Interesting how polarizing and off-topic pulling such discussion is. And also interesting how much inaccurate or old information continues to be perpetuated.
For example the fact that Pianoteq uses sampled action noise. Of course the software is closed source so we know only what the vendor says, so let's assume they "selectively disclose" but without misleading or inaccurate/false statements. They did say in the past that action noises were sampled. They said later (IIRC in version 6 or 7) that this is not anymore the case and all noises are now "modeled". They did not disclose how.
They also very clearly disclose the "big picture" of how it works, which is definitely some sort of wave synthesis, see https://www.modartt.com/pianoteq_features#tutorial (not sure why people keep speculating about this given how much the company themselves discloses).
Another topic of high speculation is that doing a full-fledged PDE simulation of the instrument would be the holy grail (if we only had fast enough computers). Since this is really close to my professional expertise, I'd take the freedom to correct a few things, even if I dislike doing so because I see it as wildly off-topic. PDEs might be more accurate, but they might even be complete crap. I suspect that the condition number for a full model of the piano would be a monstrosity, which would make the system unsolvable. Integral formulation (rather than PDE) might ameliorate the problem, but that's not guaranteed. In short: it's possible that the "real" physical model would be even worse than what we have now.
And even if it's solvable, it's not a given that the fine details which make a "good" piano "good" would be captured by such a model: it's known by many acoustic piano people that some grand pianos sound like crap and there is nothing that can be done about it.
Regarding quantum computers, it's true that they would be good only for some kind of computing, and definitely not just a generic "everything is faster" ones. Yet, solving linear systems (such as the ones that arise from the discretization of PDEs or of the integral formulations) are a promising application to gain a speed up. Promising means "there are founded hopes that it'll be faster", but it's not a sure thing. On the other hand, latency would just kill it for this application, making any solving speed improvement a moot point for real time playing. Yet, by the time we'd have something decent of that size, people might have reduced latency dramatically, but as far as I know nobody in the world is even thinking about it. Still (I don't want to sound like the IBM's guy, but here I am), I can't see any reason why somebody would want a quantum computer in their living room. As far as we can reasonably extrapolate even the most optimistic path, we can't even speculate how we can have a quantum computer in our pocket (as we now have with the classical computer most people call "phone")
But regardless, I find it extremely interesting how many people hear things differently and most (not all obviously) assume everyone else hears like them and therefore must be an idiot by liking something that to them sounds like crap falling into the water if you allow me to use this metaphor 😃
I think I am one of the few "in between" people, who hear the shortcomings of both sampling and modeling (and there definitely are others like me, with whom I've discussed, mostly privately). Some of them are even on the Modartt forum!
I've reached the following conclusion, which I think is best expressed by a computer graphic example. Sampling is like very high resolution "flat" images, such as the ones for google maps. On my phone there is a 3D option which pretends to show them in 3D but the effect is pathetic to my eyes: the facades of the buildings are not recognized as such and even the buildings themselves have no 3D appearance at all (in most locations -- there are some exceptions in famous big cities). Yet, if I stick to the 2D view, it's quite convincing and I may think to be in an airplane looking down (but with no option to change course and see it from a different perspective). On the other hand, Pianoteq is more like the early ray-tracing programs: it gives a very detailed 3D appearance, with the appropriate volumetric (and even stereoscopic) appearance, let me say of a room with objects on a table. I can move my position at will and have the view change perfectly, put my eyes exactly where I want and so on. Yet, the surfaces are very crude: yes, there are textures which can be slapped on surfaces, but that wooden table looks more like a piece of plastic than real wood (note, this is just an example, I am not claiming that sampling is "2D flat" and pianoteq is "3D stereo", just that they are able to do better different things).
So I can see how some people like more of one approach or the other. Unlike images, I can't point my finger on "look, these are incorrect pixels". When I tried (with Yammy/Laurel) some people even though I was wasting their time. Heck, we don't even have a shared vocabulary to describe things. But I can definitely hear the limitations of both approaches, and to me NEITHER thing sound ANYWHERE close enough to a real acoustic piano. And I mean "of a recording of a real acoustic piano played by the same speakers", of course a speaker is different from the soundboard, but that is a DIFFERENT issue to me.
The real problem, in my opinion, is that people focus on one aspect only (either unconsciously or by physiological differences in their auditive path) and therefore this unfortunately situation arises, then they assume everyone should be like them and if they aren't then they must be stupid.... I see this happening in all fields of life, so it's not surprising it's happening in piano too, but I digress.... better go back living under my rock than wasting time writing messages like this.... why did I do it?
