|x|2T2ρ)ℓ−κκ−1,Tρ<|x|<√2Tρ. |
Therefore, using the change of variable x=Tρy and the fact that ℓ>κκ−1, one obtains
∫Tρ<|x|<√2Tρ|G|−1κ−1|∇G|κκ−1dx≤CT−ρκκ−1∫Tρ<|x|<√2Tρg(|x|2T2ρ)ℓ−κκ−1dx=CT−ρκκ−1+Nρ∫1<|y|<√2g(|y|2)ℓ−κκ−1dy, |
i.e.
∫Tρ<|x|<√2Tρ|G|−1κ−1|∇G|κκ−1dx≤CT−ρκκ−1+Nρ. | (2.9) |
Using (2.4), (2.8) and (2.9), (2.7) follows.
Lemma 2.3. Let κ>1 and ℓ>2κκ−1. Then
∫Q|ξT|−1κ−1|ξTtt|κκ−1dxdt≤CT−2κκ−1+1+Nρ. | (2.10) |
Proof. By (2.1) and the properties of the cut-off functions f and g, one has
∫Q|ξT|−1κ−1|ξTtt|κκ−1dxdt=(∫T0F(t)−1κ−1|F″(t)|κκ−1dt)(∫0<|x|<√2TρG(x)dx). | (2.11) |
An elementary calculation shows that
F(t)−1κ−1|F″(t)|κκ−1≤CT−2κκ−1f(tT)ℓ−2κκ−1,0<t<T, |
which yields (since ℓ>2κκ−1)
∫T0F(t)−1κ−1|F″(t)|κκ−1dt≤CT−2κκ−1+1. | (2.12) |
On the other hand, using the change of variable x=Tρy, one obtains
∫0<|x|<√2TρG(x)dx=∫0<|x|<√2Tρg(|x|2T2ρ)ℓdx=TNρ∫0<|y|<√2g(|y|2)ℓdy, |
which yields
∫0<|x|<√2TρG(x)dx≤CTNρ. | (2.13) |
Therefore, using (2.11), (2.12) and (2.13), (2.10) follows.
Lemma 2.4. Let κ>1 and ℓ>κκ−1. Then
∫Q|ξT|−1κ−1|ξTt|κκ−1dxdt≤CT−κκ−1+1+Nρ. | (2.14) |
Proof. By (2.1) and the properties of the cut-off functions f and g, one has
∫Q|ξT|−1κ−1|ξTt|κκ−1dxdt=(∫T0F(t)−1κ−1|F′(t)|κκ−1dt)(∫0<|x|<√2TρG(x)dx). | (2.15) |
An elementary calculation shows that
F(t)−1κ−1|F′(t)|κκ−1≤CT−κκ−1f(tT)ℓ−κκ−1,0<t<T, |
which yields (since ℓ>κκ−1)
∫T0F(t)−1κ−1|F′(t)|κκ−1dt≤CT1−κκ−1. | (2.16) |
Therefore, using (2.13), (2.15) and (2.16), (2.14) follows.
Proposition 3.1. Suppose that problem (1.1) admits a global weak solution u with (u,∇u)∈Lploc(Q)×Lqloc(Q). Then there exists a constant C>0 such that
∫Qw(x)φdxdt≤Cmin{A(φ),B(φ)}, | (3.1) |
for all φ∈C2c(Q), φ≥0, where
A(φ)=∫Qφ−1p−1|φtt|pp−1dxdt+∫Qφ−1p−1|φt|pp−1dxdt+∫Qφ−1p−1|Δφ|pp−1dxdt |
and
B(φ)=∫Qφ−1p−1|φtt|pp−1dxdt+∫Qφ−1p−1|φt|pp−1dxdt+∫Qφ−1q−1|∇φ|qq−1dxdt. |
Proof. Let u be a global weak solution to problem (1.1) with
(u,∇u)∈Lploc(Q)×Lqloc(Q). |
Let φ∈C2c(Q), φ≥0. Using (1.6), one obtains
∫Q|u|pφdxdt+∫Qw(x)φdxdt≤∫Q|u||φtt|dxdt+∫Q|u||Δφ|dxdt+∫Q|u||φt|dxdt. | (3.2) |
On the other hand, by ε-Young inequality, 0<ε<13, one has
∫Q|u||φtt|dxdt=∫Q(|u|φ1p)(φ−1p|φtt|)dxdt≤ε∫Q|u|pφdxdt+C∫Qφ−1p−1|φtt|pp−1dxdt. | (3.3) |
Here and below, C denotes a positive generic constant, whose value may change from line to line. Similarly, one has
∫Q|u||Δφ|dxdt≤ε∫Q|u|pφdxdt+C∫Qφ−1p−1|Δφ|pp−1dxdt | (3.4) |
and
∫Q|u||φt|dxdt≤ε∫Q|u|pφdxdt+C∫Qφ−1p−1|φt|pp−1dxdt | (3.5) |
Therefore, using (3.2), (3.3), (3.4) and (3.5), it holds that
(1−3ε)∫Q|u|pφdxdt+∫Qw(x)φdxdt≤CA(φ). |
Since 1−3ε>0, one deduces that
∫Qw(x)φdxdt≤CA(φ). | (3.6) |
Again, by (1.6), one has
∫Q(|u|p+|∇u|q)φdxdt+∫Qw(x)φdxdt=∫Quφttdxdt−∫QuΔφdxdt−∫Quφtdxdt=∫Quφttdxdt+∫Q∇u⋅∇φdxdt−∫Quφtdxdt, |
where ⋅ denotes the inner product in RN. Hence, it holds that
∫Q(|u|p+|∇u|q)φdxdt+∫Qw(x)φdxdt≤∫Q|u||φtt|dxdt+∫Q|∇u||∇φ|dxdt+∫Q|u||φt|dxdt. | (3.7) |
By ε-Young inequality (ε=12), one obtains
∫Q|u||φtt|dxdt≤12∫Q|u|pφdxdt+C∫Qφ−1p−1|φtt|pp−1dxdt, | (3.8) |
∫Q|u||φt|dxdt≤12∫Q|u|pφdxdt+C∫Qφ−1p−1|φt|pp−1dxdt, | (3.9) |
∫Q|∇u||∇φ|dxdt≤12∫Q|∇u|qφdxdt+C∫Qφ−1q−1|∇φ|qq−1dxdt. | (3.10) |
Next, using (3.7), (3.8), (3.9) and (3.10), one deduces that
12∫Q|∇u|qφdxdt+∫Qw(x)φdxdt≤CB(φ), |
which yields
∫Qw(x)φdxdt≤CB(φ). | (3.11) |
Finally, combining (3.6) with (3.11), (3.1) follows.
Proposition 3.2. Suppose that problem (1.1) admits a global weak solution u with (u,∇u)∈Lploc(Q)×Lqloc(Q). Then
∫Qw(x)ξTdxdt≤Cmin{A(ξT),B(ξT)}, | (3.12) |
for all T>0, where ξT is defined by (2.1).
Proof. Since for all T>0, ξT∈C2c(Q), ξT≥0, taking φ=ξT in (3.1), (3.12) follows.
Proposition 3.3. Let w∈L1loc(RN), w≥0 and w≢0. Then, for sufficiently large T,
∫Qw(x)ξTdxdt≥CT, | (3.13) |
where ξT is defined by (2.1).
Proof. By (2.1) and the properties of the cut-off functions f and g, one has
∫Qw(x)ξTdxdt=(∫T0f(tT)ℓdt)(∫0<|x|<√2Tρw(x)g(|x|2T2ρ)ℓdx). | (3.14) |
On the other hand,
∫T0f(tT)ℓdt=T∫10f(s)ℓds. | (3.15) |
Moreover, for sufficiently large T (since w,g≥0),
∫0<|x|<√2Tρw(x)g(|x|2T2ρ)ℓdx≥∫0<|x|<1w(x)g(|x|2T2ρ)ℓdx. | (3.16) |
Notice that since w∈L1loc(RN), w≥0, w≢0, and using the properties of the cut-off function g, by the domianted convergence theorem, one has
limT→∞∫0<|x|<1w(x)g(|x|2T2ρ)ℓdx=∫0<|x|<1w(x)dx>0. |
Hence, for sufficiently large T,
∫0<|x|<1w(x)g(|x|2T2ρ)ℓdx≥C. | (3.17) |
Using (3.14), (3.15), (3.16) and (3.17), (3.13) follows.
Now, we are ready to prove the Fujita-type result given by Theorem 1.1. The proof is by contradiction and makes use of the nonlinear capacity method, which was developed by Mitidieri and Pohozaev (see e.g. [11]).
Proof of Theorem 1.1. (ⅰ) Suppose that problem (1.1) admits a global weak solution u with (u,∇u)∈Lploc(Q)×Lqloc(Q). By Propositions 3.2 and 3.3, for sufficiently large T, one has
CT≤min{A(ξT),B(ξT)}, | (3.18) |
where ξT is defined by (2.1),
A(ξT)=∫Qξ−1p−1T|ξTtt|pp−1dxdt+∫Qξ−1p−1T|ξTt|pp−1dxdt+∫Qξ−1p−1T|ΔξT|pp−1dxdt | (3.19) |
and
B(ξT)=∫Qφ−1p−1|ξTtt|pp−1dxdt+∫Qξ−1p−1T|ξTt|pp−1dxdt+∫Qξ−1q−1T|∇ξT|qq−1dxdt. | (3.20) |
Taking ℓ>max{2pp−1,qq−1}, using Lemmas 2.1, 2.3, 2.4 with κ=p, and Lemma 2.2 with κ=q, one obtains the following estimates
∫Qξ−1p−1T|ξTtt|pp−1dxdt≤CT−2pp−1+1+Nρ, | (3.21) |
∫Qξ−1p−1T|ξTt|pp−1dxdt≤CT−pp−1+1+Nρ, | (3.22) |
∫Qξ−1p−1T|ΔξT|pp−1dxdt≤CT−2ρpp−1+Nρ+1, | (3.23) |
∫Qξ−1q−1T|∇ξT|qq−1dxdt≤CT−ρqq−1+Nρ+1. | (3.24) |
Hence, by (3.19), (3.21), (3.22) and (3.23), one deduces that
A(ξT)≤C(T−2pp−1+1+Nρ+T−pp−1+1+Nρ+T−2ρpp−1+Nρ+1). |
Observe that
−2pp−1+1+Nρ<−pp−1+1+Nρ. | (3.25) |
So, for sufficiently large T, one deduces that
A(ξT)≤C(T−pp−1+1+Nρ+T−2ρpp−1+Nρ+1). |
Notice that the above estimate holds for all ρ>0. In particular, when ρ=12, it holds that
A(ξT)≤CT−pp−1+1+N2 | (3.26) |
Next, by (3.20), (3.21), (3.22), (3.24) and (3.25), one deduces that
B(ξT)≤C(T−pp−1+1+Nρ+T−ρqq−1+Nρ+1). |
Similarly, the above inequality holds for all ρ>0. In particular, when ρ=p(q−1)q(p−1), it holds that
B(ξT)≤CT−pp−1+1+Np(q−1)q(p−1). | (3.27) |
Therefore, it follows from (3.18), (3.26) and (3.27) that
0<C≤min{T−pp−1+N2,T−pp−1+Np(q−1)q(p−1)}, |
which yields
0<C≤T−pp−1+N2:=Tα(N) | (3.28) |
and
0<C≤T−pp−1+Np(q−1)q(p−1):=Tβ(N). | (3.29) |
Notice that for N∈{1,2}, one has
α(N)={−p−12(p−1)ifN=1,−1p−1ifN=2, |
which shows that
α(N)<0,N∈{1,2}. |
Hence, for N∈{1,2}, passing to the limit as T→∞ in (3.28), a contradiction follows (0<C≤0). This proves part (ⅰ) of Theorem 1.1.
(ⅱ) Let N≥3. In this case, one has
α(N)<0⟺p<1+2N−2. |
Hence, if p<1+2N−2, passing to the limit as T→∞ in (3.28), a contradiction follows. Furthermore, one has
β(N)<0⟺q<1+1N−1. |
Hence, if q<1+1N−1, passing to the limit as T→∞ in (3.29), a contradiction follows. Therefore, we proved part (ⅱ) of Theorem 1.1.
(ⅲ) Let
p>1+2N−2andq>1+1N−1. | (3.30) |
Let
u(x)=ε(1+r2)−δ,r=|x|,x∈RN, | (3.31) |
where
max{1p−1,2−q2(q−1)}<δ<N−22 | (3.32) |
and
0<ε<min{1,(2δ(N−2δ−2)1+2qδq)1min{p,q}−1}. | (3.33) |
Note that due to (3.30), the set of \delta satisfying (3.32) is nonempty. Let
w(x) = -\Delta u-|u|^p-|\nabla u|^q,\quad x\in \mathbb{R}^N. |
Elementary calculations yield
\begin{eqnarray} w(x)& = &2\delta \varepsilon \left[(1+r^2)^{-\delta-1}-2r^2(\delta+1)(1+r^2)^{-\delta-2}+(N-1)(1+r^2)^{-\delta-1}\right]\\ && -\varepsilon^p(1+r^2)^{-\delta p}-2^q\delta^q\varepsilon^qr^q(1+r^2)^{-(\delta+1)q}. \end{eqnarray} | (3.34) |
Hence, one obtains
\begin{eqnarray*} w(x) &\geq & 2\delta \varepsilon \left[(1+r^2)^{-\delta-1}-2(\delta+1)(1+r^2)^{-\delta-1}+(N-1)(1+r^2)^{-\delta-1}\right]\\ &&-\varepsilon^p(1+r^2)^{-\delta p}-2^q\delta^q\varepsilon^q(1+r^2)^{-(\delta+1)q+\frac{q}{2}}\\ & = & 2\delta \varepsilon (N-2\delta-2)(1+r^2)^{-\delta-1}-\varepsilon^p(1+r^2)^{-\delta p}-2^q\varepsilon^q\delta^q(1+r^2)^{-\left(\delta+\frac{1}{2}\right)q}. \end{eqnarray*} |
Next, using (3.32) and (3.33), one deduces that
w(x)\geq \varepsilon \left[2\delta (N-2\delta-2)-\varepsilon^{p-1}-2^q\varepsilon^{q-1}\delta^q\right](1+r^2)^{-\delta-1} \gt 0. |
Therefore, for any \delta and \varepsilon satisfying respectively (3.32) and (3.33), the function u defined by (3.31) is a stationary solution (then global solution) to (1.1) for some w > 0 . This proves part (ⅲ) of Theorem 1.1.
Proposition 4.1. Let N\geq 3 , \sigma < N and w\in I_\sigma^+ . Then, for sufficiently large T ,
\begin{equation} \int_Q w(x)\xi_T\,dx\,dt \geq C T^{\rho(N-\sigma)+1}. \end{equation} | (4.1) |
where \xi_T is defined by (2.1).
Proof. By (3.14) and (3.15), one has
\begin{equation} \int_Q w(x)\xi_T\,dx\,dt = CT\int_{0 \lt |x| \lt \sqrt{2}T^\rho} w(x) g\left(\frac{|x|^2}{T^{2\rho}}\right)^\ell\,dx. \end{equation} | (4.2) |
On the other hand, by definition of I_\sigma^+ , and using that g(s) = 1 , 0\leq s\leq 1 , for sufficiently large T , one obtains (since w, g\geq 0 )
\begin{eqnarray*} \int_{0 \lt |x| \lt \sqrt{2}T^\rho} w(x) g\left(\frac{|x|^2}{T^{2\rho}}\right)^\ell\,dx &\geq & \int_{0 \lt |x| \lt T^\rho} w(x) g\left(\frac{|x|^2}{T^{2\rho}}\right)^\ell\,dx\\ & = & \int_{0 \lt |x| \lt T^\rho} w(x)\,dx\\ &\geq & \int_{\frac{T^\rho}{2} \lt |x| \lt T^\rho} w(x)\,dx\\ &\geq & C \int_{\frac{T^\rho}{2} \lt |x| \lt T^\rho} |x|^{-\sigma}\,dx \\ & = & C T^{\rho(N-\sigma)}. \end{eqnarray*} |
Hence, using (4.2), (4.1) follows.
Now, we are ready to prove the new critical behavior for the problem (1.1) stated by Theorem 1.2.
Proof of Theorem 1.2. (ⅰ) Suppose that problem (1.1) admits a global weak solution u with (u, \nabla u)\in L^p_{loc}(Q)\times L^q_{loc}(Q) . By Propositions 3.2 and 4.1, for sufficiently large T , one obtains
\begin{equation} CT^{\rho(N-\sigma)+1}\leq \min\{A(\xi_T),B(\xi_T)\}, \end{equation} | (4.3) |
where \xi_T , A(\xi_T) and B(\xi_T) are defined respectively by (2.1), (3.19) and (3.20). Next, using (3.26) and (4.3) with \rho = \frac{1}{2} , one deduces that
\begin{equation} 0 \lt C\leq T^{\frac{-p}{p-1}+\frac{\sigma}{2}}. \end{equation} | (4.4) |
Observe that
\frac{-p}{p-1}+\frac{\sigma}{2} \lt 0 \Longleftrightarrow \sigma \lt \frac{2p}{p-1}. |
Hence, if \sigma < \frac{2p}{p-1} , passing to the limit as T\to \infty in (4.4), a contradiction follows. Furthermore, using (3.27) and (4.3) with \rho = \frac{p(q-1)}{q(p-1)} , one deduces that
\begin{equation} 0 \lt C\leq T^{\frac{p}{p-1}\left(\frac{\sigma(q-1)}{q}-1\right)}. \end{equation} | (4.5) |
Observe that
\frac{p}{p-1}\left(\frac{\sigma(q-1)}{q}-1\right) \lt 0\Longleftrightarrow \sigma \lt \frac{q}{q-1}. |
Hence, if \sigma < \frac{q}{q-1} , passing to the limit as T\to \infty in (4.5), a contradiction follows. Therefore, part (ⅰ) of Theorem 1.2 is proved.
(ⅱ) Let
\begin{equation} \sigma \geq \max\left\{\frac{2p}{p-1},\frac{q}{q-1}\right\}. \end{equation} | (4.6) |
Let u be the function defined by (3.31), where
\begin{equation} \frac{\sigma-2}{2} \lt \delta \lt \frac{N-2}{2} \end{equation} | (4.7) |
and \varepsilon satisfies (3.33). Notice that since \sigma < N , the set of \delta satisfying (4.7) is nonempty. Moreover, due to (4.6) and (4.7), \delta satisfies also (3.32). Hence, from the proof of part (iii) of Theorem 1.2, one deduces that
w(x) = -\Delta u-|u|^p-|\nabla u|^q \gt 0,\quad x\in \mathbb{R}^N. |
On the other hand, using (3.34) and (4.7), for |x| large enough, one obtains
w(x)\leq C (1+|x|^2)^{-\delta-1} \leq C |x|^{-2\delta-2}\leq C|x|^{-\delta}, |
which proves that w\in I_\sigma^- . This proves part (ⅱ) of Theorem 1.2.
We investigated the large-time behavior of solutions to the nonlinear damped wave equation (1.1). In the case when N\in\{1, 2\} , we proved that for all p > 1 , problem (1.1) admits no global weak solution (in the sense of Definition 1). Notice that from [5], the same result holds for the problem without gradient term, namely problem (1.5). This shows that in the case N\in\{1, 2\} , the nonlinearity |\nabla u|^q has no influence on the critical behavior of problem (1.5). In the case when N\geq 3 , we proved that, if 1 < p < 1+\frac{2}{N-2} or 1 < q < 1+\frac{1}{N-1} , then problem (1.1) admits no global weak solution, while if p > 1+\frac{2}{N-2} and q > 1+\frac{1}{N-1} , global solutions exist for some w > 0 . This shows that in this case, the Fujita critical exponent for the problem (1.1) is given by
p_c(N,q) = \left\{\begin{array}{lll} 1+\frac{2}{N-2} &\mbox{if}& q \gt 1+\frac{1}{N-1},\\ \infty &\mbox{if}& q \lt 1+\frac{1}{N-1}. \end{array} \right. |
From this result, one observes two facts. First, in the range q > 1+\frac{1}{N-1} , from [5], the critical exponent p_c(N, q) is also equal to the critical exponent for the problem without gradient term, which means that in this range of q , the nonlinearity |\nabla u|^q has no influence on the critical behavior of problem (1.5). Secondly, one observes that the gradient term induces an interesting phenomenon of discontinuity of the Fujita critical exponent p_c(N, q) jumping from 1+\frac{2}{N-2} to \infty as q reaches the value 1+\frac {1}{N-1} from above. In the same case N\geq 3 , we determined also the second critical exponent for the problem (1.1) in the sense of Lee and Ni [10], when p > 1+\frac{2}{N-2} and q > 1+\frac{1}{N-1} . Namely, we proved that in this case, if \sigma < \max\left\{\frac{2p}{p-1}, \frac{q}{q-1}\right\} and w\in I_\sigma^+ , then there is no global weak solution, while if \max\left\{\frac{2p}{p-1}, \frac{q}{q-1}\right\}\leq \sigma < N , global solutions exist for some w\in I_\sigma^- . This shows that the second critical exponent for the problem (1.1) in the sense of Lee and Ni is given by
\sigma^* = \max\left\{\frac{2p}{p-1},\frac{q}{q-1}\right\}. |
We end this section with the following open questions:
(Q1). Find the first and second critical exponents for the system of damped wave equations with mixed nonlinearities
\begin{eqnarray*} \left\{\begin{array}{lllll} u_{tt}-\Delta u +u_t & = & |v|^{p_1}+|\nabla v|^{q_1} +w_1(x) &\mbox{in}& (0,\infty)\times \mathbb{R}^N,\\ v_{tt}-\Delta v +v_t& = & |u|^{p_2}+|\nabla u|^{q_2} +w_2(x) &\mbox{in}& (0,\infty)\times \mathbb{R}^N, \end{array} \right. \end{eqnarray*} |
where p_i, q_i > 1 , w_i\in L^1_{loc}(\mathbb{R}^N) , w_i\geq 0 and w_i\not\equiv 0 , i = 1, 2 .
(Q2). Find the Fujita critical exponent for the problem (1.1) with w\equiv 0 .
(Q3) Find the Fujita critical exponent for the problem
u_{tt}-\Delta u +u_t = \frac{1}{\Gamma(1-\alpha)}\int_0^t (t-s)^{-\alpha}|u(s,x)|^{p}\,ds +|\nabla u|^{q} +w(x),\quad t \gt 0,\,x\in \mathbb{R}^N, |
where N\geq 1 , p, q > 1 , 0 < \alpha < 1 and w\in L^1_{loc}(\mathbb{R}^N) , w\geq 0 . Notice that in the limit case \alpha \to 1^- , the above equation reduces to (1.1).
(Q4) Same question as above for the problems
u_{tt}-\Delta u +u_t = |u|^p+\frac{1}{\Gamma(1-\alpha)}\int_0^t (t-s)^{-\alpha}|\nabla u(s,x)|^{q}\,ds+w(x) |
and
u_{tt}-\Delta u +u_t = \frac{1}{\Gamma(1-\alpha)}\int_0^t (t-s)^{-\alpha}|u(s,x)|^{p}\,ds +\frac{1}{\Gamma(1-\beta)}\int_0^t(t-s)^{-\beta}|\nabla u(s,x)|^{q} +w(x), |
where 0 < \alpha, \beta < 1 .
The author is supported by Researchers Supporting Project RSP-2020/4, King Saud University, Saudi Arabia, Riyadh.
The author declares no conflict of interest.
[1] | Iraq diabetes report 2000-2045. Available from: https://www.diabetesatlas.org/data/en/country/96/iq.html |
[2] |
Tigga NP, Garg S (2023) Speech Emotion Recognition for multiclass classification using Hybrid CNN-LSTM. International Journal of Microsystems and Iot 1: 9–17. https://doi.org/10.5281/zenodo.8158288 doi: 10.5281/zenodo.8158288
![]() |
[3] |
Jaber HA, Rashid MT (2019) HD-sEMG gestures recognition by SVM classifier for controlling prosthesis. Iraqi Journal of Computers, Communications, Control and System Engineering (IJCCCE) 19: 10–19. https://doi.org/10.33103/uot.ijccce.19.1.2 doi: 10.33103/uot.ijccce.19.1.2
![]() |
[4] |
Abgeena A, Garg S (2023) A novel convolution bi-directional gated recurrent unit neural network for emotion recognition in multichannel electroencephalogram signals. Technol Health Care 31: 1215–1234. https://doi.org/10.3233/THC-220458 doi: 10.3233/THC-220458
![]() |
[5] |
Mujumdar A, Vaidehi V (2019) Diabetes prediction using machine learning algorithms. Procedia Computer Science 165: 292–299. https://doi.org/10.1016/j.procs.2020.01.047 doi: 10.1016/j.procs.2020.01.047
![]() |
[6] |
Madan P, Singh V, Chaudhari V, Albagory Y, Dumka A, Singh R, et al. (2022) An optimization-based diabetes prediction model using cnn and bi-directional lstm in real-time environment. Applied Sciences 12: 3989. https://doi.org/10.3390/app12083989 doi: 10.3390/app12083989
![]() |
[7] |
Chang V, Bailey J, Xu QA, Sun Z (2023) Pima indians diabetes mellitus classification based on machine learning (ml) algorithms. Neural Computing and Applications 36: 16157–16173. https://doi.org/10.1007/s00521-022-07049-z doi: 10.1007/s00521-022-07049-z
![]() |
[8] | Noori NA, Yassin AA (2021) A comparative analysis for diabetic prediction based on machine learning techniques. Journal of Basrah Researches (Sciences) 47. |
[9] |
Ahamed BS, Arya MS, Nancy AO (2022) Prediction of type-2 diabetes mellitus disease using machine learning classifiers and techniques. Frontiers in Computer Science 4: 835242. https://doi.org/10.1155/2022/9220560 doi: 10.1155/2022/9220560
![]() |
[10] | Butt UM, Letchmunan S, Ali M, Hassan FH, Baqir A, Sherazi H (2021) Machine learning based diabetes classification and prediction for healthcare applications. Journal of healthcare engineering 2021. https://doi.org/10.1155/2021/9930985 |
[11] |
Zou Q, Qu K, Luo Y, Yin D, Ju Y, Tang H (2018) Predicting diabetes mellitus with machine learning techniques. Frontiers in genetics 9: 515. https://doi.org/10.3389/fgene.2018.00515 doi: 10.3389/fgene.2018.00515
![]() |
[12] |
Naz H, Ahuja S (2020) Deep learning approach for diabetes prediction using pima indian dataset. Journal of Diabetes and Metabolic Disorders 19: 391–403. https://doi.org/10.1007/s40200-020-00520-5 doi: 10.1007/s40200-020-00520-5
![]() |
[13] |
Olisah CC, Smith L, Smith M (2022) Diabetes mellitus prediction and diagnosis from a data preprocessing and machine learning perspective. Comput Meth Prog Bio 220: 106773. https://doi.org/10.1016/j.cmpb.2022.106773 doi: 10.1016/j.cmpb.2022.106773
![]() |
[14] | Rashid A (2020) Diabetes dataset, Mendeley Data. |
[15] |
Mirjalili S (2015) Dragonfly algorithm: A new meta-heuristic optimization technique for solving single-objective, discrete, and multi-objective problems. Neural Comput Appl 27: 1053–1073. https://doi.org/10.1007/s00521-015-1920-1 doi: 10.1007/s00521-015-1920-1
![]() |
[16] | Mafarja MM, Eleyan D, Jaber I, Hammouri A, Mirjalili S (2017) Binary Dragonfly Algorithm for Feature Selection. 2017 International Conference on New Trends in Computing Sciences (ICTCS), 12–17. |
[17] | Alhakeem ZM, Ali RS (2019) Fast channel selection method using crow search algorithm. Proceedings of the International Conference on Information and Communication Technology, 210–214. https://doi.org/10.1145/3321289.3321309 |
[18] |
Hochreiter S, Schmidhuber J (1997) Long Short-Term Memory. Neural Computation 9: 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735 doi: 10.1162/neco.1997.9.8.1735
![]() |
[19] | Dalianis H (2018) Evaluation Metrics and Evaluation, Cham: Springer International Publishing. https://doi.org/10.1007/978-3-319-78503-5_6 |
1. | Sen Ming, Jiayi Du, Bo Du, Existence of Blow-Up Solution to the Cauchy Problem of Inhomogeneous Damped Wave Equation, 2025, 17, 2073-8994, 1009, 10.3390/sym17071009 |