This continuous Galerkin method discretizes the spatial variables and time variable simultaneously. Therefore, we can postulate a balance of energy over the interval of time, not only at some moments. In the formulation of the method, we integrate the physical quantities analytically in the time interval rather than numerically, as in the classical finite element method. The numerical method presented will be used to solve the evolutionary processes of the vibrations. In this paper, the velocity formulation of the space–time finite element method is used. The equations of motion are discretized both in space and time. This means that in the present numerical scheme the velocity is distributed in a finite space–time element according to interpolation functions and nodal velocities in two adjacent time layers used as parameters. The analytic form of the velocity function in space and time allows the integration and differentiation with respect to these variables. Functions of displacement and acceleration determined in this way, however, still depend on the nodal velocities. Thus, as the result we have both inertia and stiffness matrices multiplied by velocities. The integration of the velocity function with respect to time results in the displacement function and contains the term with initial displacements in time layer. Initial displacements in each time step of time stepping scheme in the final equation of force equilibrium express nodal forces at the beginning of time interval. According to the equations of motion (
1) calculated acceleration and displacement functions allow the analytical determination of the energy of the system. The energy of the external forces is derived from the right hand side of the Eq. (
1). Classical energy minimization and assembly of the global system leads to the following matrix solution scheme
$$\begin{aligned} \left( \mathbf{M}_g+\mathbf{K}_g\right) \left\{ \begin{array}{l} \mathbf{v}_i\\ \mathbf{v}_{i+1} \end{array} \right\} + \mathbf{E}_g\,\mathbf{w}_i=\mathbf{F}_g\ . \end{aligned}$$
(5)
Here,
\(i\) and
\(i+1\) denote the known and calculated state, respectively. The problem is reduced to the numerical solution of the system of algebraic Eq. (
5). The vector
\(\mathbf{v}\) contains the velocity of the nodal displacements and angles of rotation, and the vector
\(\mathbf{w}\) contains the nodal displacements and the angles of rotation.
\(\mathbf{M}_g\),
\(\mathbf{K}_g\), and
\(\mathbf{E}_g\) are the global matrices of inertia, stiffness, and nodal forces.
\(\mathbf{F}_g\) is the global vector of external forces. The global matrices are assembled from the local matrices
\(\mathbf{M}\),
\(\mathbf{K}\), and
\(\mathbf{E}\), which will be derived further. These elemental matrices are merged in appropriate locations of the global matrices, based on the topology of the mesh. Similarly, the global load vector is assembled from elemental force vectors
\(\mathbf{F}\). This vector can assume zero values or values describing the distributed external load in finite elements. According to the current position of the moving load, the vector
\(\mathbf{F}_g\) varies in each time step. The vector
\(\mathbf{F}_g\) has zero components at the beginning and in each successive time step gains new non-zero values contributed by vectors
\(\mathbf{F}\). The current vector of displacements and rotation angles can be computed using
$$\begin{aligned} \mathbf{w}_{i+1}=\mathbf{w}_i+h\mathbf{v}_{i+1}\ . \end{aligned}$$
(6)
The partition of the space–time area into elements of simplex shape allows us to obtain the stationary solution numerically. In order to demonstrate the properties of the method, let us consider a single simplex finite element as depicted in Fig.
3a. In this case, two space–time subdomains
\(\Omega _A\!=\!\{(x,t)\!\!: 0\le x\le b,\;\; hx/b\le t\le h\}\) and
\(\Omega _B\!=\!\{(x,t)\!\!: 0\le x\le b,\;\; 0\le t\le hx/b\}\) are defined. We assume a linear distribution of the velocity displacements
\(v=\dot{w}\) and the velocity rotation angles
\(\psi =\dot{\theta }\) inside the triangular element:
$$\begin{aligned} \begin{array}{l} v=\dot{w}(x,t)=a_1x+a_2t+a_3,\\ \psi =\dot{\theta }(x,t)=a_1x+a_2t+a_3. \end{array} \end{aligned}$$
(7)
In the finite element shown in Fig.
3a, the interpolation of nodal velocities can be written
$$\begin{aligned} \begin{array}{l} \displaystyle v_A(x,t)=\left( 1-t/h\right) v_1+\left( t/h-x/b\right) v_3+\left( x/b\right) v_4,\\ \displaystyle v_B(x,t)=\left( 1-x/b\right) v_1+\left( x/b-t/h\right) v_2+\left( t/h\right) v_4\,. \end{array} \end{aligned}$$
(8)
The same shape functions were assumed for the velocities of the nodal rotations. Below, calculations for part A of the space–time finite element are presented. We must emphasize here that in the case of part B, the procedure will be totally analogous. Nodal displacements and rotation angles can be written as the integrals of the velocities:
$$\begin{aligned} \begin{array}{l} w_A(x,t)=w_{A0}(x)+\int v_A(x,t)\,\text{ d }t,\\ \theta _A(x,t)=\theta _{A0}(x)+\int \psi _A(x,t)\,\text{ d }t\,. \end{array} \end{aligned}$$
(9)
In order to determine the virtual energy of the problem, we multiply the equations of motion (
1), by the virtual functions
\(v_A^*\) and
\(\psi _A^*\) and integrate the resulting power over the space–time domain
\(\Omega _A\). We consider a distributed load and assume that the load is moved in each time step from one finite element to another, without intermediate steps. The Heaviside step function describing the external load is replaced by a constant function. Thus, virtual energy of the part A of the space–time is given by the following form
$$\begin{aligned} \begin{array}{l} \rho A\int _{\Omega _A} v_A^*\dot{v}_A\,\text{ d }\Omega _A-kGA\int _{\Omega _A} v_A^*\left( w_A^{\prime \prime }-\theta _A^{\prime }\right) \text{ d }\Omega _A\\ \quad +c\int _{\Omega _A} v_A^*w_A\,\text{ d }\Omega _A=P_0\int _{\Omega _A} v_A^*\text{ d }\Omega _A,\\ \rho I\int _{\Omega _A}\psi _A^*\dot{\psi }_A\,\text{ d }\Omega _A-EI\int _{\Omega _A}\psi _A^*\theta _A^{\prime \prime }\text{ d }\Omega _A\\ \quad -kGA\int _{\Omega _A}\psi _A^*\left( w_A^{\prime }-\theta _A\right) \text{ d }\Omega _A=0\,. \end{array} \end{aligned}$$
(10)
The proper choice of virtual functions is a fundamental question of the space–time approach. Various functions in time result in solution schemes of different accuracy and stability. Virtual function review in the case of multiplex shaped space–time finite element approach is presented in [
6]. In this case the virtual linear shape function is assumed. After integration by parts, we have
$$\begin{aligned} \begin{array}{l} \rho A\int _{\Omega _A} v_A^*\dot{v}_A\,\text{ d }\Omega _A+kGA\int _{\Omega _A}\left( {v_A^*}^\prime w_A^\prime +v_A^*\theta _A^\prime \right) \text{ d }\Omega _A\\ \quad +c\int _{\Omega _A} v_A^*w_A\,\text{ d }\Omega _A=P_0\int _{\Omega _A} v_A^*\text{ d }\Omega _A,\\ \rho I\int _{\Omega _A}\psi _A^*\dot{\psi }_A\,\text{ d }\Omega _A+EI\int _{\Omega _A}{\psi _A^*}^\prime \theta _A^\prime \text{ d }\Omega _A\\ \quad -kGA\int _{\Omega _A}\psi _A^*\left( w_A^\prime -\theta _A\right) \text{ d }\Omega _A=0\,. \end{array} \end{aligned}$$
(11)
The classical energy minimization of (
11) leads us to the matrices
\(\mathbf{M}_A\),
\(\mathbf{K}_A\), and
\(\mathbf{E}_A\) for part A of the simplex element. We do the same for part B. After aggregating the matrices of parts A and B, we get the local matrices of inertia, stiffness, and nodal forces:
$$\begin{aligned} \mathbf{M}= \left. \frac{\rho b}{6h}\left[ \begin{array}{r@{\quad }r@{\quad }r@{\quad }r@{\quad }r@{\quad }r@{\quad }r} -A &{} &{} 0 &{} &{} 0 &{} &{} 0\\ 0 &{} &{} -I &{} &{} 0 &{} &{} 0\\ -A &{} &{} 0 &{} &{} -A &{} &{} 0\\ 0 &{} &{} -I &{} &{} 0 &{} &{} -I \end{array} \right| \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} A &{} 0 &{} 0 &{} 0\\ 0 &{} I &{} 0 &{} 0\\ A &{} 0 &{} A &{} 0\\ 0 &{} I &{} 0 &{} I \end{array} \right] ,\nonumber \\ \end{aligned}$$
(12)
$$\begin{aligned} \mathbf{K}&= \left. \frac{EIh}{3b}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 0 &{} 0 &{} 0 \end{array} \right| \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 1 &{} 0 &{} -1\\ 0 &{} 0 &{} 0 &{} 0\\ 0 &{} -1 &{} 0 &{} 1 \end{array} \right] \nonumber \\&+\left. \frac{cbh}{120}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 9 &{} 0 &{} 0 &{} 0\\ 0 &{} 0 &{} 0 &{} 0\\ 11 &{} 0 &{} 5 &{} 0\\ 0 &{} 0 &{} 0 &{} 0 \end{array} \right| \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 2 &{} 0 &{} 4 &{} 0\\ 0 &{} 0 &{} 0 &{} 0\\ -2 &{} 0 &{} 11 &{} 0\\ 0 &{} 0 &{} 0 &{} 0 \end{array} \right] \nonumber \\&+\left. \frac{kGAh}{384}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 23b &{} 0 &{} 0\\ 0 &{} -16 &{} 0 &{} 16\\ 16 &{} 52b &{} -16 &{} 5b \end{array} \right| \right. \nonumber \\&\left. \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 144/b &{} -40 &{} -144/b &{} 40\\ 40 &{} -3b &{} -40 &{} 20b\\ -144/b &{} -72 &{} 144/b &{} 72\\ 72 &{} -8b &{} -72 &{} 39b \end{array} \right] , \end{aligned}$$
(13)
$$\begin{aligned} \mathbf{E}&= \frac{EI}{2b}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 0 &{} 0 &{} 0 &{} 0\\ 0 &{} 1 &{} 0 &{} -1\\ 0 &{} 0 &{} 0 &{} 0\\ 0 &{} -1 &{} 0 &{} 1 \end{array} \right] \frac{cb}{24}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 3 &{} 0 &{} 1 &{} 0\\ 0 &{} 0 &{} 0 &{} 0\\ 3 &{} 0 &{} 5 &{} 0\\ 0 &{} 0 &{} 0 &{} 0 \end{array} \right] \nonumber \\&+\frac{kGA}{16}\left[ \begin{array}{l@{\quad }l@{\quad }l@{\quad }l} 8/b &{} -2 &{} -8/b &{} 2\\ 2 &{} b &{} -2 &{} b\\ -8/b &{} -6 &{} 8/b &{} 6\\ 6 &{} 3b &{} -6 &{} 3b \end{array} \right] \!. \end{aligned}$$
(14)
The stiffness matrix
\(\mathbf{K}\) (
13) and nodal forces matrix
\(\mathbf{E}\) (
14) can be split into three parts: a part related to bending, a part related to foundation, and a part related to shear. Assuming the linear shape function (
7) in virtual energy (
11), we indicate that the bending strain component is constant and the shear strain component varies linearly. In this case the exact integration of the shear strain components in virtual energy (
11) results in the element which is too stiff. This over-stiffness of the element is known and is called locking. The shear components in relation with the bending component are high. A reduced integration of the shear part is a classical literature remedy. Bending terms are integrated exactly while shear terms are integrated with only one point of the Gauss quadrature. More details can be found in [
24,
26], e. g. As mentioned previously, the matrices (
12), (
13), and (
14) were assembled into the global matrices
\(\mathbf{M}_g\),
\(\mathbf{K}_g\), and
\(\mathbf{E}_g\).