{Note: links to articles in the scientific literature are behind paywalls; they should be able to be accessed on campus, or offcampus using the UD Proxify button that can be installed on your browser.}

The VSEPR and Valence Bond theory that students are taught in introductory chemistry courses are simplistic models that are often good enough to explain or predict “day to day” chemistry.  However, sometimes their prediction is inaccurate (e.g an H₂S bond angle closer to 90°), and sometimes flat-out wrong (e.g. H₂O having two same-energy lone pairs on the oxygen, in sp³ hybrid orbitals). 

In this review on orbitals in phosphorous compounds, the author makes the following statement regarding the invocation of d orbitals in bonding schemes:

There are those issues which are settled, those which are in an advanced state of understanding but which remain to be confirmed and those issues which are still wide open.

In the first category we find mostly discarded concepts {such as} the involvement of d orbitals in main group bonding.

The paper is rather advanced (I have only perused it and can’t claim to fully understand the details) but may be of interest to a curious student.  If nothing else, it shows examples of competing models of chemical bonding and how models can be judged based on how well they conform to “real world” data.

In this paper (provocatively entitled “Chemical Bonding to Hypercoordinate Second-Row Atoms: d

Orbital Participation versus Democracy”), the authors preface their work with this statement:

Although there remain some differences in the emphasis placed on  the role played by d basis functions, the consensus view which emerges from most of the reliable ab initio  investigations published in recent years is that the bonding in a  molecule such as SF₆ has very little to do with the availability of d atomic orbitals. Nevertheless, the existence of PF₅, but not of NF₅, is still often rationalized to high school students, and to many undergraduates, in terms of the availability of d orbitals and the possibility of obtaining “an expanded octet”. Indeed, models based on  d²sp³, dsp², and dsp³ hybrid orbitals are still in widespread use amongst professional chemists and are described in many of the most widely used textbooks. It is tempting to speculate as to why such models continue to survive when there is so  much theoretical evidence which does not support them.

It’s not just hybrid orbitals that incorporate d- orbitals that are fiction; the whole concept of “hybrid orbitals” is dodgy.

Prof. Cramer from the University of Minnesota, in this solution to a homework involving computation of molecular orbitals for water, puts it more bluntly:

Your Lewis structure, if you drew it correctly, says that there is one pair of electrons in one O–H σ bond, one pair in another identical such σ bond, and two pairs each of which is equivalent that constitute the lone pairs on oxygen. The two lone pairs and the O–H bonds should by pointing towards the apices of a tetrahedron because they are all considered to be sp³ hybridized. Right?

My God, how you've been lied to before now....

As you can see, the MOs look nothing like the Lewis picture. Instead, amongst other details, there is one lone pair that is pure p (not sp³), another that is, if anything, sp²-like, but also enjoys contribution from hydrogen 1s components. There is one orbital that looks like both O–H σ bonds are present, but another that has an odd "bonding-allover" character to it.

However, he goes on to explain that, when the water molecule is taken out of isolation and interacts with other molecules, the tetrahedral-like geometry that you would predict from VSEPR is observed (e.g. in the structure of ice crystals). 

If you look at the structures in that final link, you can see elements of what we talked about in class present in those structures.  MO 1 correlates to the filled 1s orbital of oxygen. Those electrons (n=1) are not valence electrons (n=2 for oxygen) and don’t really contribute to bonding in the molecule.  As I said in class, we can safely ignore non-valence electrons when we want to consider bonding in molecules.  MO 3 looks like a pair of O-H σ bonds. MO 6 and 7 resemble σ* antibonding orbitals.  The key point here is that the orbitals that describe the lone pairs on oxygen (MO 4 and 5) are clearly different and do not resemble a pair of sp³ orbitals (what we colloquially call the “bunny ear” lone pair depiction).  The MO theory predicts that the two oxygen lone pairs are in non-equivalent orbitals, and this correlates with experimental evidence.

The models we teach you to predict or rationalize molecular structure/reactivity are an approximation.  They have a lot of practical utility, but you should appreciate that they are just a model of reality, not reality itself.