Research article Special Issues

Segmentation of Left Ventricle with a Coupled Length Regularization and Sparse Composite Shape Prior: a Variational Approach

  • Received: 25 May 2015 Accepted: 05 September 2015 Published: 14 September 2015
  • Segmentation of left ventricles in Cine MR images plays an important role in analyzing cardiac functions. In this study, we propose a variational method that incorporates both prior knowledge on geometrical coupling and shapes of the endo- and epicardium. Specifically, we dynamically maintain and update a smoothly varying distance between the endo- and epicardial contours, represented by a pair of level set functions, with a novel coupling energy embedded in the length regularization. We encode the shape prior with a sparse composite model based on a set of training templates. A robust fidelity with Gaussian mixture models is employed to provide robust intensity estimates in each subregion under insufficient local gradient information. Quantitative evaluation of the proposed method demonstrates competitive/better DSC and APD accuracy compared to other state-of-the-art approaches.

    Citation: Wenyang Liu, Dan Ruan. Segmentation of Left Ventricle with a Coupled Length Regularization and Sparse Composite Shape Prior: a Variational Approach[J]. AIMS Medical Science, 2015, 2(4): 295-302. doi: 10.3934/medsci.2015.4.295

    Related Papers:

    [1] Panyu Deng, Jun Zheng, Guchuan Zhu . Well-posedness and stability for a nonlinear Euler-Bernoulli beam equation. Communications in Analysis and Mechanics, 2024, 16(1): 193-216. doi: 10.3934/cam.2024009
    [2] Yonghui Zou . Global regularity of solutions to the 2D steady compressible Prandtl equations. Communications in Analysis and Mechanics, 2023, 15(4): 695-715. doi: 10.3934/cam.2023034
    [3] Isaac Neal, Steve Shkoller, Vlad Vicol . A characteristics approach to shock formation in 2D Euler with azimuthal symmetry and entropy. Communications in Analysis and Mechanics, 2025, 17(1): 188-236. doi: 10.3934/cam.2025009
    [4] Zhigang Wang . Serrin-type blowup Criterion for the degenerate compressible Navier-Stokes equations. Communications in Analysis and Mechanics, 2025, 17(1): 145-158. doi: 10.3934/cam.2025007
    [5] Reinhard Racke . Blow-up for hyperbolized compressible Navier-Stokes equations. Communications in Analysis and Mechanics, 2025, 17(2): 550-581. doi: 10.3934/cam.2025022
    [6] Hongxia Lin, Sabana, Qing Sun, Ruiqi You, Xiaochuan Guo . The stability and decay of 2D incompressible Boussinesq equation with partial vertical dissipation. Communications in Analysis and Mechanics, 2025, 17(1): 100-127. doi: 10.3934/cam.2025005
    [7] Yang Liu, Xiao Long, Li Zhang . Long-time dynamics for a coupled system modeling the oscillations of suspension bridges. Communications in Analysis and Mechanics, 2025, 17(1): 15-40. doi: 10.3934/cam.2025002
    [8] Shengbing Deng, Qiaoran Wu . Existence of normalized solutions for the Schrödinger equation. Communications in Analysis and Mechanics, 2023, 15(3): 575-585. doi: 10.3934/cam.2023028
    [9] Yuxuan Chen . Global dynamical behavior of solutions for finite degenerate fourth-order parabolic equations with mean curvature nonlinearity. Communications in Analysis and Mechanics, 2023, 15(4): 658-694. doi: 10.3934/cam.2023033
    [10] Huiyang Xu . Existence and blow-up of solutions for finitely degenerate semilinear parabolic equations with singular potentials. Communications in Analysis and Mechanics, 2023, 15(2): 132-161. doi: 10.3934/cam.2023008
  • Segmentation of left ventricles in Cine MR images plays an important role in analyzing cardiac functions. In this study, we propose a variational method that incorporates both prior knowledge on geometrical coupling and shapes of the endo- and epicardium. Specifically, we dynamically maintain and update a smoothly varying distance between the endo- and epicardial contours, represented by a pair of level set functions, with a novel coupling energy embedded in the length regularization. We encode the shape prior with a sparse composite model based on a set of training templates. A robust fidelity with Gaussian mixture models is employed to provide robust intensity estimates in each subregion under insufficient local gradient information. Quantitative evaluation of the proposed method demonstrates competitive/better DSC and APD accuracy compared to other state-of-the-art approaches.


    In this paper, we consider the non-isentropic compressible Euler equations with a source term in the following Euler coordinate system:

    $ {ρt+(ρu)x=0,(ρu)t+(ρu2+p(ρ,S))x=βρ|u|αu,St+uSx=0,
    $
    (1.1)

    where$ \; \rho, u, S $ and$ \; p(\rho, S) $ are the density, velocity, entropy and pressure of the considered gas, respectively. $ x\in[0, L] $ is the spatial variable, and $ L > 0 $ is a constant denoting the duct's length. $ p(\rho, S) = a e^{S} \rho ^{\gamma } $, with constants $ a > 0 $ and $ \gamma > 1 $. And, the term $ \beta \rho |u|^{\alpha} u $ represents the source term with $ \alpha, \; \beta \in R $. Especially, the source term denotes friction when $ \beta < 0 $.

    System (1.1) is equipped with initial data:

    $ (ρ,u,S)|t=0=(ρ0(x),u0(x),S0(x)),
    $
    (1.2)

    and boundary conditions:

    $ (ρ,u,S)|x=0=(ρl(t),ul(t),Sl(t)).
    $
    (1.3)

    If $ S = Const. $, the system (1.1) is the isentropic Euler equations with a source term. In the past few decades, the problems related to the isentropic compressible Euler equations with different kinds of source terms have been studied intensively. We refer the reader to [1,2,3,4,5,6,7,8,9,10] to find the existence and decay rates of small smooth (or large weak) solutions to Euler equations with damping. The global stability of steady supersonic solutions of 1-D compressible Euler equations with friction $ \beta\rho|u|u $ was studied in [11]. For the singularity formation of smooth solutions, we can see [12,13,14,15] and the references therein. Moreover, the authors in [16] established the finite-time blow-up results for compressible Euler system with space-dependent damping in 1-D. Recently, time-periodic solutions have attracted much attention. However, most of these temporal periodic solutions are driven by the time-periodic external force; see [17,18] for examples. The first result on the existence and stability of time-periodic supersonic solutions triggered by boundary conditions was considered in [19]. Then, the authors of [20] studied the global existence and stability of the time-periodic solution of the isentropic compressible Euler equations with source term $ \beta \rho |u|^{\alpha}u $.

    If $ S\neq Const. $, much less is known. In [21,22,23,24,25,26], the authors used characteristics analysis and energy estimate methods to study 1-D non-isentropic p-systems with damping in Lagrangian coordinates. Specifically, the global existence of smooth solutions for the Cauchy problem with small initial data has been investigated in [21,22]. The influence of the damping mechanism on the large time behavior of solutions was considered in [23,24]. For the results of the initial-boundary value problem, see [25,26]. The stability of combination of rarefaction waves with viscous contact wave for compressible Navier-Stokes equations with temperature-dependent transport coefficients and large data was obtained in [27]. As for the problems about non-isentropic compressible Euler equations with a vacuum boundary, we refer the reader to [28,29]. In [30,31,32], the relaxation limit problems for non-isentropic compressible Euler equations with source terms in multiple space dimensions were discussed.

    In this paper, we are interested in the dynamics of non-isentropic Euler equations with friction. Exactly speaking, we want to prove the global existence and stability of temporal periodic solutions around the supersonic steady state to non-isentropic compressible Euler equations with the general friction term $ \beta \rho |u|^{\alpha} u $ for any $ \alpha, \; \beta\in R $. It is worth pointing out that the temporal periodic non-isentropic supersonic solution considered in this paper is driven by periodic boundary conditions.

    We choose the steady solution $ \tilde{W}(x) = (\tilde{\rho}(x), \tilde{u}(x), \tilde{S}(x))^{\top} $ (with $ \tilde{u}(x) > 0 $) as a background solution, which satisfies

    $ {(˜ρ˜u)x=0,(˜ρ˜u2+p(˜ρ,˜S))x=β˜ρ˜uα+1,˜u˜Sx=0,(˜ρ,˜u,˜S)|x=0=(ρ,u,S0).
    $
    (1.4)

    The equation $ (1.4)_{3} $ indicates that the static entropy in the duct must be a constant. That is, $ \tilde{S}(x) = S_{0} $. Moreover, when $ (\alpha, \beta) $ lies in different regions of $ R^{2} $, the source term $ \beta \tilde{\rho} \tilde{u}^{\alpha+1} $ affects the movement of flow dramatically. We analyze the influence meticulously and gain the allowable maximal duct length for subsonic or supersonic inflow.

    Based on the steady solution, we are interested in two problems. The first one is, if $ \rho_{l}(t)-\rho_{-} $, $ u_{l}(t)-u_{-} $, $ S_{l}(t)-S_{0} $ and $ \rho_{0}(x)- \tilde{\rho}(x) $, $ u_{0}(x)-\tilde{u}(x) $, $ S_{0}(x)-S_{0} $ are small in some norm sense, can we obtain a classical solution of the problem described by (1.1)–(1.3) for $ [0, \infty) \times [0, L] $ while this classical solution remains close to the background solution? If the first question holds, our second one is whether the small classical solution is temporal-periodic as long as the inflow is time-periodic at the entrance of ducts?

    We use $ \bar{W}(t, x) = (\bar{\rho}(t, x), \bar{u}(t, x), \bar{S}(t, x))^\top = (\rho(t, x)-\tilde{\rho}(x), u(t, x)-\tilde{u}(x), S(t, x)-S_{0})^\top $ to denote the perturbation around the background solution, and, correspondingly,

    $ \bar{W}_{0}(x) = (\bar{\rho}_{0}(x),\bar{u}_{0}(x),\bar{S}_{0}(x)) = (\rho_{0}(x)-\tilde{\rho}(x),u_{0}(x)-\tilde{u}(x),S_{0}(x)-S_{0}), $
    $ \bar{W}_{l}(t) = (\bar{\rho}_{l}(t),\bar{u}_{l}(t),\bar{S}_{l}(t)) = (\rho_{l}(t)-\rho_{-},u_{l}(t)-u_{-},S_{l}(t)-S_{0}), $

    that is,

    $ t=0:{ρ0(x)=ˉρ0(x)+˜ρ(x),u0(x)=ˉu0(x)+˜u(x),0xL,S0(x)=ˉS0(x)+S0,
    $
    (1.5)

    and

    $ x=0:{ρl(t)=ˉρl(t)+ρ,ul(t)=ˉul(t)+u,t0.Sl(t)=ˉSl(t)+S0.
    $
    (1.6)

    The main conclusions of this article are as follows:

    Theorem 1.1. For any fixed non-sonic upstream state $ (\rho_{-}, u_{-}, S_{0}) $ with $ \rho_{-} \neq \rho^{*} = \left[\frac{(\rho_{-}u_{-})^2}{a \gamma e^{S_{0}} }\right]^ {\frac{1}{\gamma +1}} > 0 $ and $ u_{-} > 0 $, the following holds:

    1) There exists a maximal duct length $ L_{m} $, which only depends on $ \alpha, \beta, \gamma $ and $ (\rho_{-}, u_{-}, S_{0}) $, such that the steady solution $ \tilde{W}(x) = (\tilde{\rho}(x), \tilde{u}(x), S_{0})^{\top} $ of the problem (1.1) exists in $ [0, L] $ for any $ L < L_{m} $;

    2) The steady solution $ (\tilde{\rho}(x), \tilde{u}(x), S_{0})^{\top} $ keeps the upstream supersonic/subsonic state and $ \tilde{\rho} \tilde{u} = \rho_{-}u_{-} > 0 $;

    3) $ \|(\tilde{\rho}(x), \tilde{u}(x), S_{0})\|_{C^{2}([0, L])} < M_{0} $, where $ M_{0} $ is a constant only depending on $ \alpha $, $ \beta $, $ \gamma $, $ \rho_{-} $, $ u_{-} $, $ S_{0} $ and $ L $;

    4) If $ \beta > 0 $, $ \alpha \leq 1 $ and the upstream is supersonic, the maximal duct length $ L_{m} $ can be infinite and a vacuum cannot appear in any finite place of ducts;

    5) When $ \beta > 0 $, $ \alpha\geq-\gamma $ and the upstream is subsonic, the maximal duct length $ L_{m} $ can also be infinite, and the flow cannot stop in any place of ducts.

    Theorem 1.2. Assume that the length of duct $ L < L_{m} $ and the steady flow is supersonic at the entrance of a duct, i.e., $ \rho_{-} < \rho^{*} = \left[\frac{(\rho_{-} u_{-})^2}{a \gamma e^{S_{0}} }\right]^ {\frac{1}{\gamma +1}} $. Then, there are constants $ \varepsilon_{0} $ and $ K_{0} $ such that, if

    $ ˉW0(x)C1([0,L])=(ρ0(x)˜ρ(x),u0(x)˜u(x),S0(x)S0)C1([0,L])ε<ε0,
    $
    (1.7)
    $ ˉWl(t)C1([0,+))=ρl(t)ρ,ul(t)u,Sl(t)S0C1([0,+))ε<ε0,
    $
    (1.8)

    and the $ C^{0}, \; C^{1} $ compatibility conditions are satisfied at point $ (0, 0) $, there is a unique $ C^{1} $ solution $ W(t, x) = (\rho(t, x), u(t, x), S(t, x))^{\top} $ for the mixed initial-boundary value problems (1.1)–(1.3) in the domain $ G = \{(t, x)|t\geq0, \; x\in[0, L]\} $, satisfying

    $ ˉW(t,x)C1(G)=ρ(t,x)˜ρ(x),u(t,x)˜u(x),S(t,x)S0C1(G)K0ε.
    $
    (1.9)

    Remark 1.1. Since the flows at $ \{x = L\} $ are entirely determined by the initial data on $ x\in[0, L] $ and the boundary conditions on $ \{x = 0\} $ under the supersonic conditions, we only need to present the boundary conditions on $ \{x = 0\} $ in Theorem 1.2.

    If we further assume that the boundaries $ \rho_{l}(t), u_{l}(t), S_{l}(t) $ are periodic, then the $ C^{1} $ solution obtained in Theorem 1.2 is a temporal periodic solution:

    Theorem 1.3. Suppose that the assumptions of Theorem 1.2 are fulfilled and the flow at the entrance $ x = 0 $ is temporal-periodic, i.e., $ W_{l}(t+P) = W_{l}(t) $; then, the $ C^{1} $ solution $ W(t, x) = (\rho(t, x), u(t, x), S(t, x))^{\top} $ of the problem described by $ (1.1) $–$ (1.3) $ is also temporal-periodic, namely,

    $ W(t+P,x)=W(t,x)
    $
    (1.10)

    for any $ t > T_{1} $ and $ x \in [0, L] $, where $ T_{1} $ is a constant defined in (4.3).

    The organization of this article is as follows. In the next section, we study the steady-state supersonic and subsonic flow. The wave decomposition for non-isentropic Euler equations is introduced in Section 3. In Section 4, based on wave decomposition, we prove the global existence and stability of smooth solutions under small perturbations around the steady-state supersonic flow. And, in Section 5, with the help of Gronwall's inequality, we obtain that the smooth supersonic solution is a temporal periodic solution, after a certain start-up time, with the same period as the boundary conditions.

    In this section, the steady-state flow is considered for some positive constants upstream $ (\rho_{-}, u_{-}, S_{0}) $ on the left side. In [11], the authors considered the differential equation in which the Mach number varies with the length of the duct. In [20], the authors investigated the steady-state equation with sound speed and flow velocity. Different from the methods used in [11] and [20], and motivated by [33], we rewrite (1.4) as the equations related to momentum and density in this paper, namely,

    $ {˜mx=0,(˜m2˜ρ+p(˜ρ,S0))x=β˜mα+1˜ρα,(˜ρ,˜m)|x=0=(ρ,ρu),
    $
    (2.1)

    where $ \tilde{m} = \tilde{\rho} \tilde{u} $ represents momentum. The advantage of this method is that the vacuum and stagnant states can be considered. Now, we analyze this problem in three cases:

    Case 1: $ \alpha \neq 1 $ and $ \alpha \neq -\gamma $.

    In this case, (2.1) becomes

    $ {˜m=const.=ρu,F1(˜ρ,˜m)x=β˜mα+1,
    $
    (2.2)

    where

    $ F1(˜ρ,˜m)=˜m2α1˜ρα1+aγeS0γ+α˜ργ+α.
    $
    (2.3)

    Then, we get

    $ F1(˜ρ,˜m)˜ρ=˜ρα(˜m2˜ρ2+aγeS0˜ργ1)=˜ρα2(˜ρ2p˜ρ˜m2).
    $
    (2.4)

    Let $ G(\tilde{\rho}, \tilde{m}) = \tilde{\rho}^2 p_{\tilde{\rho}} -\tilde{m}^{2} $. For any fixed $ \tilde{m} > 0 $, we have that $ \underset{ \tilde{\rho} \rightarrow 0}{\lim} G(\tilde{\rho}, \tilde{m}) = -\tilde{m}^{2} < 0 $. From the definition of $ p(\tilde{\rho}, S_{0}) $, we obtain

    $ \tilde{\rho}^{2} p_{\tilde{\rho}} \;{\text{ is a strictly increasing function for}}\; \tilde{\rho} > 0. $

    Thus, when $ \tilde{\rho}\rightarrow +\infty $, $ G(\tilde{\rho}, \tilde{m})\rightarrow +\infty $. Then, there exists $ \rho^{*} = \left[\frac{(\rho_{-} u_{-})^2}{a \gamma e^{S_{0}} }\right]^ {\frac{1}{\gamma +1}} > 0 $ such that $ G\left(\rho^{*}, \tilde{m}\right) = 0 $ (i.e., $ \left(\rho^*\right)^2 p_{\tilde{\rho}} \left(\rho^*\right) = \tilde{m}^{2} $). That is, when $ \tilde{\rho} = \rho^{*} $, the fluid velocity is equal to the sound speed (i.e., $ \tilde{u} = \tilde{c} = \sqrt{\frac {\partial p}{\partial \tilde{\rho}}} = \sqrt{a\gamma} e^{\frac{S_{0}}{2}} \tilde{\rho}^{\frac{\gamma -1}{2}} $). Therefore, we have

    $ F1(˜ρ,˜m)˜ρ=˜ρα2(p˜ρ˜ρ2˜m2)<0p˜ρ˜ρ2<˜m2
    $
    (2.5)

    and

    $ F1(˜ρ,˜m)˜ρ=˜ρα2(p˜ρ˜ρ2˜m2)>0p˜ρ˜ρ2>˜m2.
    $
    (2.6)

    We conclude that $ \frac{\partial F_1(\tilde{\rho}, \tilde{m})}{\partial \tilde{\rho}} < 0 $ for $ \tilde{\rho} < \rho^{*} $ and $ \frac{\partial F_1(\tilde{\rho}, \tilde{m})}{\partial \tilde{\rho}} > 0 $ for $ \tilde{\rho} > \rho^{*} $. Furthermore, we have

    $ lim˜ρ0F1(˜ρ,˜m)=0,lim˜ρ+F1(˜ρ,˜m)=+,F1(ρ,˜m)<0,forα>1;
    $
    (2.7)
    $ lim˜ρ0F1(˜ρ,˜m)=+,lim˜ρ+F1(˜ρ,˜m)=+,F1(ρ,˜m)>0,forγ<α<1;
    $
    (2.8)

    and

    $ lim˜ρ0F1(˜ρ,˜m)=+,lim˜ρ+F1(˜ρ,˜m)=0,F1(ρ,˜m)<0,forα<γ.
    $
    (2.9)

    Then, for any fixed $ \tilde{m} = \rho_{-}u_{-} > 0 $, according to different regions of $ \alpha \in R $, we draw the graphs of $ F_1(\tilde{\rho}, \tilde{m}) $. See Figure 1 below.

    Figure 1.  Plot of $ \tilde{\rho}\rightarrow F_{1}(\tilde{\rho}, m) $.

    Integrating $ (2.2)_{2} $ over $ (0, x) $, we obtain

    $ F1(˜ρ(x),˜m)F1(ρ,˜m)=β˜mα+1x.
    $
    (2.10)

    If $ \beta < 0 $, by $ (2.10) $, $ F_{1}(\tilde{\rho}, \tilde{m}) $ will decrease as the length of ducts increases, until it arrives at the minimum $ F_{1}(\rho^{*}, \tilde{m}) $, no matter whether the upstream is supersonic (i.e., $ \rho_{-} < \rho^{*} $) or subsonic (i.e., $ \rho_{-} > \rho^{*} $). Therefore, we get the maximal length of ducts

    $ Lm=1β[u1α1α+aγeS0γ+αργ1uα1+(aγeS0)1αγ+1(ρu)(1α)(γ1)γ+1(1α11γ+α)]
    $
    (2.11)

    for a supersonic or subsonic flow before it gets choked, which is the state where the flow speed is equal to the sonic speed.

    However, if $ \beta > 0 $, $ \alpha > 1 $ and the upstream is supersonic (i.e., $ \rho_{-} < \rho^{*} $), by (2.7), (2.10) and Figure 1 (i), we know that $ \tilde{\rho} $ is decreasing as duct length $ x $ increases. Then, we get the maximal length of ducts

    $ Lm=1β(u1αα1aγeS0γ+αργ1uα1)
    $
    (2.12)

    for a supersonic flow before it reaches the vacuum state. If $ -\gamma < \alpha < 1 $ or $ \alpha < -\gamma $, by (2.8)–(2.10) and Figure 1(ii) and (iii), $ \tilde{\rho} $ is decreasing as the duct length $ x $ increases for supersonic upstream, too. But, the vacuum will never occur for any duct length $ L $.

    Moreover, if $ \beta > 0 $, $ \alpha < -\gamma $ and the upstream is subsonic (i.e., $ \rho_{-} > \rho^{*} $), combining (2.9), (2.10) with Figure 1(iii), $ \tilde{\rho} $ is increasing as the duct length $ x $ increases. At the same time, $ F_{1}(\tilde{\rho}, \tilde{m}) $ is increasing and approaching its supremum 0. Then, we get the maximal length of the duct $ L_{m} $, which is still as shown in (2.12). When $ L > L_{m} $, the fluid velocity is zero, that is, the fluid stagnates in a finite place. While, if $ -\gamma < \alpha < 1 $ or $ \alpha > 1 $, again, by (2.7), (2.8), (2.10) and Figure 1(i) and (ii), $ \tilde{\rho} $ is also increasing as the duct length $ x $ increases, but $ F_{1}(\tilde{\rho}, \tilde{m}) $ goes to infinity as $ \tilde{\rho} $ grows. In this case, although the fluid is slowing down, it does not stagnate at any finite place.

    Case 2: $ \alpha = 1 $.

    Now, (2.2) turns into

    $ {˜m=ρu,F2(˜ρ,˜m)x=β˜m2,
    $
    (2.13)

    where

    $ F_2\left(\tilde{\rho},\tilde{m}\right) = -\tilde{m}^{2} \ln \tilde{\rho}+\frac{a \gamma e^{S_{0}}}{\gamma+1} \tilde{\rho}^{\gamma+1}. $

    And, we get

    $ lim˜ρ0F2(˜ρ,˜m)=+,lim˜ρ+F2(˜ρ,˜m)=+,
    $
    (2.14)
    $ F2(˜ρ,˜m)˜ρ=˜ρ(˜m2˜ρ2+aγeS0˜ργ1),
    $
    (2.15)

    and

    $ F2(˜ρ(x),˜m)F2(ρ,˜m)=β˜m2x.
    $
    (2.16)

    Similarly, the function $ F_{2}\left(\tilde{\rho}(x), \tilde{m}\right) $ gets its minimum at point $ \tilde{\rho} = \rho^{*} $. If $ \beta < 0 $, combining (2.14) with (2.16), we get the maximal length of ducts

    $ Lm=1β(γ+1)(lnρ1γu2aγeS0+aγeS0ργ1u21)
    $
    (2.17)

    for a supersonic or subsonic flow before it gets choked. While, if $ \beta > 0 $, the flow remains in its entrance state for any duct length $ L > 0 $, no matter whether it is supersonic or subsonic.

    Case 3: $ \alpha = -\gamma $.

    In this case, $ (2.1) $ changes into

    $ {˜m=ρu,F3(˜ρ,˜m)x=β˜m1γ,
    $
    (2.18)

    where

    $ F_3(\tilde{\rho},\tilde{m}) = \frac{\tilde{m}^2}{1+\gamma} \tilde{\rho}^{-\gamma-1}+a \gamma e^{S_{0}} \ln \tilde{\rho}. $

    Then, we have

    $ lim˜ρ0F3(˜ρ,˜m)=+,lim˜ρ+F3(˜ρ,˜m)=+,
    $
    (2.19)
    $ F3(˜ρ,˜m)˜ρ=˜ργ(˜m2˜ρ2+aγeS0˜ργ1),
    $
    (2.20)

    and

    $ F3(˜ρ(x),˜m)F3(ρ,˜m)=β˜m1γx.
    $
    (2.21)

    Similar to the other two cases, the function $ F_{3}\left(\tilde{\rho}(x), \tilde{m}\right) $ gets its minimum at point $ \tilde{\rho} = \rho^{*} $. If $ \beta < 0 $, by (2.19) and (2.21), we obtain the maximal length of ducts

    $ Lm=1β(1+γ)[uγ+1+aγeS0(ρu)γ1ln(aγeS01ργ1u2)]
    $
    (2.22)

    for a supersonic or subsonic flow before it gets choked. While, if $ \beta > 0 $, again, by (2.19) and (2.21), the flow also keeps the upstream supersonic or subsonic state for any duct length $ L > 0 $.

    To sum up, we draw the following conclusion from the above analysis:

    Lemma 2.1. If $ \rho_{-} \neq \rho^{*} > 0, \; u_{-} > 0, \; c^{*} = (a \gamma e^{S_{0}})^{\frac{1}{\gamma+1}}(\rho_{-} u_{-})^{\frac{\gamma-1}{\gamma+1}} > 0 $ and the duct length $ L < L_{m} $, where $ L_{m} $ is the maximal allowable duct length given in (2.11), (2.12), (2.17) and (2.22), then the Cauchy problem (1.4) admits a unique smooth positive solution $ (\tilde{\rho}(x), \tilde{u}(x), S_{0}) ^{\top} $ which satisfies the following properties:

    1) $ 0 < \rho_{-} < \tilde{\rho}(x) < \rho^{*} $ and $ c^{*} < \tilde{u}(x) < u_{-} $, \quad if $ \beta < 0 $ and $ \rho_{-} < \rho^{*} $;

    2) $ 0 < \rho^{*} < \tilde{\rho}(x) < \rho_{-} $ and $ u_{-} < \tilde{u}(x) < c^{*} $, \quad if $ \beta < 0 $ and $ \rho_{-} > \rho^{*} $;

    3) $ 0 < \tilde{\rho}(x) < \rho_{-} $ and $ c^{*} < u_{-} < \tilde{u}(x) < +\infty $, \quad if $ \beta > 0 $ and $ \rho_{-} < \rho^{*} $;

    4) $ 0 < \rho_{-} < \tilde{\rho}(x) < +\infty $ and $ 0 < \tilde{u}(x) < u_{-} < c^{*} $, \quad if $ \beta > 0 $ and $ \rho_{-} > \rho^{*} $;

    5) $ \tilde{\rho} \tilde{u} = \rho_{-}u_{-} $;

    6) $ \|(\tilde{\rho}(x), \tilde{u}(x), S_{0})\|_{C^{2}([0, L])} < M_{0}, $ where $ M_{0} $ is a constant only depending on $ \alpha $, $ \beta $, $ \gamma $, $ \rho_{-} $, $ u_{-} $, $ S_{0} $ and $ L $.

    Remark 2.1. The following is worth pointing out:

    1) When $ \beta > 0 $ and the upstream is supersonic, a vacuum can occur at the finite place for $ \alpha > 1 $, while a vacuum will never happen in any finite ducts for $ \alpha\leq 1 $;

    2) When $ \beta > 0 $ and the upstream is subsonic, fluid velocity can be zero at the finite place for $ \alpha < -\gamma $, while the movement of fluid will never stop in the duct for $ \alpha \geq -\gamma $;

    3) For the case of $ \beta = 0 $, we refer the reader to [19] for details.

    Thus, from Lemma 2.1 and Remark 2.1, we can directly get Theorem 1.1.

    In order to answer the two problems proposed in the introduction, we introduce a wave decomposition for system (1.1) in this section. Here, we choose the steady supersonic solution $ \tilde{W}(x) = (\tilde{\rho}(x), \tilde{u}(x), \tilde{S}(x))^{\top} $ (with $ \tilde{u}(x) > 0 $) as the background solution, which satisfies (1.4). For system (1.1), the corresponding simplification system has the form

    $ {ρt+ρxu+ρux=0,ut+uux+aγeSργ2ρx+aeSργ1Sx=βuα+1,St+uSx=0.
    $
    (3.1)

    Let us denote $ W(t, x) = \bar{W}(t, x)+\tilde{W}(x) $, where $ \bar{W} = (\bar{\rho}, \bar{u}, \bar{S})^\top $ is the perturbation around the background solution. Substituting

    $ ρ(t,x)=ˉρ(t,x)+˜ρ(x),u(t,x)=ˉu(t,x)+˜u(x),S(t,x)=ˉS(t,x)+S0
    $
    (3.2)

    into (3.1) yields

    $ {ˉρt+uˉρx+ρˉux+˜ρxˉu+˜uxˉρ+˜u˜ρx+˜ρ˜ux=0,ˉut+uˉux+ˉu˜ux+˜ux˜u+aγeSργ2(ˉρx+˜ρx)+aeSργ1ˉSx=β(ˉu+˜u)α+1,¯St+uˉSx=0.
    $
    (3.3)

    Combining this with (1.4), system (3.3) can be simplified as

    $ {ˉρt+uˉρx+ρˉux=˜uxˉρ˜ρxˉu,ˉut+uˉux+aγeSργ2ˉρx+aeSργ1ˉSx=Θ(ρ,˜ρ,S,S0)eˉSˉρ˜ρx˜uxˉug(u,˜u)ˉu,¯St+uˉSx=0,
    $
    (3.4)

    where $ \Theta(\rho, \tilde{\rho}, S, S_{0}) e^{\bar{S}} \bar{\rho} = a \gamma (e^{S} \rho ^{\gamma-2}-e^{S_{0}} \tilde{\rho}^{\gamma-2}) $ and $ g(u, \tilde{u}) \bar{u} = -\beta [(\bar{u}+\tilde{u})^{\alpha+1}-\tilde{u}^{\alpha+1}] $. $ g(u, \tilde{u}) $ can be represented as follows:

    $ g(u,\tilde{u}) = -\beta(\alpha+1) \int_{0}^{1} (\theta \bar{u}+\tilde{u})^{\alpha} {\mathrm{d}} \theta. $

    Obviously, system (3.4) can be expressed as the following quasi-linear equations:

    $ ˉWt+A(W)ˉWx+H(˜W)ˉW=0,
    $
    (3.5)

    where

    $ A(W)=(uρ0aγeSργ2uaeSργ100u),
    $
    (3.6)
    $ H(˜W)=(˜ux˜ρx0Θ(ρ,˜ρ,S,˜S)eˉS˜ρx˜ux+g(u,˜u)0000).
    $
    (3.7)

    Through simple calculations, the three eigenvalues of system (3.5) are

    $ λ1(W)=uc,λ2(W)=u,λ3(W)=u+c,
    $
    (3.8)

    where $ c = \sqrt{a\gamma} e^{\frac{S}{2}} \rho^{\frac{\gamma -1}{2}} $. The three right eigenvectors $ r_{i}(W)\; (i = 1, 2, 3) $ corresponding to $ \lambda_{i}\; (i = 1, 2, 3) $ are

    $ {r1(W)=1ρ2+c2(ρ,c,0),r2(W)=1ρ2+γ2(ρ,0,γ),r3(W)=1ρ2+c2(ρ,c,0).
    $
    (3.9)

    The left eigenvectors $ l_{i}(W)\; (i = 1, 2, 3) $ satisfy

    $ li(W)rj(W)δij,ri(W)ri(W)1,(i,j=1,2,3),
    $
    (3.10)

    where $ \delta_{i j} $ represents the Kroneckeros symbol. It is easy to get the expression for $ l_{i}(W) $ as follows:

    $ {l1(W)=ρ2+c22(ρ1,c1,0),l2(W)=ρ2+γ22(ρ1,0,γ1),l3(W)=ρ2+c22(ρ1,c1,0).
    $
    (3.11)

    Besides, $ l_{i}(W) $ and $ r_{i}(W) $ have the same regularity.

    Let

    $ μi=li(W)ˉW,ϖi=li(W)ˉWx,μ=(μ1,μ2,μ3),ϖ=(ϖ1,ϖ2,ϖ3);
    $
    (3.12)

    then,

    $ ˉW=3k=1μkrk(W),ˉWx=3k=1ϖkrk(W).
    $
    (3.13)

    Noticing (3.5) and (3.12), we have

    $ dμidit=d(li(W)ˉW)dit=d(ˉW)ditli(W)ˉW+λi(W)˜Wli(W)ˉWli(W)H(˜W)ˉW,
    $
    (3.14)

    where

    $ li(W)=(W1(li(W))W2(li(W))W3(li(W))).
    $
    (3.15)

    By using (3.5) and (3.13), we get

    $ d(ˉW)dit=ˉWt+λi(W)(ˉW)x=3k=1(λi(W)λk(W))ϖkrk(W)H(˜W)ˉW.
    $
    (3.16)

    Thus, noting $ \nabla(l_{i}(W) r_{j}(W)) = 0 $ and $ \nabla l_{i}(W) r_{j}(W) = -l_{i}(W) \nabla r_{j}(W) $, we get

    $ dμidit=μit+λi(W)μix=3j,k=1Φijk(W)ϖjμk+3j,k=1˜Φijk(W)μjμk3k=1˜˜Φik(W)μk,
    $
    (3.17)

    where

    $ Φijk(W)=(λj(W)λi(W))li(W)Wrj(W)rk(W),˜Φijk(W)=li(W)H(˜W)Wrj(W)rk(W),˜˜Φik(W)=λi(W)li(W)˜WWrk(W)+li(W)H(˜W)rk(W),
    $
    (3.18)

    and

    $ Φiik(W)0,k=1,2,3.
    $
    (3.19)

    Similarly, we have from (3.10) and (3.13) that

    $ dϖidit=d(li(W)ˉWx)dit=3k=1ϖkd(li(W))ditrk(W)+li(W)d(ˉWx)dit,
    $
    (3.20)

    and

    $ d(li(W))ditrk(W)=li(W)d(rk(W))dit=3s=1li(W)(rk(W))Wsd(Ws)dit=3s=1Cksi(W)(dˉWsdit+d˜Wsdit),
    $
    (3.21)

    where $ C_{ksi} (W) = l_{i} (W) \frac{\partial \left(r_{k} (W) \right)}{\partial W_{s}} $. It is concluded from (3.16) that

    $ d(ˉWs)dit=3j=1(λi(W)λj(W))ϖjrjs(W)H(˜W)ˉW.
    $
    (3.22)

    Therefore,

    $ 3k=1ϖkd(li(W))ditrk(W)=3j,k,s=1ϖkCksi(λj(W)λi(W))ϖjrjs(W)3k,s=1Cksiλi˜Wsxϖk+3k,s=1CksiϖkH(˜W)ˉW.
    $
    (3.23)

    Then,

    $ li(W)dˉWxdit=li(W)(ˉWxt+A(W)ˉWxx)=3k,s=1li(W)(A(W))Ws(ˉWs+˜Ws)xϖkrkli(W)(H(˜W)ˉW)x,
    $
    (3.24)

    where we used (3.5). By differentiating

    $ A(W) r_{k}(W) = \lambda_{k} (W) r_{k}(W) $

    with respect to $ W_{s} $ and multiplying the result by $ l_{i}(W) $, we get

    $ li(W)(A(W))Wsrk=li(W)(λk)Wsrk+li(W)λk(rk)Wsli(W)A(W)(rk)Ws=(λk)Wsδik+(λkλi)Cksi(W).
    $
    (3.25)

    Thus,

    $ dϖidit=3k=1ϖkd(li(W))ditrk(W)+li(W)d(ˉWx)dit=3j,k=1Υijk(W)ϖjϖk+3j,k=1˜Υijk(W)ϖkli(W)H(˜W)xˉW,
    $
    (3.26)

    where

    $ Υijk(W)=(λj(W)λk(W))li(W)Wrk(W)rj(W)Wλk(W)rj(W)δik,˜Υijk(W)=λk(W)li(W)Wrk(W)˜W+li(W)WrkH(˜W)rjμj(W)Wλk(W)δik˜Wli(W)H(˜W)rk(W).
    $

    In view of Lemma 2.1, it is clear that the term $ H(\tilde{W})_{x} $ in (3.26) is meaningful.

    For the convenience of the later proof, we can rewrite system (3.5) as

    $ ˉWx+A1(W)ˉWt+A1(W)H(˜W)ˉW=0
    $
    (3.27)

    by swapping the variables $ t $ and $ x $. Here, we represent the eigenvalues, left eigenvectors and right eigenvectors of the matrix $ A^{-1}(W) $ as $ \hat{\lambda}_{i} $, $ \hat{l}_{i}(W) $ and $ \hat{r}_{i}(W), i = 1, 2, 3 $, respectively.

    Let

    $ ˆμi=ˆli(W)ˉW,ˆϖi=ˆli(W)ˉWt,ˆμ=(ˆμ1,ˆμ2,ˆμ3),ˆϖ=(ˆϖ1,ˆϖ2,ˆϖ3).
    $
    (3.28)

    Similar to the above arguments, we can get similar results by combining (3.27) and (3.28):

    $ dˆμidit=ˆμix+ˆλi(W)ˆμit=3j,k=1ˆΦijk(W)ˆϖjˆμk+3j,k=1ˆ˜Φijk(W)ˆμjˆμk3k=1ˆ˜˜Φik(W)ˆμk,
    $
    (3.29)

    with

    $ ˆΦijk(W)=(ˆλj(W)ˆλi(W))ˆli(W)Wˆrj(W)ˆrk(W),ˆ˜Φijk(W)=ˆλj(W)ˆli(W)H(˜W)Wˆrj(W)ˆrk(W),ˆ˜˜Φik(W)=ˆli(W)˜WWˆrk(W)+ˆλi(W)ˆli(W)H(˜W)ˆrk(W),
    $

    and

    $ dˆϖidit=ˆϖix+ˆλi(W)ˆϖit=3j,k=1ˆΥijk(W)ˆϖjˆϖk+3j,k=1ˆ˜Υijk(W)ˆϖkˆli(W)(A1H(˜W))tˉW,
    $
    (3.30)

    where

    $ ˆΥijk(W)=(ˆλj(W)ˆλk(W))ˆli(W)Wˆrk(W)ˆrj(W)Wˆλk(W)ˆrj(W)δik,ˆ˜Υijk(W)=ˆli(W)Wˆrk(W)˜W+ˆli(W)Wˆrk(W)A1H(˜W)ˆrjˆμj(W)ˆli(W)A1(W)H(˜W)ˆrk(W).
    $

    The wave decomposition for the initial data

    $ \bar{W}(t,x)|_{t = 0} = \bar{W}_{0}(x) = (\bar{\rho}_{0}(x),\bar{u}_{0}(x),\bar{S}_{0}(x))^{\top} $

    and boundary conditions

    $ \bar{W}(t,x)|_{x = 0} = \bar{W}_{l}(t) = (\bar{\rho}_{l}(t), \bar{u}_{l}(t),\bar{S}_{l}(t)) ^{\top} $

    have the following form:

    $ μ0=(μ10,μ20,μ30),ϖ0=(ϖ10,ϖ20,ϖ30),ˆμl=(ˆμ1l,ˆμ2l,ˆμ3l),ˆϖl=(ˆϖ1l,ˆϖ2l,ˆϖ3l),
    $
    (3.31)
    $ μl=(μ1l,μ2l,μ3l),ϖl=(ϖ1l,ϖ2l,ϖ3l),
    $
    (3.32)

    with

    $ μi0=li(W0)¯W0,ϖi0=li(W0)x(ˉW0),ˆμil=ˆli(Wl)ˉWl,ˆϖil=ˆli(Wl)t(ˉWl),
    $
    (3.33)
    $ μil=li(Wl)¯Wl,ϖil=li(Wl)x(ˉWl),
    $
    (3.34)

    where

    $ W_{0} = (\rho_{0},u_{0},S_{0})^{\top},\; W_{l} = (\rho_{l},u_{l},S_{l})^{\top}. $

    In this section, based on wave decomposition, we prove the global existence and stability of smooth solutions under small perturbations around the steady-state supersonic flow in region $ G = \{(t, x)|t\geq0, \; x\in[0, L]\} $. The initial data and boundary conditions satisfy the compatibility conditions at point (0, 0) (see [11]).

    In order to verify Theorem 1.2, we first establish a uniform prior estimate of the supersonic classical solution. That is, we assume that

    $ |μi(t,x)|Kε,|ϖi(t,x)|Kε,(t,x)G,i=1,2,3,
    $
    (4.1)

    when

    $ (ˉρ0,ˉu0,ˉS0)C1([0,L])<ε,(ˉρl,ˉul,ˉSl)C1([0,+))<ε,
    $
    (4.2)

    where $ \varepsilon $ is a suitably small positive constant. Here and hereafter, $ K $, $ K_{i} $ and $ K_{i}^{*} $ are constants that depend only on $ L $, $ \varepsilon $, $ \left\| (\tilde{\rho}, \tilde{u}, S_{0})) \right\|_{C^{2}([0, L])} $ and $ T_{1} $, defined by

    $ T1=mint0,x[0,L]i=1,2,3Lλi(W(t,x))>0.
    $
    (4.3)

    Here, $ \lambda_{1} $, $ \lambda_{2} $ and $ \lambda_{3} $ are the three eigenvalues of system (3.5). Combining (3.9) and (3.13), (4.1) means

    $ |ˉW(t,x)|Kε,|ˉWx(t,x)|Kε,(t,x)G.
    $
    (4.4)

    In what follows, we will show the validity of the hypothesis given by (4.1).

    Let $ x = x^{*}_{j}(t)\; (j = 1, 2, 3) $ be the characteristic curve of $ \lambda_{j} $ that passes through (0, 0):

    $ dxj(t)dt=λj(W(t,xj(t))),xj(0)=0.
    $
    (4.5)

    Since $ \lambda_{3}(W) > \lambda_{2}(W) > \lambda_{1}(W) $, we have that $ x = x^{*}_{3}(t) $ lies below $ x = x^{*}_{2}(t) $ and $ x = x^{*}_{2}(t) $ lies below $ x = x^{*}_{1}(t) $. In what follows, we divide domain $ G = \{(t, x)|t\geq0, \; x\in[0, L]\} $ into several different regions.

    Region 1: The region $ G_{1} = \left\{(t, x) \mid 0 \leq t \leq T_{1}, \; 0 \leq x \leq L, \; x \geq x_{3}^{*}(t)\right\} $.

    For any point $ (t, x) \in G_{1} $, integrating the $ i $-th equation in (3.17) along the $ i $-th characteristic curve about $ t $ from 0 to $ t $, we have

    $ |μi(t,x(t))|=|μi(0,bi)|+t03j,k=1|Φijk(W)ϖjμk|dτ+t03j,k=1|˜Φijk(W)μjμk|dτ+t03k=1|˜˜Φik(W)μk|dτ|μi0(bi)|+K1t0|μ(τ,x(τ))|dτ,i=1,2,3,
    $
    (4.6)

    where we have used (4.3) and (4.4) and assumed that the line intersects the $ x $ axis at $ (0, b_{i}) $. Similarly, integrating the $ i $-th equation in (3.26) along the $ i $-th characteristic curve and assuming that the line intersects the $ x $ axis at $ (0, b_{i}) $ again, we get

    $ |ϖi(t,x(t))|=|ϖi(0,bi)|+t03j,k=1|Υijk(W)ϖjϖk|dτ+t03j,k=1|˜Υijk(W)ϖk|dτ+t0|li(W)H(˜W)xˉW|dτ|ϖi0(bi)|+K2t0|ϖ(τ,x(τ))|dτ+t0[12|˜uxxρc(ΘeˉS˜ρx)xc˜ρxxρ±˜uxx±gx(u,˜u)||μ1|+12|˜uxxρc(ΘeˉS˜ρx)x+c˜ρxxρ˜uxxgx(u,˜u)||μ3|]dτ|ϖi0(bi)|+K2t0|ϖ(τ,x(τ))|dτ+K2t0|μ(τ,x(τ))|dτ,i=1,3,
    $
    (4.7)

    and

    $ |ϖ2(t,x(t))|=|ϖ2(0,b2)|+t03j,k=1|Υijk(W)ϖjϖk|dτ+t03j,k=1|˜Υijk(W)ϖk|dτ+t0|l2(W)H(˜W)xˉW|dτ|ϖ20(b2)|+K3t0|ϖ(τ,x(τ))|dτ+t0[ρ2+γ22ρ2+c2(|˜uxxc˜ρxxρ||μ1|+|˜uxx+c˜ρxxρ||μ3|)]dτ|ϖ20(b2)|+K3t0|ϖ(τ,x(τ))|dτ+K3t0|μ(τ,x(τ))|dτ,
    $
    (4.8)

    where $ \Theta = \Theta(\rho, \tilde{\rho}, S, S_{0}) $. Adding (4.6)–(4.8) together, for any $ i = 1, 2, 3 $, and using Gronwall's inequality, one gets

    $ |μ(t,x)|+|ϖ(t,x)|eK4T1(μ0C0([0,L])+ϖ0C0([0,L])).
    $
    (4.9)

    Due to the boundedness of $ T_{1} $, the arbitrariness of $ (t, x)\in G_{1} $ and (4.9), it holds that

    $ max(t,x)G1{|μ(t,x)|+|ϖ(t,x)|}K(μ0C0([0,L])+ϖ0C0([0,L])).
    $
    (4.10)

    Region 2: The region $ G_{2} = \left\{(t, x) \mid t \geq 0, \; 0 \leq x \leq L, \; 0 \leq x \leq x_{1}^{*}(t)\right\} $.

    We make the change of variables $ t $ and $ x $. For any point $ (t, x) \in G_{2} $, integrating (3.29) along the $ i $-th characteristic curve about $ x $, it follows that

    $ |ˆμi(t(x),x)||ˆμil(ti)|+K5x0|ˆμ(t(ς),ς)|dς,i=1,2,3,
    $
    (4.11)

    where we assumed that the line intersects the $ t $ axis at the point $ (t_{i}, 0) $. Similarly, repeating the above procedure for (3.30), we get

    $ |ˆϖi(t(x),x)||ˆϖil(ti)|+K6x0|ˆϖ(t(ς),ς)|dς+K6x0|ˆμ(t(ς),ς)|dς,i=1,2,3.
    $
    (4.12)

    Summing up (4.11) and (4.12) for $ i $ = 1, 2, 3 and applying Gronwall's inequality, we obtain

    $ max(t,x)G2{|ˆμ(t,x)|+|ˆϖ(t,x)|}K(ˆμlC0([0,+))+ˆϖlC0([0,+))),(t,x)G2,
    $
    (4.13)

    where we exploit the arbitrariness of $ (t, x)\in G_{2} $.

    Region 3: The region $ G_{3} = \left\{(t, x) \mid 0 \leq t \leq T_{1}, \; 0 \leq x \leq L, \; x_{2}^{*}(t) \leq x \leq x_{3}^{*}(t)\right\} $.

    For any point $ (t, x) \in G_{3} $, integrating the 1st and 2nd equations in (3.17) and (3.26) along the 1st and 2nd characteristic curve, we get

    $ |μ1(t,x(t))||μ10(x1)|+K7t0|μ(τ,x(τ))|dτ,
    $
    (4.14)
    $ |ϖ1(t,x(t))||ϖ10(x1)|+K8t0|ϖ(τ,x(τ))|dτ+t0[12|2˜uxxρc(ΘeˉS˜ρx)xc˜ρxxρ+gx(u,˜u)||μ1|+12|ρc(ΘeˉS˜ρx)x+c˜ρxxρgx(u,˜u)||μ3|]dτ|ϖ10(x1)|+K8t0|ϖ(τ,x(τ))|dτ+K8t0|μ(τ,x(τ))|dτ,
    $
    (4.15)
    $ |μ2(t,x(t))||μ20(x2)|+K9t0|μ(τ,x(τ))|dτ,
    $
    (4.16)

    and

    $ |ϖ2(t,x(t))||ϖ20(x2)|+K10t0|ϖ(τ,x(τ))|dτ+K10t0|μ(τ,x(τ))|dτ,
    $
    (4.17)

    where we assumed that the line intersects the $ x $ axis at points $ (0, x^{\prime}_{1}) $ and $ (0, x^{\prime}_{2}) $, respectively. Similarly, integrating the 3rd equations in (3.17) and (3.26) along the 3rd characteristic curve, one has

    $ |μ3(t,x(t))||μ3l(t3)|+K11tt3|μ(τ,x(τ))|dτ|μ3l(t3)|+K11t0|μ(τ,x(τ))|dτ,
    $
    (4.18)

    and

    $ |ϖ3(t,x(t))||ϖ3l(t3)|+K12tt3|ϖ(τ,x(τ))|dτ+tt3[12|ρc(ΘeˉS˜ρx)xc˜ρxxρgx(u,˜u)||μ1|+12|2˜uxx+ρc(ΘeˉS˜ρx)x+c˜ρxxρ+gx(u,˜u)||μ3|]dτ|ϖ3l(t3)|+K12t0|ϖ(τ,x(τ))|dτ+K12t0|μ(τ,x(τ))|dτ,
    $
    (4.19)

    where the point $ (t^{\prime}_{3}, 0) $ is the intersection of the line and the $ t $ axis.

    Since the boundary data are small enough, we sum up $ (4.14)-(4.19) $ and apply Gronwall's inequality to obtain the following:

    $ max(t,x)G3{|μ(t,x)|+|ϖ(t,x)|}K(μ0C0([0,L])+ϖ0C0([0,L])+μlC0([0,+))+ϖlC0([0,+))),
    $
    (4.20)

    where we exploit the arbitrariness of $ (t, x)\in G_{3} $.

    Region 4: The region $ G_{4} = \left\{(t, x) \mid 0 \leq t \leq T_{1}, \; 0 \leq x \leq L, \; x_{1}^{*}(t) \leq x \leq x_{2}^{*}(t)\right\} $.

    For any point $ (t, x) \in G_{4} $, integrating the 1st equations in (3.17) and (3.26) along the 1st characteristic curve, we get

    $ |μ1(t,x(t))||μ10(x1)|+K13t0|μ(τ,x(τ))|dτ,
    $
    (4.21)

    and

    $ |ϖ1(t,x(t))||ϖ10(x1)|+K14t0|ϖ(τ,x(τ))|dτ+K14t0|μ(τ,x(τ))|dτ,
    $
    (4.22)

    where we assumed that the line intersects the $ x $ axis at $ (0, x^{\prime \prime}_{1}) $. Similarly, integrating the 2nd and 3rd equations in (3.17) and (3.26) along the 2nd and 3rd characteristic curve, one has

    $ |μ2(t,x(t))||μ2l(t2)|+K15t0|μ(τ,x(τ))|dτ,
    $
    (4.23)
    $ |ϖ2(t,x(t))||ϖ2l(t2)|+K16t0|ϖ(τ,x(τ))|dτ+K16t0|μ(τ,x(τ))|dτ,
    $
    (4.24)
    $ |μ3(t,x(t))||μ3l(t3)|+K17t0|μ(τ,x(τ))|dτ,
    $
    (4.25)

    and

    $ |ϖ3(t,x(t))||ϖ3l(t3)|+K18t0|ϖ(τ,x(τ))|dτ+K18t0|μ(τ,x(τ))|dτ,
    $
    (4.26)

    where the line intersects the $ t $ axis at points $ (t^{\prime \prime}_{2}, 0) $ and $ (t^{\prime \prime}_{3}, 0) $, respectively.

    Noticing that the boundary data are small enough, we sum $ (4.21) $–$ (4.26) $ and then apply Gronwall's inequality to obtain

    $ max(t,x)G4{|μ(t,x)|+|ϖ(t,x)|}K(μ0C0([0,L])+ϖ0C0([0,L])+μlC0([0,+))+ϖlC0([0,+))),
    $
    (4.27)

    where we exploit the arbitrariness of $ (t, x)\in G_{4} $.

    From (4.10), (4.13), (4.20) and (4.27), we have proved that the assumption of (4.1) is reasonable. Therefore, we have obtained a uniform $ C^{1} $ a priori estimate for the classical solution. Thanks to the classical theory in [34], we further obtain the global existence and uniqueness of $ C^{1} $ solutions (see [11,35,36,37,38,39]) for problems (1.1)–(1.3). This proves Theorem 1.2.

    In this section, we show that the smooth supersonic solution $ W(t, x) = (\rho(t, x), u(t, x), S(t, x))^\top $ is temporal-periodic with a period $ P > 0 $, after a certain start-up time $ T_{1} $, under the temporal periodic boundary conditions. Here, we have assumed that $ W_{l}(t+P) = W_{l}(t) $ with $ P > 0 $.

    For system (1.1), Riemann invariants $ \xi $, $ \eta $ and $ \zeta $ are introduced as follows:

    $ ξ=u2γ1c,η=S,ζ=u+2γ1c.
    $
    (5.1)

    Then, system (1.1) can be transformed into the following form:

    $ {ξt+λ1(ξ,ζ)ξx=β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx,ηt+λ2(ξ,ζ)ηx=0,ζt+λ3(ξ,ζ)ζx=β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx,
    $
    (5.2)

    where

    $ \lambda_{1} = u-c = \frac{\gamma +1}{4} \xi+\frac{3- \gamma }{4} \zeta,\quad \lambda_{2} = u = \frac{1}{2} (\xi+\zeta),\quad \lambda_{3} = u+c = \frac{3-\gamma }{4} \xi+\frac{\gamma +1}{4} \zeta $

    are three eigenvalues of system (1.1). For supersonic flow (i.e., $ u > c $), we know that $ \lambda_{3} > \lambda_{2} > \lambda_{1} > 0 $. Obviously, (1.2)–(1.3) can be written as

    $ ξ(0,x)=ξ0(x),η(0,x)=η0(x),ζ(0,x)=ζ0(x),0xL,
    $
    (5.3)
    $ ξ(t,0)=ξl(t),η(t,0)=ηl(t),ζ(t,0)=ζl(t),t0,
    $
    (5.4)

    where $ \xi_{l} (t+P) = \xi_{l} (t), \; \eta_{l} (t+P) = \eta_{l} (t) $ and $ \zeta_{l} (t+P) = \zeta_{l} (t) $ with $ P > 0 $.

    We swap $ t $ and $ x $ so that the problem described by (5.2)–(5.4) takes the following form:

    $ {ξx+1λ1ξt=1λ1[β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx],ηx+1λ2ηt=0,ζx+1λ3ζt=1λ3[β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx],ξ(t,0)=ξl(t),η(t,0)=ηl(t),ζ(t,0)=ζl(t),
    $
    (5.5)

    where $ t > 0 $ and $ x\in[0, L] $. Next, we set

    $ V=(ξ˜ξ,η˜η,ζ˜ζ),Λ(t,x)=(1λ1(ξ(t,x),ζ(t,x))0001λ2(ξ(t,x),ζ(t,x))0001λ3(ξ(t,x),ζ(t,x)));
    $
    (5.6)

    then, the Cauchy problem (5.5) can be simplified as follows:

    $ Vx+Λ(t,x)Vt=Λ(t,x)(β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx0β(ξ2+ζ2)α+1+γ116γ(ζξ)2ηx)(1˜λ1[β(˜ξ2+˜ζ2)α+1+γ116γ(˜ζ˜ξ)2˜η]01˜λ3[β(˜ξ2+˜ζ2)α+1+γ116γ(˜ζ˜ξ)2˜η]),
    $
    (5.7)

    where

    $ ˜ξ=˜u2γ1˜c,˜η=˜S,˜ζ=˜u+2γ1˜c,˜λ1=λ1(˜ξ,˜ζ)=γ+14˜ξ+3γ2˜ζ,˜λ2=λ2(˜ξ,˜ζ)=12˜ξ+12˜ζ,˜λ3=λ3(˜ξ,˜ζ)=3γ4˜ξ+γ+14˜ζ.
    $

    According to

    $ \|\rho-\tilde{\rho}\|_{C^{1}(G)}+\|u-\tilde{u}\|_{C^{1}(G)}+\|S-\tilde{S}\|_{C^{1}(G)} < K_{0} \varepsilon $

    and (5.1), we can easily obtain

    $ ξ(t,x)˜ξ(x)C1(G)+η(t,x)˜η(x)C1(G)+ζ(t,x)˜ζ(x)C1(G)<J1ε,
    $
    (5.8)

    where the constant $ J_{1}(>0) $ depends solely on $ \; \tilde{\rho}, \tilde{u}, \gamma $ and $ L $.

    In order to prove that $ W(t+P, x) = W(t, x) $, for any $ t > T_{1} $ and $ x \in [0, L] $, we first prove that the following conclusions hold:

    $ ξ(t+P,x)=ξ(t,x),η(t+P,x)=η(t,x),ζ(t+P,x)=ζ(t,x),t>T1,x[0,L],
    $
    (5.9)

    where $ T_{1} $ is the start-up time, which is defined in (4.3).

    Let

    $ \; N(t,x) = V(t+P,x)-V(t,x); $

    then, according to (5.7), we obtain

    $ {Nx+Λ(t,x)Nt=R(t,x),N(t,0)=0,t>0,
    $
    (5.10)

    where

    $ R(t,x)=Λ(t+P,x)(β(ξ(t+P,x)2+ζ(t+P,x)2)α+1+(γ1)(ζ(t+P,x)ξ(t+P,x))2ηx(t+P,x)16γ0β(ξ(t+P,x)2+ζ(t+P,x)2)α+1+(γ1)(ζ(t+P,x)ξ(t+P,x))2ηx(t+P,x)16γ)Λ(t,x)(β(ξ(t,x)2+ζ(t,x)2)α+1+(γ1)(ζ(t,x)ξ(t,x))2ηx(t,x)16γ0β(ξ(t,x)2+ζ(t,x)2)α+1+(γ1)(ζ(t,x)ξ(t,x))2ηx(t,x)16γ)[Λ(t+P,x)Λ(t,x)]Vt(t+P,x).
    $
    (5.11)

    Using the continuity of $ \lambda_{i}\; (i = 1, 2, 3) $ and (5.8), after some calculations, we obtain the following estimates:

    $ |Vt(t+P,x)|J2ε,
    $
    (5.12)
    $ |ξ(t+P,x)+ζ(t+P,x)|J3,
    $
    (5.13)
    $ |Λ(t,x)|J4,
    $
    (5.14)
    $ |Λ(t+P,x)Λ(t,x)|J5|N(t,x)|,
    $
    (5.15)
    $ |Λt(ξ(t,x),η(t,x))|J6ε,
    $
    (5.16)

    and

    $ |R(t,x)||Λ(t,x)|(J7|β||N(t,x)|+γ116γJ8J9|N(t,x)|0J7|β||N(t,x)|+γ116γJ8J9|N(t,x)|)+|Λ(t+P,x)Λ(t,x)|((J32)α+1|β|+γ116J23J80(J32)α+1|β|+γ116J23J8)+|Λ(t+P,x)Λ(t,x)||Vt(t+P,x)|J10|N(t,x)|,
    $
    (5.17)

    where the constants $ J_{i}\; (i = 2, \cdots, 10) $ depend only on $ \tilde{\rho}, \tilde{u}, \gamma $ and $ L $.

    In the above calculation, we have used

    $ |(ξ(t+P,x)2+ζ(t+P,x)2)α+1(ξ(t,x)2+ζ(t,x)2)α+1|=|uα+1(t+P,x)uα+1(t,x)|=|u(t+P,x)u(t,x)||(α+1)||10[u(t,x)+θ(u(t+P,x)u(t,x))]αdθ|J7|N(t,x)|,forα1;
    $
    $ |(\frac{\xi(t+P,x)}{2}+\frac{\zeta(t+P,x)}{2})^{\alpha+1}-(\frac{\xi(t,x)}{2}+\frac{\zeta(t,x)}{2})^{\alpha+1}| = 0 \leq J_{7} |N(t, x)|,\; \quad {\text{for}}\; \alpha = -1. $

    Now, fix a point $ (t^{*}, x^{*}) $ with $ t^{*} > T_{1} $ and $ 0 < x^{*} < L $. Let $ \Gamma_{1}: t = \check{t}_{1}(x) $ and $ \Gamma_{3}: t = \check{t}_{3}(x) $ be two characteristic curves passing through point $ (t^{*}, x^{*}) $, that is,

    $ dˇt1dx=1λ1(ξ(ˇt1,x),ζ(ˇt1,x)),ˇt1(x)=t,
    $
    (5.18)

    and

    $ dˇt3dx=1λ3(ξ(ˇt3,x),ζ(ˇt3,x)),ˇt3(x)=t,
    $
    (5.19)

    where $ x \in [0, x^{*}] $. Since $ \lambda_{3}(W) > \lambda_{1}(W) $, $ \Gamma_{1} $ lies below $ \Gamma_{3} $. Set

    $ Ψ(x)=12ˇt3(x)ˇt1(x)|N(t,x)|2dt,
    $
    (5.20)

    where $ 0 \leq x < x^{*} $. According to the definition of $ T_{1} $, and combining $ t^{*} > T_{1} $ and $ 0 \leq x^{*} \leq L $, we obtain that $ (\check{t}_{1}(0), \check{t}_{3}(0))\subset (0, +\infty) $. Then, it follows from (5.10) that $ N(t, 0) \equiv 0 $. Thus, $ \Psi(0) = 0 $.

    Taking the derivative of $ \Psi(x) $ with regard to $ x $ gives

    $ Ψ(x)=ˇt3(x)ˇt1(x)N(t,x)Nx(t,x)dt+12|N(ˇt3(x),x)|21λ3(ξ(ˇt3(x),x),ζ(ˇt3,x))12|N(ˇt1(x),x)|21λ1(ξ(ˇt1(x),x),ζ(ˇt1(x),x))ˇt3(x)ˇt1(x)N(t,x)Λ(t,x)Nt(t,x)dt+ˇt3(x)ˇt1(x)N(t,x)R(t,x)dt+12N(t,x)Λ(t,x)N(t,x)|t=ˇt3(x)t=ˇt1(x)=12ˇt3(x)ˇt1(x)[(N(t,x)Λ(t,x)N(t,x))tN(t,x)Λt(t,x)N(t,x)]dt+ˇt3(x)ˇt1(x)N(t,x)R(t,x)dt+12N(t,x)Λ(t,x)N(t,x)|t=ˇt3(x)t=ˇt1(x)=12ˇt3(x)ˇt1(x)N(t,x)Λt(t,x)N(t,x)dt+ˇt3(x)ˇt1(x)N(t,x)R(t,x)dt(J6ε+2J10)Ψ(x),
    $
    (5.21)

    where we used (5.16) and (5.17).

    Therefore, using Gronwall's inequality, we obtain that $ \Psi(x) \equiv 0 $. In addition, according to the continuity of $ \Psi(x) $, we obtain that $ \Psi(x^{*}) = 0 $; then, $ N(t^{*}, x^{*}) = 0 $. Using the arbitrariness of $ (t^{*}, x^{*}) $, we get

    $ N(t, x) \equiv 0,\quad \forall t > T_{1},\; x \in [0, L]. $

    Thus, (5.9) holds. Then, from $ (5.1) $ and $ c = \sqrt{a\gamma} e^{\frac{S}{2}} \rho^{\frac{\gamma -1}{2}} $, it follows that

    $ W(t+P,x) = W(t,x) $

    for any $ t > T_{1} $ and $ x \in [0, L] $, where $ T_{1} $ is the start-up time defined in (4.3). This proves Theorem 1.3.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    This work was supported in part by the Natural Science Foundation of China Grant No. 12101372, No. 12271310, and the Natural Science Foundation of Shandong Province Grant No. ZR2022MA088.

    The authors declare there is no conflict of interest.

    [1] Lelieveldt B, Geest R, Lamb H, et al. (2001) Automated observer-independent acquisition of cardiac short-axis MR images: A pilot study 1. Radiology 221: 537–542.
    [2] Lynch M, Ghita O, Whelan PF (2006) Left-ventricle myocardium segmentation using a coupled level-set with a priori knowledge. Comput Mem Imag Grap 30: 255–262.
    [3] Kohlberger T, Funka-Lea G, Desh V (2007) Soft level set coupling for LV segmentation in gated perfusion SPECT. In Medical Image Computing and Computer-Assisted Intervention (MICCAI) 4791: 327–334.
    [4] Feng C, Li C, Zhao D, et al. (2013) Segmentation of the left ventricle using distance regularized two-layer level set approach. In Medical Image Computing and Computer-Assisted Intervention (MICCAI) 8149: 477–484.
    [5] Tsai A, Yezzi A, Wells W, et al. (2003) A shape-based approach to the segmentation of medical imagery using level sets. IEEE Trans Med Imaging 22: 137–154.
    [6] Cremer D, Osher S, Soatto S (2006) Kernel density estimation and intrinsic alignment for shape priors in level set segmentation. Int J Comput Vision 69: 335–351. doi: 10.1007/s11263-006-7533-5
    [7] Liu W, Ruan D (2014) Segmentation with a shape dictionary. In Biomedical Imaging (ISBI), IEEE International Symposium on: 357–360.
    [8] Wang L, Li C, Sun Q, et al. (2008) Brain MR image segmentation using local and global intensity fitting active contours/surfaces. In Medical Image Computing and Computer-Assisted Intervention (MICCAI) 5241: 384–392.
    [9] Verma N, Muralidhar GS, Bovik AC, et al. (2011) Model-driven, probabilistic level set based segmentation of magnetic resonance images of the brain. In Engineering in Medicine and Biology Society (EMBC), IEEE Annual International Conference on: 2821–2824.
    [10] Li C, Xu C, Gui C, et al. (2005) Level set evolution without re-initialization: a new variational formulation. In Computer Vision and Pattern Recognition (CVPR), IEEE Computer Society Conference on 1: 430–436.
    [11] Osher S, Fedkiw R (2003) Level set methods and dynamic implicit surfaces. Springer.
    [12] Dempster AP, Laird NM, Rubin DB (1977) Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc Series B (Methodological) 39: 1–38.
    [13] Boyd S, Parikh N, Chu E, et al. (2011) Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends in Machine Learning 3: 1– 122.
    [14] Jolly MP, Xue H, Grady L, et al. (2009) Combining registration and minimum surfaces for the segmentation of the left ventricle in cardiac Cine MR images. In Medical Image Computing and Computer-Assisted Intervention (MICCAI) 5762: 910–918.
    [15] Constantinides C, Chenoune Y, Kachenoura N, et al. (2009) Semi-automated cardiac segmentation on Cine magnetic resonance images using GVF-snake deformable models. The MIDAS Journal-Cardiac MR Left Ventricle Segmentation Challenge.
    [16] Folkesson J, Samset E, Kwong R, et al. (2008) Unifying statistical classification and geodesic active regions for segmentation of cardiac MRI. IEEE Trans Inf Technol Biomed 12: 328–334. doi: 10.1109/TITB.2007.912179
  • This article has been cited by:

    1. Yakui Wu, Qiong Wu, Yue Zhang, Time decay estimates of solutions to a two-phase flow model in the whole space, 2024, 13, 2191-950X, 10.1515/anona-2024-0037
    2. Xue Wang, Guoxian Chen, A positivity-preserving well-balanced wet-dry front reconstruction for shallow water equations on rectangular grids, 2024, 198, 01689274, 295, 10.1016/j.apnum.2024.01.012
    3. Leilei Tong, Global existence and decay estimates of the classical solution to the compressible Navier-Stokes-Smoluchowski equations in ℝ3 , 2024, 13, 2191-950X, 10.1515/anona-2023-0131
    4. Xiao Han, Hui Wei, Multiplicity of the large periodic solutions to a super-linear wave equation with general variable coefficient, 2024, 16, 2836-3310, 278, 10.3934/cam.2024013
    5. Xixi Fang, Shuyue Ma, Huimin Yu, Temporal periodic solutions of non-isentropic compressible Euler equations with geometric effects, 2024, 13, 2191-950X, 10.1515/anona-2024-0049
  • Reader Comments
  • © 2015 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(5013) PDF downloads(1038) Cited by(2)

Figures and Tables

Figures(3)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog