Getting layout right (even if only a structure is considered) with HTML5/CSS3 is still more like an art or black magic.

On the other hand, there are other GUI systems (like wxWindows and Tcl/Tk) and some GUI research (like The Auckland Layout Model, ALM, and [other methods][1]), which hint at the possibility of formalization for the layout managers (geometry managers).

Are there any comprehensible formal models for HTML5/CSS, which provide ultracompact (abstract) way to describe structure, "physics" and "geometry" of resizeable webpages, using language of blocks? Also html/css can be generated from it, which works more or less as described in standard browsers. Also, a model can be derived given HTML/CSS (browsers do it by their algorithms, so this seems to be theoretically possible).

By "ultracompact" and abstract it is understood: much more compact than HTML/CSS and also more domain-oriented, "speaking" the language of webpage's dynamics in response to resizing or changed content, that is, higher level than HTML/CSS constructs.

For an analogy, it is possible to write a program to make a textual search, based on some complex rules, but the same task can be performed by a much more compact regular expression. So, is there similar compact language for HTML/CSS layout?

The goals of such a model could be:

 - to verify existing design (model checking)
 - to build robust design given higher level specifications
 - to check whether a set of requirements is consistent with HTML5/CSS3 engine (e.g., does not require writing javascript to make an adjustment too complex for the declarative languages)
 - to be a solid platform for even higher level research on qualities ("to check the harmony with algebra.")

Also it could be a language to use for certain GUI-related abstractions, like is usual in programming language domain, where we do not need to use concrete syntax to express an idea of for-loop and we do have all kinds of nice, proven results about main concepts of algorithmic constructions.

Of course, web-browsers possess algorithmic model for rendering, e.g. popular and simplified description can be found [here][2], but as pointed above it does not have the properties listed above.