Obstacles to collaborative research between experimentalists, theorists, and computational researchers

I'm a computational chemist. I've worked with many experimental collaborators with wildly varying experiences. Barriers certainly can and most definitely exist. Here are some of the major ones.

1. Lack of understanding: This is perhaps the most common and basic reason and results from a lack of experience or knowledge in experimental techniques by computationalists and vice versa, quite simply because they were not trained in this area. As a computationalist, it is natural that I may misunderstand how certain experimental techniques work (especially the state of the art) because I do not spend most of my time working on it. This can result in me over-trusting a piece of experimental evidence, or underestimating the time and effort it can take to conduct the experiments. And believe me, this can definitely happen the other way round. One example, an experimental collaborator asked for some 'quick calculations' in a month and I had to point out that it would easily take a year and all of our computational budget to complete it.
2. Politics: In a collaboration, there is always the question of who is the major contributor. Yes, it is possible to have multiple corresponding and first authors, but it rarely changes this fact. In my field (chemistry), usually the experimental side is the major participant of a collaborative work. This means a computational PI will benefit significantly less, and his student will usually not be a first author. This generally leads to conflicts regarding contribution of time and resources given the unequal recognition of work. Sometimes, in order to solve this issue, particularly for long-term collaborations, the experimentalist and computationalist can 'take turns' directing the project and claiming the major contribution to a paper. (I am not supporting or criticizing this phenomenon, but it definitely exists in the field.)
3. Geographic/temporal reasons: Quite simply, many if not most collaborations occur between groups in different universities and different countries. The simple fact is you cannot communicate them like you would with your group members in the same room. This results in the need to schedule meetings and email communications which are often difficult (professors are busy!) and are invariably an inefficient way of transferring information. I've has cases where several weeks or even months of work were invalidated/or made irrelevant because I was only informed of new experimental findings many months later in a meeting.
4. Different objectives/expectations: There is some overlap with the political reason, but sometimes it can be purely due to academic reasons. It can be common for experimental and computational groups to have different expectations for what knowledge/results a work is supposed to yield. For example, a study can be trivially easy for a computationally group (and possibly uninteresting) to accomplish (via modelling), but can be extremely difficult for an experimental group to synthesize/characterize. Alternatively, a relatively simple reaction for an experimental group to conduct can be a multi-year effort by several researchers (or it can even be unsolvable/impractical by current techniques!).

Is it possible to overcome these barriers? Certainly. I've had collaborators who are well aware of these issues (from experience), and end up being very enjoyable to work with. As an aspiring researcher, the best thing you can do is keep these issues in mind when collaborating with other groups. If possible, try to learn the techniques which your collaborators are currently using, and maintain a steady stream of communication with whoever your direct counterpart is (usually, if you are a student, keep in contact with the corresponding student in the other group instead of the professor directly - though you can/should CC him and your advisor to keep them in the loop).

Edit: I decided to add some examples of hypothetical collaborators from the perspective of a computational researcher (for fun). Any resemblance to actual persons, living or dead, or actual events is purely coincidental.

The Good: The well-informed and reasonable collaborator. Knows their stuff (even on your side), up to date on literature, knows what is practical and not. Knows if you are BSing them or not. Won't do you wrong if you do your job. Keep them well-fed with updates. Expect to do good science with them.

The Bad: The villain/antagonist. Only works with you because you are part of the same grant or project, and only sees you as a way of getting higher impact factor. Only interested in positive results, gives unreasonable deadlines, does not like to hear the phrase "but that can't be done!' Never satisfied, keep him well-fed, but for your own survival.

The Ugly: Nice guy, easy to work with. Okay with whatever you have. Takes anything you offer, but not terribly interested in the science behind the results. Likes you because you can help him get higher impact factor journals. Expect him to disappear one day when he has found someone better or no longer needs your help.

The main source of the "friction" in talks among computational, theoretical, and experimental researchers is language. One group may use the term very differently from another.

For instance, I can say that I "calculate" a given quantity using a simulation. To me, this implies that I have performed a simulation, and know that the results that I get have a certain amount of inherent uncertainty because of the intrinsic variability of the particles I'm studying and that a slightly different starting point will lead to a very different result. However, all of that nuance may be lost on a theoretician or experimentalist who hears "calculate" and thinks I've just crunched some numbers and have a "one size fits all" answer to the problem.

However, the problem is not merely among the different groups, but also within them. Even different computational scientists can have difficulty understanding one another without some "translation." For instance, I can remember my two computational advisors arguing over a point for a while before we all realized that they were talking about exactly the same thing. The only difference was that one was using the "language" of control systems, and the other was using the "language" of materials science to describe the same phenomenon.

I'm a computational epidemiologist, and so work fairly heavy with both "experimentalists" in the form of either observational epidemiology or clinical trials, computational folks, and more theoretical mathematical biology researchers.

Generally, I've found this collaborative research to be quite successful, but there are some difficulties:

1. Difficulty understanding scope. Often, I find clinical colleagues end up wanting to add lots of detail to computational models that vastly increases the difficulty in implementation without really understanding what that means. Equally, my more theoretical colleagues don't have a good grasp of the nuances of things like IRBs, how long data takes to collect, etc. and vastly underestimate how hard and expensive it is to actually collect data.
2. Competing pressures and priorities. Is 15% effort on a grant a ton of support, or barely covering your time? What's needed for tenure and promotion? What's a "good journal"? Do the theory people get out a bunch of papers early, while the experimentalists sit on their hands waiting for data.
3. Difficulty understanding what's interesting. For example, there may be an intensely important clinical questions on really unique data sets that actually use pretty mundane methods from a computational/theoretical perspective.
4. Different tendencies in different fields. Do you use LaTeX? How familiar is everyone with programming? Are papers the currency of choice, or presentations? What even is publishable?
5. Different Incentive Structures. Are the theoreticians penalized for having papers with multiple co-authors? Are you evaluated based on student outcomes on grants, etc.?
6. Timing and communicating. "What data do you need?" "I don't know, what data do you have?" is probably the most common conversation I end up having.