On steady-state preserving spectral methods for homogeneous Boltzmann equations (Q2018267)

It is considered a simplified version of a Boltzmann type equation \[ \frac{\partial f}{\partial t}=\mathcal{{Q}}(f,f) \] where \(f(t,x,v),\) \(x\in \mathbb{R}^d,\) \(v\in \mathbb{R}^d\) (\(d>1\)), \[ \mathcal{{Q}}(f,f)(v)=\int_{\mathbb{R}^d}\int_{\mathbb{S}^{d-1}}B\biggm(\frac{v-v'}{|v-v_*|}\omega, |v-v'|\biggm)[f(v'_*)f(v')-f(v_*)f(v)]d\omega dv_*, \] \(\displaystyle v'=\frac{v+v_*}{2}+\frac{|v-v_*|}{2}\sigma,\) \(\displaystyle v'_*=\frac{v+v_*}{2}-\frac{|v-v_*|}{2}\sigma.\) Using the decomposition \(f=M+g\) with \(M\) the local Maxwellian equilibrium and orthogonal projection \(\mathcal{{P}}_N:L^2([-\pi,\pi]^3)\rightarrow \mathbb{P}^N=span\{e^{ik\cdot v}| -~N\leqslant k_j\leqslant N, j=1,\dots,d\}\) such that \(\langle f-\mathcal{{P}}_N f,\phi\rangle_{L^2}=0\) \(\forall \phi\in \mathbb{P}^N\) new problem is made: \[ \begin{cases} \displaystyle\frac{\partial g_N}{\partial t}=\mathcal{L}^\lambda_N(M_N,g_N)+\mathcal{Q}^\lambda_N(g_N,g_N),\\ f_N=M_N+g_N \end{cases} \] where \(M_N=\mathcal{{P}} M,\) \(g_N=\mathcal{{P}} g,\) \(\mathcal{L}^\lambda_N(M_N,g_N)=\mathcal{{P}} \mathcal{L}^\lambda(M_N,g_N)\) and \(\mathcal{Q}^\lambda_N(g_N,g_N)_N=\mathcal{{P}} \mathcal{Q}^\lambda(g_N,g_N).\) This new method achieves a nice monotonous decay of the relative entropy, without the large time increase of the classical spectral method. The behavior of the \(L^2\) error is also monotonous for the steady-state preserving method, which is not the case for the classical one. And it is noticed that even if the positivity of the solution is not preserved by the new method, the number of nonpositive cells seems to be lower at a fixed time for the steady-state-preserving method, compared to the classical one.