Style of mathematical writing vs. too many lemmas I work in PDEs. I have now written 3 papers. I find my style is of the form: introduction, statement of results, paragraphs to introduce something, lemma, more text, lemma, more text, lemma, more text, theorem, concluding remarks (I missed the proofs).
I am getting sick of this type of writing. I want to write my next paper with more style and elegance than something which looks like the output from a workhouse. 
For example I have seen papers where they come up with some interesting result which they don't put into a lemma but instead put it in a paragraph. When I cite such a thing, I say something like "see the paragraph on page 10 of [1]" so I think it is a bad idea to put results outside environments. 
Does anyone have any ideas? Basically I want to write a more unpredictable paper instead of the usual routine which I wish to take myself out of. This is good because readers will also find it more interesting.
 A: The mathematical papers I find most enjoyable (as well as accessible) are those written in order of increasing technicality.  So, consider an order along the lines of Introduction and Conclusion (containing motivation, related work, main ideas, informal statement of results and applications, interesting open problems).  Essential definitions. Formal statement of results.  Applications/corollaries.  Further definitions. Elementary arguments (may be several sections).  Technical estimates (may be several sections).  Of course, the logical dependences require careful cross-referencing. 
A: The only disadvantage of Dieudonne's style is that equation numbers are extremely long. I use this style with at most one point in subsections, theorem, lemmas (like 41.18), with local number insider subsections: (7), which I cite like that inside the subsection and cite like (41.18.7) from outside the subsection. All my books are written like that, and most papers. Unfortunately the Latex numbering does not support local numbers, so I use an extra program (called tex-nmb) for renumbering (written by Andreas Kriegl). 
For a short paper I try to have one theorem in the beginning, one lemma, one corollary, one remark, without any numbers. 
A: Since the OP is looking for an alternative text structure, here's a possibility that I've seen in Dieudonné's multi-volume Eléments d'Analyse (Treatise on Analysis). He does not always put statements of lemmas or theorems, proofs and discussion in separate environments. Instead, breaks the text into numbered Chapters, Sections, and Subsections in the following way. Chapters and Sections have titles, while Subsections are anonymous and referred to only by number, like (23.3.10), with equations even further subnumbered, like (21.4.4.1). The point is that each Subsection is short enough to contain only one main idea, which could be a statement with or without proof, a remark, a definition, some motivational paragraphs, or an example. A definition or a statement could be highlighted in italics, but not in a separate environment.
It seems to me a decent compromise between citability and a more flexible text structure. As long as the Subsections are short enough, with the most important text somehow highlighted, the reader does not have to spend too much time looking for the information that is being referred to.
