About Changes: Sixty-Four Studies for Six Harps

Author(s): James Tenney

Source: Perspectives of New Music, Vol. 25, No. 1/2, 25th Anniversary Issue (Winter -
Summer, 1987), pp. 64-87
Published by: Perspectives of New Music
Stable URL: http://www.jstor.org/stable/833092
Accessed: 25/07/2009 18:51

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=pnm.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

Perspectives of New Music is collaborating with JSTOR to digitize, preserve and extend access to Perspectives
of New Music.

http://www.jstor.org
 ABOUT CHANGES:SIXTY-FOUR
STUDIESFORSIXHARPS

JAMESTENNEY

A. INTRODUCTION

M rY INTENTIONS IN this work were both exploratory and didactic. That is,
I wanted to investigatethe new harmonicresourcesthat have become
availablethrough the concept of "harmonic space" much more thoroughly
than I had in any earlierwork. At the sametime I wantedto explorethese har-
monic resourceswithin a formalcontext which would clearlydemonstratecer-
tain theoretical ideas and compositional methods alreadydeveloped in my
computer music of the early1960s, including the use of stochastic (or con-
strained-random)processes applied to severalholarchicalperceptual levels,
both monophonicallyand polyphonically.The referencesto the I Ching or
"Book of Changes"(in the titlesof the individualStudies)derivefrom correla-
tions which were made partlyfor poetic/philosophicalreasons,but also-and
perhapsmore importantly-as a means of ensuring that all possiblecombina-
 About Changes 65

tions of parametricstates would be included in the work as a whole. I must

confessthat I frequentlythought of the twenty-fourPreludesand Fuguesof J.
S. Bach's WelUrtemed Clavieras a kind of model for what I wanted to do with
the work, althoughit seems highly unlikelythat these Studiesthemselveswill
everbetraythat factto a listener.A largemainframecomputerwas used in the
compositionprocess,generatingcoded numericaloutput which was then tran-
scribedinto standardmusicalnotation. Two separateFORTRANIV programs
were involved, the firstdealingwith characteristics
of the set of sixty-fourStud-
ies as a whole, the second determiningthe detailsof each individualStudy.

B. GENERALFEATURES

The harpsaretuned a sixth of a semitone (16.66... cents)apart,so the ensem-

ble is capableof producinga temperedmicrotonalset of seventy-twopitchesin
each octave. This tuning system (which I call the 72-set) providesvery good
approximationsof most of the importantjust intervalswithin the 11-limit,with
the worst casebeing the three-centerrorfor the 5/4 majorthird (383Cinstead
of 386?). The relationsbetween some of these just intervalsand their nearest
approximationsin the 72-set are shown in Table1 (where interval sizes are
rounded off to the nearestcent).

pc number

ratio size 72-set size error

8/7 231? 14 233? + 2?
7/6 267? 16 267(? ?0(
6/5 316? 19 317? + 1?
11/9 347(? 21 3501? + 3?
5/4 386? 23 383( - 3?
9/7 435? 26 433? -2?
4/3 498( 30 500? + 2(
11/8 5511 33 550? - 1?
7/5 583? 35 583? + 0?
(etc.-largerintervals
which are of these havethe same
octave-complements
absolutevaluesforerror)

TABLE 1: A COMPARISON OF SOME IMPORTANT JUST INTERVALS

WITH THEIR APPROXIMATIONS IN THE 72-SET
66 of NewMusic
Perspectives

Eachof the Studiesis correlatedwith (andnamedafter)one of the sixty-four

hexagramsin the I Chingor "Book of Changes." This correlationis basedon
the configurationof adjacentdigramsin the hexagram,as follows:of the three
disjunctdigramsin each hexagram,the lower one is associatedwith pitch, the
middle one with temporal density, and the upper one with dynamic level.
Each digrammay take one of four differentforms, and each of these is inter-
preted to mean one of four possible "states" in a parameter-low (- -),
medium (= =), high (_-=), and full ( ). Thus, for example, the hex-
agramassociatedwith the fifth Study is number59 ("Dispersion"),which has
the followingform:

I upper digram:dynamicstate = full

] middle digram:temporaldensitystate = medium

_ lower digram:pitch state = high

Relativemeansand rangescorrespondingto the four differentstatesareshown

in Example1.

1.0
.9
| .8 HIGH =
'E .7 .7 + .3

.6 (--)
MEDIUM = FULL =

i .5 .5 + .3 .5 + .5

.4
(==) ( )
0 LOW =
.3 .3 + .3
.
.2 (--)

.1
0.0

EXAMPLE 1: RELATIVE MEANS AND RANGES

CORRESPONDING TO THE FOUR DIGRAM STATES

Actually, the parametricstates of each Study are determined by twohex-

agrams-the first one (for which the Study is named) correspondingto the
 About Changes 67

parametricstatesat the beginningof the Study, the second to those at the end.
Where these terminalstates differin a given parameter,a gradualtransition
from one to the other is produced by the programusing a half-cosineinter-
polation function. At lower holarchicallevels, linearinterpolationis also used
for such changesof state duringthe course of a temporalgestalt-unit(or TG).
In both cases,two mean-valuesareused for eachTG-an initialand a finalone,
and these terminalvaluesareconnected by the interpolationfunction. For this
purpose, the followingformulaeareused:

linearinterpolation:

Vt = v1 + (v2 - v1) * (t- tl)/(t2 - tl)

half-cosineinterpolation:
vl + v-2 vl - v2 t - t)
vt = 2
+
2
os
t2 tl

wherev, is the valuein the parameterat time t, vl the initialvalue(at time t1)
and v2 the finalvalue (at time t2).

The first programgeneratestwo non-repeatingrandom sequencesof hex-

agramnumbers,one for initialstates,the other for finalstates,so everypossible
combinationof parametricstates occurs once at the beginning of one of the
Studies,and once at the end of (usually)a differentone. "Changinglines" for
the initialhexagramare then inferred-such as would effect its transformation
into the finalhexagram.Becauseof this indirectway of derivingchanginglines,
they occur more often than they do when the I Chinghexagramsareobtained
in the traditionalways, where the probabilityof a changingline is one in four,
or 25%; here approximately50%of the lines arechanging.
On the basisof the initialand finalparametricstatesof each Study, the first
programalsodetermines(1)whetherit is-tobe monophonicor polyphonic,and
then (2) the averageverticaldensityofits elements,(3) the overalldurationof the
Study, (4) its averageclang-duration,and (5) the initialand finaltonic locations
for the Study, asdescribedmorefullybelow. To determinewhethera Studywas
to be monophonicor polyphonic,it wasfirstconsideredpotentially polyphonicif
at leastone parameterwas in the "full" state, eitherat the beginningor at the
end. When this was the case, a weighted randomdecisionwas made, with the
weightingadjustedin such a way that approximatelyhalfof the sixty-fourStud-
ies would be polyphonic,the other halfmonophonic.
Both temporaldensityandverticaldensityvaryexponentiallyin the Studies-
i.e. the probabledistributionof valuesin these two parameterswill be uniform
on a logarithmicscale.Thus, forexample,the averagetemporaldensitymTd of a
68 of NewMusic
Perspectives

TG will be computed as mTd = 2mS,where S is the stochasticallycontrolled

variable,andmSitsaveragevalue.Similarlyforverticaldensity:mVd = 2mz.But
while the meanvaluesfor temporaldensitydepend directlyon input data,those
for verticaldensity are determinedby a formulawhich relatesthem to pitch-
range, averagetemporal density, and the number of polyphonic strata, as
follows:

mZ = .5 + (1 - VmS/1.6)* /nP/195/ /EF

where mS is the averagevalueof the temporaldensity exponent and 1.6 is the

maximumvalueit canhavein anyStudy;nP is one-halfof the numberof pitches
in the range(always? 195); and Nst is the numberof polyphonicstratain the
Study(either1or 2). The averageverticaldensityof anyStudythus variesdirectly
with the pitch-range,and inverselywith the averagetemporaldensityand the
numberof strata.
The totaldurationof eachStudyvariesdirectlywith the average"volume" of
the three-dimensionalspaceoutlined by the rangesin the threebasicparameters
(pitch,temporaldensity,anddynamiclevel),andinverselywith the averageden-
sityof eventswithin thisspace.Thisvolumeis proportionalto the productof the
averagerangesin the threeparameters,and the "densityof events" to the prod-
uct of (average)temporaldensity,verticaldensity, and the numberof strata,as:

Volume nP * nS* nL
Density mTd * mVd* Nst

whereallvariables(exceptNst)arearithmeticaveragesof the correspondingvari-

ablesin the initialand finalstatesof the Study. The resultsof this computation
are later re-scaledto yield a minimum durationof 1'20" and a maximum of
2'40", so the averagedurationfor the Studiesin the set is about two minutes.
EachStudyis organizedinto TGs at two holarchicallevelsbetween those of
individualelementsand the Studyas a whole-clangs and segments.Here I have
deliberatelyavoidedTG-articulations at both the sequence-andsection-levels,in
an effortto enhancethe sense of continuityand the perceptibilityof contour at
the segment-leveland over the whole Study. The averageclang-durationsin
individualStudiesweremade to depend (inversely)on theiraveragedensities(as
definedabove), andscaledto yielda minimumdurationof2.4/\/2 = 1.697, and
a maximumdurationof 2.4 * /2 = 3.394 seconds.
The harmonicorgnization of the Studieswill be describedin more detail
later,but a briefsummaryhere may help clarifycertainother operationscarried
out by this firstof the two programs.The pitch classes(pcs) availablewithin a
given clangconstitutea "mode" of (usually)seven differentpcs, one of which is
treatedasa localor temporarytonic or root. In monophonicStudies,a new root
and a new mode arechosen for each new clang.In polyphonicStudies-whose
 About Changes 69

clang-boundariesare not, in general,synchronous-a new root and mode are

chosen wheneverthe startingtime of a new clangin one stratumis laterthan
half-waythroughthe durationof the concurrentclangin the other stratum-so
pcs in the two strataaredrawnfrom the sameset more than halfof the time. In
both monophonic and polyphonic Studies, the seriesof root-progressionsis
controlled in such a way that each Study ends with a dominant-to-tonic
"cadence" on the same root (the "global tonic") with which it began. Initial
tonic pcs areorderedin a waywhich distributesthe seventy-twopcsgivenby the
tuning system over the sixty-fourStudies as uniformlyas possible, by simply
omitting every ninth pc in the seriesfrom 0 to 71. The final tonic location is
determinedin a way which will be explainedlater.
The output of this firstprogramconsistsof sixty-fourblocksof data,eachof
which is used asinput to the secondprogramto generatethe detailsof one of the
Studies.Eachblock includesthe followingdata:the numbersof the hexagrams
defininginitialand finalparametricstatesfor the Study, its total durationand
averagedang-duration,its initialtonic pc and the numberof unit steps (in har-
monic space)to the dominantof the "targettonic," the numberof polyphonic
strata,and the initialand finalmean valuesand rangesfor pitch, temporalden-
sity, dynamiclevel, and verticaldensity.

C. INDIVIDUALSTUDIES

In generatingthe output data for an individualStudy, the second program

works"from the top, down." That is, it firstdeterminesthe durationandother
parametricstatevaluesfor the firstsegment, then for the firstclangin that seg-
ment, and then for successiveelementsin that clang.When all the elementsin
the firstclanghave been generated,it determinesthe statevaluesfor the second
clang and for its elements. After the last element of the last clang in this first
segment has been generated,the programproceedsto the second segment, its
firstclang and the latter'ssuccessiveelements, and so on. In the case of poly-
phonic Studies,these operationsarecarriedout "in parallel,"in such a waythat
successiveelements' parametricvaluesaregeneratedalternatelyfrom the two
polyphonicstrata.This was necessaryto maintainharmoniccoherencebetween
the two strata,sincepitchesin the two stratawereto be drawnfrom the sameset
of availablepitch classesat anygiven moment, wheneverthis was possible.
The number of segmentsin a Study is approximatelyequal to the average
numberof clangsin a segment, and the averagesegment-durationapproximates
the geometricmeanof clang-and Study-durations,althoughindividualsegment
durationsvaryrandomlywithin a rangeof + 25%of this averageduration.For
each segment, an initialand finalmean valuein each of the other parameters-
pitch, temporaldensity, dynamiclevel and verticaldensity-are chosen within
the availablerangearound the current"global" mean for the Study, which is
70 of NewMusic
Perspectives

determined-as explainedearlier-bya half-cosineinterpolationbetween the ini-

tial and final mean valuesfor that parametergiven by the input data for the
Study.Eachof the terminalmean-valuesforthe sgmentiscomputedasthe arith-
metic averageof two randomvalues,which resultsin a tendencytowarda "tri-
angular"frequencydistribution,ratherthan a uniformone-peaking at the cur-
rentglobalmeanand decreasinglinearlytowardthe upperandlower boundaries
of the currentrangein that parameter.This wasdone to lowerthe probabilityof
extrememean valuesat the segmentlevel, which would have resultedin overly
narrowrangesat the clanglevel.
The averageclang-durationfor each Study is given in the input datafor that
Study, but-as with segment-durations-thedurationsof individualclangswere
made to vary randomlywithin a range of + 25% around the averagevalue.
Parametricmeansfor eachclangarechosenwithin segment-meansin relationto
the currentmean of the segment-as with segment-meansin relationto the cur-
rentglobalmeanof the Study-except thathere (a)the currentsegment-meanis
determinedby linear(ratherthan half-cosine)interpolationbetween the termi-
nal values,(b) only a singlevaluein the parameteris used for a clang(thatis, its
parametricmean will be constantthroughoutthe clang), (c) this valueis deter-
mined by a singlerandomnumber(so the frequencydistributionof clang-means
would tend to be uniform),and (d) the clang-meanfortemporaldensityis made
equalto the currentsegment-meanitself,ratherthan being allowedto varyran-
domly aroundthat mean, in orderto ensurea sufficientrangeof element-dura-
tions within each clang.
In all of my earlierstochasticmusic, the articulationof successiveTGs was
effectedvia the "similarityfactor"only-involving differencesin mean-valuesin
variousparameters.In an effortto incorporatethe "proximityfactor"aswell, in
the articulationof successiveclangs,a new procedurewasused herewhich inter-
poses a delaybeforethe beginningof each new clang(effectivelyprolongingthe
durationof the finalelementin the precedingclang),accordingto the following
formula:

Delay = (Dmax - Dur) * (1. - Pdst/Pdmx)

whereDur is the element-duration,Dmax the maximumelement-durationpos-

sible in that clang, Pdst the pitch-distancebetween the two clang-means,and
Pdmx the largestvaluethis can have. The magnitudeof the delayis thus deter-
mined by the relativedistancebetween the pitch-meansof the two clangs,and
by the differencebetween the durationof the lastelementin the firstclangand
the maximumelement-durationallowedfor that clang(givenits temporalden-
sity mean and range).The smallerthe distancebetween the pitch-meansof the
two clangs (relativeto the maximum value it could have, given the available
rangeof clangpitch-meanswithin the segmentat that moment), the longerthe
delayis likelyto be. Thus, for example,if the distancebetween the pitch-means
 About Changes 71

of the two clangshappensto be zero (i.e. if the two clangshavethe samepitch-

mean, which could occur, althoughit's not verylikely),the amountof the delay
will be such that the (modified)durationof the lastelementin the firstclangwill
be equalto the maximumelement-durationin thatclang.If, on the other hand,
this distancehappensto be at maximum,the delaywill be zero, andthe duration
of that lastelementwill remainunmodified.
The hierarchical(or holarchical),recursivecharacterof the program,already
describedfor segmentsand clangs,continuesat the element-level,althoughele-
ment-durationsaregeneratedmore simplythan were clang-and segment-dura-
tions (asthe reciprocalof a temporaldensityvaluefor the element),and element
dynamiclevelsaremade equalto the clang-meanin that parameter(so dynamic
levels remainconstantthroughout a clang). The valuederivedat this level for
verticaldensity-truncatedto the next-lowerinteger-determinesthe numberof
pitchesin the element. As with clangsand segments, parametricvalues(other
than dynamics)for an element are drawnfrom the availablerangearound the
clang-mean, but for the pitch-parameter,other, specificallyharmonic pro-
ceduresintervenehere to determinea set of availablepitchclasses(or pcs) before
the actualpitchesareselected.These procedureswill be describedin the section
that follows.

D. HARMONICPROCEDURES

My intentionsin this work, with respectto harmony,includedthe following:

1) that one of the pcs in everyclangshould functionas a temporarytonic

or root in relationto all the other pcs in that clang-which latterare
interpretedas a kind of temporary"mode" for that clang;
2) that the root pc would changefrom clangto cang, by meansof a root-
progressionchosen stochasticallyfrom a set of possibleroot-progres-
sionswith pre-setrelativeprobabilitiesassignedto them;

3) that the pcs in a mode should tend to form relativelycompactsets in

harmonicspace, both in relationto the other pcs in that clangand to
those in the previousclang,and finally
4) that the "random walk" characterof the seriesof root-progressions
should graduallybe "focused" in such a way that each Study would
end with a dominant-to-tonic progressionto the same root pc with
which it began-and in the samemode.

To achievetheseintentionsrequireda carefulanalysisof the 72-set andits several

possiblemappingsin harmonicspace.For example,the pcs in the 72-set can be
72 of NewMusic
Perspectives

mapped in pitch-classprojection spacesof 2, 3, or 4 (or more!) dimensions,

accordingto the prime-limitbeingconsidered.ForChanges,I decidedto assume
an 11-limit (five-dimensional)harmonicspacefor the modes, a 7-limit (four-
dimensional)harmonicspacefor root-progressions,and to locatethe final,"tar-
get tonic" on the same 3,5-plane as the initialtonic (which impliesa 5-limit,
three-dimensionalspace for this relationbetween initial and final tonic loca-
tions). Examples2 and 3 show some of these mappingsof the pcs in the 72-set,
in pitch-classprojectionspacesof two and three dimensions(correspondingto
prime-limitsof 5 and 7, respectively).Note that-because the 72-set is an equal-
temperedsystem-its latticestructureis peridic in harmonicspace (no matter
what the dimensionalitymaybe of thatspaceinto whichit is mapped).Thatis, it
repeatsitselfendlesslyin alldirections.It was decidedto use asthe targettonic in
eachof theseStudiesone of the manylocationsof thattonic in the 3,5-plane,in a
direction(in relationto the initialtonic) similarto the directionin which Bach's
harmonicprogressionstend to move in a mappingof the 12-set in harmonic
space-i.e. towardthe left alongthe 3-axis(viadescendingfifths-e.g. V-I) and
upwardalongthe 5-axis(lessquickly,andmostlyviathe descendingminorthird
progression-e.g. I-vi). Example4 shows the configurationof recurringtonics
(in relationto an initial1/1 or "O"),in an abbreviatedbut extendedmappingon
the 3,5-plane.The locationused foreachStudywasone of the threeindicatedby
the arrows-which one of the three depending on the estimated number of
clangs(andthus, the numberof root-progressions)in that Study. The numbers
in parenthesesgive the numberof unit steps, along the 3- and 5-axes, respec-
tively, from the initialto the finaltonic location.
Eachof the sixty-fourStudiesbegins (andthus ends) on a differenttonic pc,
and these form an ascending integer series, beginning with 0 (E } -read:
"three-sixthsof a semitone below E") for Study Number 1, and ending with
f)
71 (DO/Eb for Study Number 64, skippingeveryninth pc in the series.The
other pcs of the mode associatedwith a root are chosen from a set of alter-
natives-for each of six "scale degrees" (in addition to the tonic)-given as
input data to the program(but common to all sixty-fourStudies).These are
arrangedin "stackedthirds" order: prime, third, fifth, seventh, ninth, elev-
enth, thirteenth, and they include from three to five alternativesfor each
degree above the tonic. These are listed in Table2, which gives both the pc
numberin the 72-set and the just ratioor ratiosmost closelyapproximatedby
that pc (in parentheses).The most importantharmonicrelationshipsamong
these variousalternativesare shown in Example5, representingtheirlocations
in harmonicspace (or, more precisely,in a pitch-classprojectionspace essen-
tiallyin 7-limitform, but with the additionalratiosofll interposedalongthe 3-
axes (andin parentheses)).The choiceof a particularpc (or ic-interval class-in
 About Changes 73

50 - 20 2 --- 3a

57 -17 -e -39 3 -9 51

$5 -2-I'71 - - ~8I---3
C---
-------- f --- - ---------- ---- 1I
----6----52--5 5 -3---- '---
17'--- 707 5

13 ' 55- ," - -- 37 -- 7 49 -- 19 -fi- 31 1 43 13

2 4 - I4 1 -- 56 - 2- 6- ?1-39 0
3/t a8

3 -
A1 3 - 33 -- 3 --- # --- i

/o -- 10O 52
-5 a;

EXAMPLE2: THE 72-SET IN THE 3,5-PLANE

74 of NewMusic
Perspectives

57 ;0

6-----715 t ---- ---- -----3- 9 la -

3l--- --- 3I

40-10
7'7'~~~~~~~~~~~~~~Y

P7'--13-59--1

1 -S9/

St a----

EXAMPLE3: THE 72-SET IN 3,5,7-SPACE

About Changes 75

Lfo

z
0
H

0
U
76 of NewMusic
Perspectives

relation to a given tonic pc) for each degree is determined by severalcondi-

tions, some of which might be describedas "rules," while othersaremore sta-
tisticalin character.The rulesincludethe following:

1) in the initial (and thus also the final)tonic set, the fifth is alwaysmade
equalto 42 (3/2), and the seventhis allowedto equal 58 (7/4) only if the
third (alreadychosen) equals16 (7/6);

2) in the dominant set preceding the final ("target") tonic, the third is
always = 23 (5/4), the seventh always = 58 (7/4);
3) the various "thirds" between adjacentdegrees may vary in size only
within specifiedranges:from a minimum of 12 (9/8) to a maximum of
26 (9/7) between prime and third or third and fifth, a minimum of 16
(7/6 or 75/64) and a maximum of 30 (4/3) between adjacentdegrees
above the fifth;

4) no "mistuned fifths" are allowed between non-adjacent degrees (as

between the thirdand seventh, fifth and ninth, and so on)-i.e. anysuch
intervalmust eitherbe preciselyequal to 42 (3/2) or differfrom it by an
intervalgreaterthan 3 (a "quarter-tone");
5) no octaves (either exact or "mistuned") are allowed between those
non-adjacentdegrees which share a common pc, or approximatethat
condition too closely (as between the third and the ninth or eleventh,
the fifth and the eleventh or thirteenth)-i.e. no "seventh" larger
than 68 is allowed, and no "ninth" smallerthan 4. Thus, any inter-
val between non-adjacent degrees must differ from an octave by at
least 4 (two-thirds of a semitone);

6) if the third = 19 (6/5), the fifth must equal 42 (3/2)-thus disallowing

both the "flat" and "raised" fifths when the third is of the ordinary
minor form;
7) the raisedfifth-46 (25/16)-is only allowedwhen the third = 23 (5/4).
prime: 0 (111)
3rds:16 (7/6), 19 (6/5), 21 (119), 23 (5/4), 26 (9/7)
5ths: 35 (7/5 or 45/32), 42 (3/2), 46 (25/16)
7ths: 58 (7/4), 61 (9/5), 63 (116), 65
(15/8)>
9ths: 5 (21120),7 (15/14,16/15or 77/72), 9 (1/11 or 35/32), 12 (9/
liths: 28 (21/16),33 (11I/8),35 (7/5 or 45/32)
13ths:49 (8/5 or 77/48), 51 (18/11or 105/64), 53 (5/3), 54 (27/16)

TABLE 2: ALTERNATPvE PCS FOR A MODE

78 of NewMusic
Perspectives

+6 16
/ \ /\

53

49
J- ^?
V
thirds: sevenths: elevenths:

16 (7/6 or 58 (7/4) 28 (21/16

75/64)
61 (9/5) 33 (11/8)
19 (6/5)
63 (11/6) 35 (45/32
or 7/5)
21(11/9)
65 (15/8)
23 (5/4)

26(9/7) thirteenths:
nnths:
49 (77/48
y 5 (21/20 or
C or 8/5)
fifths: 135/128)
51 (105/64
35 (7/5 or 7 (77/72,16/15 or 18/11)
45/32) or 15/14)
53 (5/3)
42 (3/2) 9 (35/32 or
12/11) 54 (27/16)
46 (25/16)
12 (9/8) 56 (55/32
or 12/7)
16 (75/64
A
or 7/6

I
\.

EXAMPLE 5: HARMONIC RELATIONSHIPS

AMONG ALTERNATIVE PCS FOR THE MODES
 About Changes 79

Some of these rules correspondto similarrules for chord-constructionin

both traditionaland jazz harmonicpractice(and I should perhapsadd here
something which has not been made explicit before: the pcs of a mode are
often heard simultaneously as well as successively-as "chords" as well as
"melodiclines"--thus the ambivalence(whichmay have been noticed already)
in my use of the terms "tonic" and "root"). Other rules were designed to
avoid certainambiguitiesand/or conflicts that might otherwise occur in the
creationof these modes. Although these rulesappearto be quite restrictive,a
verylargenumberof modal sets were still possible,but these were furthercon-
strained by what I have referredto (above) as "statistical" conditions, as
follows:

The pcs that remainavailablefor a given modal degreeaftertesting against

the rulesjust listed areassignedvaryingprobabilitiesdependingon the sums of
their harmonicdistancesto pcs alreadychosen for that clang-and to the pcs
actually occurring in the clang just preceding (I say "actually occurring"
because-due to the random process involved in the selection of pitches in a
clang-it is alwayspossiblethat one or more of the pcs constitutingthe mode
will not occur). The relationbetween these probabilitiesand harmonicdis-
tances varies according to the modal degree in question (the constraint is
"tighter" for the higher degrees), and whether this was the first clang in the
Study or not (the constraintis "looser" for the firstclang),but in generalthat
relationis an inverseone. That is, the lower the sum of harmonicdistances
between a pc and the others precedingit, the higher its probabilityof being
chosen-and vice versa.This constraintwas made strongerfor higherdegreesof
the mode (arrangedin "stackedthirds" order, remember)by raisingthe har-
monic-distancesum to a power correspondingto the "height" of the degree,
as follows:

Pr(j) O 1/Hdsm(j)n +k-1

where Pr(j) is the relativeprobabilityof the jth pc in the set of still-available

pcs for that degree, Hdsm(j) equalsthe sum of that pcs harmonicdistancesto
precedingpcs, n is the order-numberfor the modal degree (i.e. n = 2 for the
third, 3 for the fifth, 4 for the seventh, and so on), and k = 1 for the firstclang,
2 for alllaterclangs.The resultof all this is that therewill be a tendencyfor pcs
to form relativelycomactsetsin harmonicspace, with this tendencybecoming
strongerfor higher modal degrees-and conversely,so there is more freedom
for randomvariationin the lower degrees.
It might be noted that the sets of alternativepcs for modal degreesyield
80 of NewMusic
Perspectives

seven differentkinds of triads, only four of which are familiarin traditional

western harmony(numbers2, 4, 6 and 7, below):

1) septimalminor 0 (1/1),16 (7/6), 42 (3/2)

2) 5-limit minor " 19 (6/5) "

3) 11-limit "neutral" " 21 (11/9) "

4) 5-limit major " 23 (5/4) "

5) septimalmajor " 26 (9/7) "

6) augmented " 23 (5/4), 46 (25/16)

7) diminished " 16 (7/6), 35 (7/5)

Another possibleform of the diminishedtriad-0 (1/1),19 (6/5), 35 (7/5)-was

avoided becausethe most likelyseventh degreewith such a triadwould have
been 65 (15/8), and the perfect fourth (30 (4/3)) which this forms with 35
would have introducedan unwanted ambiguitywith respectto the root. The
sets of alternativepcs for scaledegreeswere designedto avoidpcs which might
compete with the nominalroot, and the perfectfifth and fourth-and even the
(5-limit) major third and minor sixth, though less strongly-have very clear
root-defining effects. Thus, the perfect fourth itself-30 (4/3)-was not
included as a possible eleventh in a mode, and 49 (8/5 or 77/48) was only
included as a possible thirteenth becauseof its dual character-and harmonic
distancevaluesfor this intervalwere set to correspondto its interpretationas
77/48, ratherthan 8/5. (The same thing was done for the intervalformed by
pc 26 (9/7), to avoidits interpretationas 32/25, which-because of the way in
which I calculated harmonic distances (for an explanation of which, see
below)-would havegiven it more prominencethan I thought it should have.)
The seventh-chords which can arise by way of this procedure for con-
structing modes include most of the traditionalones (major, minor, domi-
nant, half-diminished, minor-major, augmented, and so forth, but not the
diminished seventh), plus severalothers which are of interest, including the
one used by Ives as the primary chord in the "Choral" of his Three
Quarter-tonePieces-O (1/1), 21 (11/9), 42 (3/2), 63 (11/6). Ninth-chords
include-again-all of the traditionalones, plus the blues "flat 7, sharp 9,"
and a very interesting group of new ones with pc 9 as the ninth of the
mode. This pc-at the "quarter-tone" position between the 12-set's minor
and major seconds-functions in the 72-set most frequently as the major
third above the "dominant" 7th-i.e. it can be analyzed as 58 (7/4) + 23
(5/4) = 81 (mod 72) = 9 (35/32). The fact that it occurs in a "dominant"-
type pc set more often than the more familiarminor or major ninth is sug-
 About Changes 81

gestive: perhaps the latter are merely the best approximations availablein
the 12-set for this interval! Finally, the eleventh-chords include a good
approximation of Partch's "otonality hexad"-O (1/1 + 0(), 23 (5/4 -
3?), 42 (3/2 - 2?), 58 (7/4 - 2?), 12 (9/8 - 4?), and 33 (11/8 - 1?).
The basic formula for the harmonic distance between any two pitches is
Hd(a/b) = k logx(ab),wherealb is the frequencyratiorepresentingthe interval
(in its maximallyreduced,"relativeprime" form), and k simplydeterminesthe
unit of measurement(with base-2 logarithms,if k = 1, Hd is in "octaves").
The form used in this program, however, is a bit different,in two respects.
First, it is a measure of the harmonic distance between pitch-classes,rather
than actualpitches. Second, sincewe aredealingherewith a temperedsystem,
a tolerance ruleis invoked,which essentiallysaysthat we can assumethe simplest
integer ratio within the tolerancerangearound the temperedpitch to be the
harmonicallyeffectiveone (that tolerancerangeis here takento be + one-half
the size of the smalleststep in the tuning system-i.e. + 1/144of an octave or
8.33.. .). The firstqualificationmeans that we areconcernedwith a distance
not between points in the full, n-dimensionalharmonicspaceitself, but rather
between points in the (n-l)-dimensional"pitch-classprojectionspace." This,
in turn, means that the formulafor harmonicdistancemust be replacedby
another of the form: Hd (a'lb') = k log(a'b'), where a' = a/21, b' = b/2J,
and i andj arethe largestintegerexponentswhich yieldintegervaluesof a' and
b'. The second qualificationmeans that -where there are two or more rela-
tivelysimpleintegerratiosdefiningintervalswithin the tolerancerangeof a pc,
the one whose ratio-terms'product is smallestdeterminesthe harmonicdis-
tancevalueassignedto that pc. It has alreadybeen mentionedthat two excep-
tions were made to this procedure,involvingpcs 26 and 49. Pitch-class26 (@
433?) approximatesboth 32/25 (@427?) and 9/7 (@ 435?), while pc 49 (@
817() approximates 8/5 (@ 814?), 77/48 (@ 818?), and 45/28 (@ 821C).
While I wanted both of these pcs to be included among the availablealter-
natives(forthirdsand thirteenths,respectively),I wanted 26 to be treatedas a
9/7, and 49 as a 77/48, so their minimal harmonicdistancevalueswere over-
ridden in anotherpart of the programwith the higher values. (I see now, in
studyingthe programagain, that the valueI assignedto 49 was that of 45/28
ratherthan 77/48-i.e. log2(315) = 8.30 ratherthan log2(231) = 7.85-but
fortunatelythis errorturned out to be a small one, with scarcelynoticeable
effect on the finalresults.)
Once the pcs of the mode for a clang have been chosen, the programis
almost ready to proceed to the selection of actual pitches, within the range
alreadydeterminedfor that clang.As at all higherlevels,this involvesa random
process, but at this level the process is further constrainedby two kinds of
probabilitydistributions,one providingsome control over the rate of recur-
rence of each pitch, the other correlatingmodal degree with register. The
82 of NewMusic
Perspectives

probability of a given pitch being chosen by the random process at any

moment was computed as the productof two probability"factors"storedin a
two-dimensionalarraycalledPPR(N,L), where N = 1 or 2, and L is an index
for pitch (L = 1 for the lowest pitch in the harps'range,L = 452 for the high-
est). For all valuesof L, PPR(1,L)was initializedat 1., so all pitchesbeganwith
the same relative probabilities. Just after a pitch is chosen for an element,
PPR(1,L)for that pitch is reducedto a verysmallvalue,and then increasedstep
by step, with the generationof each succeedingelement (at any otherpitch),
until it is againequal to 1.0. The resultof this procedureis that the immediate
recurrenceof a given pitch is made highly unlikely (althoughnot impossible,
especiallyin long and/or dense clangs,and in a polyphonictexture), with the
probabilityof recurrenceof that pitch graduallyincreasingover the next several
elements until it is equal to what it would have been if it had not already
occurred.The other probabilityfactor-PPR(2,L)-is used to effect a correla-
tion between modal degree and register,as shown graphicallyin Example6.
Note that while the root or tonic of the mode has an equal probabilityof
occurringanywherewithin the pitch-rangeof the clang-and all other modal
degreesareequallylikelyat the upper boundaryof the clang'spitch-range-the
highermodal degreeshave low probabilitiesof occurringin the lower regions
of the clangrange(and converselyfor the lower modal degrees).

1.0
(root)

.8
(Sth)

.6 (7th)

|L4
(9th)
*.4

.2

0.
(k pitch-rangeof the clang

EXAMPLE 6: CORRELATION OF MODAL DEGREE WITH REGISTER

 About Changes 83

Finally, values are determined for the starting-time(or epoch), duration,

pitch(es), and dynamiclevel of each element in the clang. Element-durationis
computed as the reciprocalof a temporaldensity value for that element, and
the epoch is given by the sum of epoch and durationvaluesfor the previous
elementin the stratum(plusthe "delay" describedearlier,when the elementis
the firstin a new clang).These time valuesareinitiallycalculatedon a virtually
continuous scale-as in Bridge-but (unlike Bridge)I decided in this work to
quantizeor "rationalize"these valuesso they could be representedin the stan-
dard metricalrhythmicnotation in the score and parts.This was done as fol-
lows: for the epoch of each element, the programcomputes (and prints out
with the other parametricvalues for that element) the absolute differences
between the initiallycalculatedvalueand both the nearestsixteenth-noteand
the nearesttripleteighth-note. It is then left up to the person transcribingthe
numerical output data into musical notation to decide which of the two
rational approximations to use, based on the magnitude of the "error"
involved, and on the epochs and errorsfor any other elements which may
begin within the same quarter-note(sincethe two divisionsof the quarter-by
3 and by 4-cannot generallybe mixed within a given quarterin our standard
system of rhythmicnotation). Example7 shows an exampleof a page of out-
put data, with the valuesfor a single element boxed, and the "error" values
just describedshown circled.
When the ending-time of an element equals or exceeds a predetermined
ending-timefor the clang, the programcomputes a new root pc for the next
clang, and a new mode for that clang. The interval-class(ic) between this new
root and the root of the previousclangthus definesa root-progression,and is
determinedas follows: an arrayis used to store an initialset of relativeproba-
bilitiesfor allowableroot-progressions(these probabilitiesare the same for all
sixty-fourStudies), as shown graphicallyin Example8, and listed in Table3.
This set of probabilitiesis conceivedas determininga smallerset of six vector
componentsin a three-dimensionalharmonicspace,and these, in turn, can be
reducedto a single "resultant"vector which indicatesthe directionand aver-
age rate of root-movement through that space-assuming, of course, that a
largenumberof such root-progressionswill be involved.The resultis a kind of
directed"randomwalk" through the harmonicspace.
In order to furtherensure not only that this "random walk" will have-
over the long run-the appropriatedirectionand ratein relationto the location
of the "targettonic" (or rather,the dominant precedingthis tonic), but that
the movement will become graduallymore "focused," and finallyarriveat its
goal, the set of individualroot-progressionprobabilitiesis revised,for eachnew
clang, accordingto the actualdirectionand distanceremainingto the target.I
won't go into more detailhere about the mechanicsof this process,except to
note that this part of the programturned out to be more complicatedthan I
had expected it to be, and that it didn't alwayswork! That is, there always
 84 of NewMusic
Perspectives

O O
II 0 2
o o

O . 10 r0 >

0o
00 .0 0
N* O
O. . N *
U)0 C t O

-ino .00 Ua 0 ?
' 00i0A eO ? ? 0
* . NO0E -~~~~~~~~~^0 N 0. _ II O00 N
r?U?W~ 0*<)0
c~la9o
? ~ ? *

?
'a"^O O oJ in O
*o _-
00 II

* *0 t 0oO a0 0*
4
0)o 0c * * * ON 0O
(- 0
of o n o*0 ,
C4N Y 0 e . * *

~~* ~~~~ ~~*

^~~~Qo) -
- -N U)
*. -

0<00 0 h . 11 * I
oi <p i - Z-CT Uo ) z
??. * Z * * ZN
Uo --0 o0.- <4
- o ?* U n')

11 n _ 0 U
o o? ?
00

?00 1(1 00i

zmatoZU)'0 --
oo o0I " a
O * * - 0 r-) u
Uoo 1 N o

0
o O o o - _*o* 0 - - .e
ooU
00Uo *0* . . Z . Z
fIloU oo <4 i? o0 II >
0* *
2* oS o0 oN U I 0 o
000ooo o- 0 o * z z
00 o $ N< 0 U 00 U 00
Z 6N * * *
< 000 * * 0t 0 0
U-
i z 0-
1)4 03
I *IHr'
oo00 N00/0
UN -
N 00 0
zo *? * _ o0
4Q * * -' *0
2O0 o?. 0 0
00 0 0 --

e..U) 0 0 i ? ? 0 ?
O4NZ o * o0 NO 0
* NO [) \ 7 S
. CV<O H) * U . ,*
BO - 0 0 * * U) 0 0 000 .
4 H *. ) *. .
N *. U) N -0 0 0 0
--
QZ ? ? 1(- _ 0 | * 3
Z --. uJ ?0 - N o - 0 n

<0 f0 - 0-- 0 ' 0' ? * 00

-U1* JO 0o o 1 * ,
_-. vU o0nr o- * * ?o * * o
x4 I < * i oo -
00 - i a
?oo.< ,a o*c - o 0 o 0
4< * * O, 0 * O *
Xr f> - 0C? 0* 0
00 -
Wjo- 02K \0 oN * * * *
r-- o- N oo inI o 0 n
.*oo -4) * 0 * 0
U,.* * il 4 -1> o _ n * . 0o
-tro
-. f%l NO - *D -
qo"s'z - _ - * 'v ')
NN i - 0
eO N 1I * I 11 0* l
H- -Q 0 Zs ' N NO 2 <) z
0 Z0'H 0
l .0 rd
5) IrZ
.0 o o X
~~LUO~IL .4
< ..
I W-51 .1
 2: ST= 0.72. DR=
6.b.2 OCO0.0.0
16 0
M= I Q=. .

3: ST= 0.97, DR= 0.25. E = 1.22, CMS- 1.8

8.3.3 00.0 0.0.0 0.0.0 0.0.00 0.0.C 0.U.0
27 o 0 0 C 0 0
M= I, C= 2. C/3. 03 0/4.032 DYN: 3

3: ST- 1.08. OR=

5.4.3 0.0.0
9 0

M= 1. 0=2.

4: ST= 1.224 DR= 0.25. ET= 1.7. CMS= 2.3

6.2.2 0O0.0.0 0.0.0 O.C.0 0.0. 0.0.0 0.0.0
16 0 0 0 C C 0
M= 1, G 2, 1/3. 115 1/4.032 DYN: 3

0 27 53 69 16 39 62 9 -1 6 6 0
0 26 42 61 1Z 35 54

CLANG 0.358 1I.4 2.27 4 12

172.54 92.33 1.41 1.02 5.36 0.43 0.58 0.15

1: ST= I .b4 DRR 0.25. ET= 2.09 CM5= 3.5

642.3 OC.O 0.0.0 0.0.0 O.O.C 0.0 .00
-7 0 0 0 0
M= 1. G= 2, 3/. 156 3/4.094 OYN: 5

4: ST= 1.48. DI=

3.5.3 0.0.0
69 0

EXAMPLE7: THE FIRSTTWO PAGESOF OUTPUT DATA FOR S

86 of NewMusic
Perspectives

rI- - - - 5/
I* - - _ - 56

EXAMPLE8: AVAILABLEROOT-PROGRESSIONS

root-progression root-progression
inteval-class probability inteal-class probability
0 .00 30 .30
7 .01 42 .10
11 .05 49 .10
12 .05 53 .11
14 .04 56 .02
16 .04 58 .02
19 .01 60 .01
23 .03 65 .10

TABLE3: ROOT-PROGRESSIONPROBABILITIES
 About Changes 87

remaineda certaindegree of unpredictabilityin the final convergencetoward

the dominant, such that the intended targetwas actuallymissed in about one
out of three runs of the program.When this happened, the output was dis-
carded,and the Studygeneratedagainwith a new randomseed. Sincethe total
durationand the averageclang-durationof each Studywere consideredcharac-
teristicfeaturesof that Study, derivedby the firstprogramby operationson its
terminal states in the three primary parameters-and not be be altered
arbitrarilyor contingently-the series of root-progressionswas requirednot
only to arriveat its target, but to arrivethere on time. Such a percentageof
"failures"is thereforenot surprising-given the essentiallystochastic natureof
the process. In each Study, the program had four chances to succeed: if it
arrivedat the target dominant at the sixth, fifth, fourth, or third clangfrom
the end, a "cadencing" routine was initiated, which kept it rooted on the
dominantpc, and set the mode in some form of (extended)"dominant7th,"
until the next to last (or in some cases,the last)clang,at which point it effected
a progressionto the final tonic. The similaritiesbetween this procedureand
what might be inferredfrom many of the cadentialpassagesin Bach'sPreludes
should be obvious-although profound differenceswill also be evident to any
listener,I am sure.