paper.md (draft): simplify the statement on performance

paper/paper.md

@@ -35,9 +35,7 @@ numerical solution of nonlinear problems in continuum mechanics of solid bodies.
 # Statement of need
 There are well-established Python packages available for finite element analysis. These packages are either distributed as binary packages or need to be compiled on installation, like FEniCSx [@fenicsx], GetFEM [@getfem] or SfePy [@sfepy]. JAX-FEM [@jaxfem], which is built on JAX [@jax], is a pure Python package but requires many dependencies in its recommended environment. `scikit-fem` [@scikitfem] is a pure Python package with minimal dependencies but with a more general scope [@scikitfem]. FElupe is both easy-to-install as well as easy-to-use in its target domain of hyperelastic solid bodies.
 
-The performance of FElupe is good for a non-compiled package but mediocre in comparison to compiled codes. However, it is still well-suited for up to mid-sized problems, i.e. up to $10^5$ degrees of freedom, when basic hyperelastic model formulations are used. A performance benchmark for times spent on stiffness matrix assembly is included in the documentation. Internally, efficient NumPy [@numpy] based math is realized by element-wise operating trailing axes [@scikitfem]. An all-at-once approach per operation is used instead of a cell-by-cell evaluation loop. The constitutive material formulation class is backend agnostic: FElupe provides NumPy-arrays as input arguments and requires NumPy-arrays as return values. This enables backends like JAX [@jax] or PyTorch [@pytorch] to be used. Interactive views of meshes, fields and solid bodies are enabled by PyVista [@pyvista]. The capabilities of FElupe may be enhanced with additional Python packages, e.g. `meshio` [@meshio], `matadi` [@matadi], `tensortrax` [@tensortrax], `hyperelastic` [@hyperelastic] or `feplot` [@feplot].
-
-
+The performance of FElupe is good for a non-compiled package and it is well-suited for up to mid-sized problems, i.e. up to $10^5$ degrees of freedom, when hyperelastic model formulations are used. A performance benchmark for times spent on stiffness matrix assembly is included in the documentation. Internally, efficient NumPy [@numpy] based math is realized by element-wise operating trailing axes [@scikitfem]. An all-at-once approach per operation is used instead of a cell-by-cell evaluation loop. The constitutive material formulation class is backend agnostic: FElupe provides NumPy-arrays as input arguments and requires NumPy-arrays as return values. This enables backends like JAX [@jax] or PyTorch [@pytorch] to be used. Interactive views of meshes, fields and solid bodies are enabled by PyVista [@pyvista]. The capabilities of FElupe may be enhanced with additional Python packages, e.g. `meshio` [@meshio], `matadi` [@matadi], `tensortrax` [@tensortrax], `hyperelastic` [@hyperelastic] or `feplot` [@feplot].
 
 # Features
 The essential high-level parts of solving problems with FElupe include a field, a solid body, boundary conditions and a job. With the combination of a mesh, a finite element formulation and a quadrature rule, a numeric region is created. A field for a field container is further created on top of this numeric region, see \autoref{fig:field}.