Iannis Xenakis was one of the great composers, along with Stockhausen, Boulez, and the rest of the European avant-garde, when I was first learning about new music back in the 1970s. Dutifully I bought albums of his music and his piano scores. But I was most impressed with his electronic music, not surprising considering where my tastes turned out. There was a Nonesuch LP of his Electro-Acoustic [sic] music that I played until it wore out. The music on the LP has been reissued on CD, and the music is as impressive today as when it was originally released in 1972. His later electronic pieces, Persepolis and La Légende d'Eer, were equally imposing and remain among my favorite musique concrète.
It took several other purchases to make a disappointing discovery, that I didn't really take to his instrumental music as much as his electronics. Even though I picked up a half dozen or so CDs over the years, none of his instrumental music has arrived in my iTunes, which basically means that I haven't listened to it in several years. While I admire and respect his work tremendously, the last Xenakis CD I bought before the one under discussion here was five years ago, one of a set of works for large orchestra conducted by Arturo Tamayo.
What prompted me to buy the recent Neos CD of Xenakis' keyboard works was that instead of being performed in the usual manner, the five pieces on the CD (limiting my discussion here to the three piano works, but there are also two for harpsichord) were MIDI realizations by the conductor Daniel Grossmann. Now, I've done my own MIDI realizations of piano compositions. I posted one by Dennis Riley here a while back, so I have no inherent bias against the technology. There's even considerable history in recording using MIDI piano. Herbert Henck participated in Klarenz Barlow's album of his piece Çoğluotobüsişletmesi (unfortunately not reissued on CD), where Henck performed the piece on one side, and a computer realization was on the other. More recently, Carolyn Yarnell released a MIDI realization of her beautiful piece The Same Sky on Tzadik, while pianist Kathleen Supové released the same piece played on piano.
Certainly Xenakis's piece are sufficiently complex to warrant a computer realization. The score for Herma contains notations for three different pitch classes, which he uses for set operations (intersection, union, negation), as well as specific dynamic marks ranging from ppp to fff (but not including p, mp, and only one mf), often combining very loud and very soft notes simultaneously. Evryali creates textures from simultaneous but apparently separate lines, often requiring four staves (and even five, in one passage) to clarify the relationships to the pianist. And Mists, which ranks among the largest piano scores simply in terms of physical size, combines Herma's pitch classes with complex rhythmic hemiolas and cryptic pitch class notation. British composer Brian Ferneyhough writes extremely difficult music because he wants the performer's struggle to be audible in the performance, and Xenakis is sometimes in a comparable position. (I haven't seen any computer realizations of Ferneyhough's music, which might be more appropriate than Xenakis. But then, Ferneyhough is still around and might choose to prohibit such a development.) And yet, there are several recordings of live pianists playing all three of these pieces, which suggests that as difficult as they may be, they aren't completely impossible to play.
The instinctive reaction to Grossmann's album is that he's cheating, that a professional pianist ought to be able to perform Xenakis's piano music without having to resort to software solutions. Grossmann's response is in the liner notes (actually written by composer and Xenakis specialist Tom Sora), in an essay entitled "Complex — Unplayable — Authoritative". I won't go into the details, but Sora gives several rationales supporting the claim staked in his title. (You can find the essay online at Neos.) The question is whether the music is as convincing as other recordings.
One of the major differences between Grossmann's recording and real pianos is the amount of resonance in the sound clouds, where the pianist plays lots of notes, both very loud and very soft, with the sustaining pedal down. In Grossmann's recording, you hear every note, i.e., the resonance is very subdued. In Claude Helffer's recording for Montaigne, the cloud is much more cloudy, with the loud notes poking through the murk and the soft notes merely contributing color. The liner notes refer to one passage (marked "+A nuage 5 s/s" in the score, around a minute into the piece) as requiring staccato notes "to sound simultaneously with another layer containing notes that are to form a compact cloud by means of the pedal," but my score doesn't have any such staccato markings.*
Also, using the computer, Grossmann is able to use stereo effects that a single piano cannot. He uses panning in some of the more contrapuntal sections to help distinguish the various lines, such as in the arborescent sections of Mists. Even though the panning effects are most audible with headphone listening, it contributes to the same ends as Grossmann's resonance decisions. Essentially, Grossmann has opted to emphasize Xenakis' pitch relations instead of the more textural ones that Helffer achieves. This is not necessarily due to the nature of a MIDI realization. His software (unfortunately not credited, but perhaps that's just the curiosity of a former geek) probably has a setting for the amount of sustain resonance.
There is no question that Grossmann's realization plays faster and more precisely than a human, especially in Evryali, which is in fact impossible to play as notated (it includes a high C# which a half step higher than the notes actually present on hardware pianos, although it sounds like Grossmann might have found one). Examples in Evryali abound: the tremolos near the beginning, which are played in canonic imitation a few beats removed from one another, even the overall speed (which is 20% faster than Helffer's). It's one of the few places in the three pieces where you know it's a machine. But the passage is so fast and so precise that it calls attention to itself, look at me, I'm a computer! In Helffer's recording, he fudges over the passage a little bit, creating a more coloristic effect, and one would never know there was canonic imitation. Evryali comes closest to Ferneyhough's ideals, and is for that reason perhaps the best justification for a computer realization.
Each potential listener will come to grips with a MIDI realization of these works. Purists will doubtless avoid the CD altogether since it goes so much against the current performance traditions of classical music. One could argue (as Sora does) whether it best represents Xenakis' intentions. Surely Xenakis knew the range of the instruments for which he wrote, so the difficulties (even the impossibilities) had some other motivation. I've mentioned before my belief that much contemporary music is written for the performer rather than the audience, and this is as true for Xenakis' piano music as for any of the open works mentioned in that post. Grossmann spent years creating his realizations, as a human performer would have to spend a comparable amount of time to solve the difficulties in the score for him or herself. I don't believe that Grossmann's recording is any closer to "what Xenakis intended" than Helffer's or any other performance, but his approach has its unique perspectives on a fascinating body of work.
*A footnote in James Harley's Xenakis monograph alludes to some controversial corrections to the score of Herma, which may account for the difference. In any event, these "corrections" are not widely disseminated.