paper: Reword third paragraph and add citations to mpi4py, Cython, and FFTW

Author: dalcinl

docs/paper/paper.bib

@@ -1,5 +1,5 @@
 @article{dalcin18,
-  author    = {Lisandro Dalc{\'{\i}}n and
+  author    = {Lisandro Dalcin and
                Mikael Mortensen and
                David E. Keyes},
   title     = {Fast parallel multidimensional {FFT} using advanced {MPI}},
@@ -44,3 +44,40 @@ @article{strang94
  volume = {82},
  year = {1994}
 }
+
+@article{mpi4py08,
+author = {Lisandro Dalcin and Rodrigo Paz and Mario Storti and Jorge D'Elia},
+title = {{MPI} for {P}ython: Performance improvements and {MPI}-2 extensions},
+journal = {Journal of Parallel and Distributed Computing},
+volume = {68},
+number = {5},
+pages = {655--662},
+year = {2008},
+issn = {0743-7315},
+url = {https://doi.org/10.1016/j.jpdc.2007.09.005},
+doi = {10.1016/j.jpdc.2007.09.005},
+}
+
+@article{cython11,
+ author = {Behnel, Stefan and Bradshaw, Robert and Citro, Craig and Dalcin, Lisandro and Seljebotn, Dag Sverre and Smith, Kurt},
+ title = {Cython: The Best of Both Worlds},
+ journal = {Computing in Science and Engg.},
+ volume = {13},
+ number = {2},
+ pages = {31--39},
+ year = {2011},
+ issn = {1521-9615},
+ url = {https://doi.org/10.1109/MCSE.2010.118},
+ doi = {10.1109/MCSE.2010.118},
+}
+
+@article{fftw05,
+  author = 	 {Frigo, Matteo and Johnson, Steven~G.},
+  title = 	 {The Design and Implementation of {FFTW3}},
+  journal = 	 {Proceedings of the IEEE},
+  year = 	 2005,
+  volume =	 93,
+  number =	 2,
+  pages =	 {216--231},
+  note =	 {Special issue on ``Program Generation, Optimization, and Platform Adaptation''}
+}

docs/paper/paper.md

@@ -42,15 +42,17 @@ successive sequential (serial) transforms over non-distributed axes,
 combined with global array redistributions (using interprocess
 communication) that realign the arrays for further serial transforms.
 
-For global redistribution ``mpi4py-fft`` makes use of a new and completely
-generic algorithm [@dalcin18] that allows for any index sets of a
-multidimensional array to be distributed. We can distribute just one index
-(a slab decomposition), two index sets (pencil decomposition) or even more for
-higher-dimensional arrays. The required MPI communications are always handled
-under the hood by MPI for Python. For serial transforms
-``mpi4py-fft`` wraps most of the FFTW library using Cython, making it callable
-from Python. We include wrappers for complex-to-complex, real-to-complex,
-complex-to-real and real-to-real transforms.
+For global redistributions, ``mpi4py-fft`` makes use of a new and
+completely generic algorithm [@dalcin18] based on advanced MPI
+features that allows for any index sets of a multidimensional array to
+be distributed. It can distribute a single index set (slab
+decomposition), two index sets (pencil decomposition), or even more
+for higher-dimensional arrays. The required MPI communications are
+always handled under the hood by MPI for Python [@mpi4py08]. For
+serial FFT transforms, ``mpi4py-fft`` uses Cython [@cython11] to wrap
+most of the FFTW library [@fftw05] and provide support for
+complex-to-complex, real-to-complex, complex-to-real and real-to-real
+transforms.
 
 ``mpi4py-fft`` is highly configurable in how it distributes and redistributes
 arrays. Large arrays may be globally redistributed for alignement