Mach. Lea n.: Sci. Technol. 6(2025) 035022 h ps://doi.o g/10.1088/2632-2153/ad 53d
OPEN ACCESS
RECEIVED
28 May 2025
REVISED
9 July 2025
ACCEPTED FOR PUBLICATION
28 July 2025
PUBLISHED
5 Augus 2025
O iginal Con en om
his wo k may be used
unde he e ms o he
C ea i e Commons
A ibu ion 4.0 licence.
Any u he dis ibu ion
o his wo k mus
main ain a ibu ion o
he au ho (s) and he i le
o he wo k, jou nal
ci a ion and DOI.
PAPER
A gene ic algo i hm o gene a e maximally o hogonal ames in
complex space
Sebas ián Roca-Je a 1,2,∗and Juan Román-Roche1,3
1Ins i u o de Nanociencia y Ma e iales de A agón (INMA), CSIC-Uni e sidad de Za agoza, 50009 Za agoza, Spain
2Depa amen o de Física de la Ma e ia Condensada, Uni e sidad de Za agoza, 50009 Za agoza, Spain
3Depa amen o de Física Teó ica, Uni e sidad de Za agoza, 50009 Za agoza, Spain
∗Au ho o whom any co espondence should be add essed.
E-mail: s oca@uniza .es
Keywo ds: e olu iona y algo i hms, me ic geome y, quan um in o ma ion, ame heo y, packing p oblem, homson p oblem,
sic po m
Abs ac
A ame is a gene aliza ion o a basis o a ec o space o a edundan o e spanning se whose
ec o s a e linea ly dependen . F ames ind applica ions in signal p ocessing and quan um
in o ma ion heo y. We p esen a gene ic algo i hm ha can gene a e maximally o hogonal
ames (MOFs) o a bi a y size nin d-dimensional complex space. Fi s , we o malize he concep
o MOF and demons a e ha i depends on he choice o an ene gy unc ion o weigh he di e en
pai wise o e laps be ween ec o s. Then, we discuss he ela ion be ween di e en ene gy
unc ions and well-known ame a ie ies such as igh and G assmannian ames and complex
p ojec i e p-designs. Ob aining MOFs poses a global non-con ex minimiza ion p oblem. We
discuss he ela ion wi h es ablished nume ical p oblems such as he Thomson p oblem and he
p oblem o inding op imal packings in complex p ojec i e space. To ackle he minimiza ion, we
design a hyb id gene ic algo i hm ha ea u es local op imiza ion o he pa en s. To assess he
pe o mance o he algo i hm, we p opose wo isualiza ion echniques ha allow us o analyze he
cohe ence and uni o mi y o high-dimensional ames. The gene ic algo i hm is able o p oduce
highly-symme ic uni e sal ames, such as equiangula igh ames, symme ic, in o ma ionally
comple e, posi i e ope a o - alued measu emen s and maximal se s o mu ually unbiased bases,
o con igu a ions o up o d=6 and n=36, wi h un imes o he o de o se e al minu es on a
egula desk op compu e o he la ges con igu a ions.
1. In oduc ion
A he beginning o e e y in oduc o y cou se on linea algeb a, one is p esen ed wi h he concep o a basis
o a ec o space. A basis is a collec ion o ec o s ha a e linea ly independen and o m a spanning se o
he ec o space. In ac , he dimension o a ec o space is de ined as he numbe o elemen s o a basis. I he
ec o space is u he endowed wi h an inne p oduc , he condi ion o linea independence is o en
supe seded by he s onge condi ion o o hogonali y be ween he basis ec o s, hus de ining an o hogonal
basis. A consequence o hese de ini ions is ha in an inne -p oduc ( ec o ) space o dimension done can
only c ea e se s o a mos do hogonal ec o s (which would cons i u e o hogonal bases). In ligh o his
ealiza ion, i is only na u al o wonde how o gene alize he concep o an o hogonal basis o se s o mo e
han d ec o s. These o e -spanning se s a e ypically e e ed o as ames, which a e de ined and
cha ac e ized in ame heo y [1,2]. The edundancy o ames p o ides obus ness agains e o s and is hus
use ul in he encoding and e ie al o signals, wi h applica ions in elecommunica ions and (quan um)
in o ma ion heo y [3]. Fo ins ance, equiangula igh ames (ETFs) wi h d2 ec o s p o ide symme ic,
in o ma ionally comple e, posi i e ope a o - alued measu es (SIC-POVMs) ha a e op imal o quan um
s a e omog aphy [4–9].
© 2025 The Au ho (s). Published by IOP Publishing L d
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
Se e al amilies o ames wi h di e en p ope ies can be de ined, such as igh and G assmannian
ames [10]. Many o hese p ope ies can be a ibu ed o pa icula ins ances o wha we call maximally
o hogonal ames (MOFs): ames o n>d ec o s ha , al hough no comple ely o hogonal among
hemsel es, a e maximally o hogonal in he sense ha hei pai wise inne p oduc s a e minimal.
As o hogonali y is a pai wise p ope y, inc easing he o hogonali y be ween a gi en pai o ec o s will
o en come a he expense o dec easing he o hogonali y be ween o he pai s o ec o s. Depending on how
much alue one assigns o he o hogonali y o each pai , one will a o one ‘maximally o hogonal’ se o
ec o s o ano he . Thus, he i s s ep in his pape is p o iding a o mal de ini ion o MOFs o a complex
space o a bi a y dimension. Once a med wi h his de ini ion, we discuss hei p ope ies and es ablish hei
ela ion wi h no able ame amilies such as igh and G assmannian ames, ETFs [10], SIC-POVMs [11],
maximal se s o mu ually unbiased bases (MUBs) [12,13] and complex p ojec i e -designs [14,15]. We also
show ha inding MOFs cons i u es, in mos cases, a nume ical op imiza ion p oblem. We discuss he
ela ion wi h es ablished nume ical p oblems such as he Thomson p oblem [16], he p oblem o inding
op imal packings in complex p ojec i e space [17] and he p oblem o inding op imal e e ence s a es o
quan um classi ie s [18,19]. Inspi ed by he use o a gene ic algo i hm o he Thomson p oblem [20], we
p esen a gene ic algo i hm ha is capable o gene a ing MOFs o a bi a y size in a complex space o
a bi a y dimension. Then, we isualize and measu e he quali y o he ames p oduced by he gene ic
algo i hm. In he cases whe e he p oblem o inding MOFs is equi alen o he Thomson p oblem o he
p oblem o inding op imal packings in complex p ojec i e space, we compa e he gene ic algo i hm wi h
s a e-o - he-a (SOTA) nume ical me hods.
The emainde o he manusc ip is o ganized as ollows. Sec ion 2is dedica ed o de ining MOFs,
discussing hei p ope ies, and ela ing hem o no able ame amilies. In sec ion 3we discuss ela ed
nume ical p oblems. The gene ic algo i hm is p esen ed in sec ion 4. In sec ion 5, we isualize and discuss he
esul s ob ained wi h he algo i hm and i s pe o mance. We end he pape wi h he conclusions and ou look
o ou wo k in sec ion 6. We also p o ide echnical de ails and complemen a y esul s in he appendices.
2. MOFs
2.1. De ini ion
The cen al objec o ou discussion is a se o ec o s Φd,n={|ϕi⟩}n
i=1⊂Cdwi h Cda ini e complex
coo dina e space o dimension d. The canonical sesquilinea inne p oduc o wo ec o s |ϕ1⟩,|ϕ2⟩is
deno ed ⟨ϕ1|ϕ2⟩.
De ini ion 2.1. [1]. A se o ec o s Φd,n={|ϕi⟩}n
i=1⊂Cdwi h n⩾dis a ame o Cdi he e exis ame
bounds 0 <A⩽B<∞such ha o e e y | ⟩∈Cd
A∥ ∥⩽
n
∑
i=1|⟨ |ϕi⟩|2⩽B∥ ∥.(1)
I one can se A=Bin equa ion (1), he ame is said o be igh [21]. Fo n=d, a ame is jus a basis. I
|⟨ϕi|ϕi⟩|=1∀ϕi∈Φd,n,Φd,nis a uni no m ame. Wi hou loss o gene ali y, we will assume ha all
ec o s a e o uni no m in he ollowing. In a ec o space o ini e dimension, such as he ones conside ed
in his pape , a ame is essen ially a ini e spanning se .
We can o mula e an ene gy associa ed o a se
EW(Φd,n) = ∑
i=j
W(|⟨ϕi|ϕj⟩|),(2)
wi h W, he weigh ing unc ion, an a bi a y eal- alued and inc easing unc ion well-de ined in he in e al
[0,1]. Such an ene gy g ows whene e he o e lap ( he modulus o he inne p oduc ) o wo ec o s
inc eases, ce e is pa ibus, and hus i will be minimized by a se wi h la ge o e -all o hogonali y. We deno e
a minimizing se as ¯
Φd,n.
P oposi ion 2.1. A se o uni no m ec o s ¯
Φd,n={|ϕi⟩}n
i=1⊂Cd ha minimizes EW(2) is a uni no m ame.
P oo . We ha e o show ha ¯
Φd,nis a spanning se o Cd. Fi s , i n=dany se ha minimizes EWis an
o hogonal basis and hus spans Cd. In he ollowing we assume n>dand p o e he p oposi ion by con a-
dic ion. Le us assume ha ¯
Φd,nis no spanning and deno e i s o hogonal complemen as ¯
Φ⊥
d,n. No e ha
W(0)⩽W(|⟨ϕi|ϕj⟩|)⩽W(1), wi h equali y in he lowe bound i and only i |ϕi⟩and |ϕj⟩a e o hogonal o
each o he . Le us de ine
ϕ⊥
k={|ϕj=k⟩∈ ¯
Φd,n|W(|⟨ϕk|ϕj⟩|)>W(0)}.(3)
2
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
Le us choose a ec o |ϕk⟩∈ ¯
Φd,nsuch ha ϕ⊥
kis no he emp y se . A non-spanning se wi h n>d ec o s
mus con ain al leas h ee such ec o s. Then, he ame ¯
Φ′
d,n o med by eplacing |ϕk⟩by |ϕ′
k⟩∈ ¯
Φ⊥
d,nin ¯
Φd,n
has ene gy EW(¯
Φ′
d,n) = EW(¯
Φd,n)−ϵwi h
ϵ=2∑
j∈ϕ⊥
k(W(|⟨ϕk|ϕj⟩|)−W(0))>0,(4)
which implies ha ¯
Φd,ndoes no minimize EW.
The e o e we e e o any se ha minimizes EWas a MOF. Depending on he unc ional o m o he
weigh ing unc ion, W, he impac o la ge and small o e laps on he ene gy can a y and he MOF will
depend on W. We deno e i ¯
Φd,n[W]. Thus, he e is no a unique de ini ion o a MOF.
2.2. P ope ies o he di e en MOFs
2.2.1. Weigh ing unc ions: p- ame po en ial and Riesz s-ene gy
We conside wo main amilies o weigh ing unc ions: o W(x) = x2p he ene gy becomes a p- ame
po en ial [4,22,23]
FPp(Φd,n) = ∑
i=j|⟨ϕi|ϕj⟩|2p(5)
and o W(x) = D−s(x), wi h D(x) he ace o cho dal dis ance
D(x) = 2√1−x2,(6)
he ene gy becomes a p ojec i e Riesz s-ene gy [23,24]
REs(Φd,n) = ∑
i=j(2√1−|⟨ϕi|ϕj⟩|2)−s
.(7)
To be p ecise, D−s(x)is no pe se a alid weigh ing unc ion as i is no de ined in he limi x→1. This can
be ixed, by imposing ha D−s(1)=+∞, which would co espond o aking he limi om he le , o by
de ining he unc ion by pa s, such ha D−s(1−ϵ⩽x⩽1) = D−s(1−ϵ)wi h ϵas a ee pa ame e . In
p ac ice, we do he la e .
In he ollowing, we discuss he ela ion be ween he p- ame po en ial and he Riesz s-ene gy and he
p ope ies o he ames ha minimize hem.
2.2.2. Riesz s-ene gy and asymp o ic uni o mi y
The main in e es o he Riesz s-ene gy is ha i allows us o eason in e ms o he poppy-seed bagel
heo em, which s a es ha he minimal-ene gy a angemen o npa icles cons ained o a bounded
D-dimensional su ace and subjec o a po en ial o he o m −s, whe e is he dis ance be ween pa icles,
ends o be uni o mly dis ibu ed o la ge nwhen s⩾D[25]. To unde s and his esul , one mus hink
physically in e ms o sho - and long- ange in e ac ions. Fo la ge s, la ge o e laps (sho - ange
in e ac ions) a e weigh ed mo e hea ily in he ene gy ela i e o small o e laps (long- ange in e ac ions).
This implies ha in he minimiza ion p ocess ec o s p io i ize dec easing he ew la ge o e laps wi h hei
nea es neighbo s be o e ying o minimize a la ge numbe o small o e laps wi h dis an neighbo s. The
esul ing a angemen is uni o m because each ec o is mos ly conce ned wi h being apa om i s nea es
neighbo s and almos insensi i e o he global a angemen o ec o s. Fo small s, he opposi e is ue,
ec o s p io i ize maximizing he numbe o ec o s ha hey ha e a small o e lap wi h, which may come a
he cos o inc easing he o e lap wi h a ew nea es neighbo s, leading o c owding and lack o uni o mi y.
The Riesz s-ene gy is in a ian unde changes in he global phase o he ec o s, so we can iden i y
equi alence classes o all uni ec o s di e ing by a global phase, which essen ially de ines he complex
p ojec i e space CPd−1. A pa ame e iza ion o CPd−1in eal coo dina es o ms a mani old o dimension
2(d−1). Following he poppy-seed bagel heo em, we will ha e o se s⩾2(d−1) o gene a e MOFs, ¯
Φd,n[s],
ha a e asymp o ically uni o m, in he sense o uni o mly dis ibu ed on he mani old ha pa ame izes
CPd−1, o n→∞[23].
The same easoning can be applied o he ame po en ial (1). Al hough we canno in oke he
poppy-seed bagel heo em o p oduce p ecise egimes depending on he alue o plike we ha e jus done o
he Riesz s-ene gy, he a gumen on he basis o compe ing sho - and long- ange in e ac ions emains alid.
Small alues o pgi e mo e ela i e impo ance o long- ange in e ac ions and hus a o less uni o m
ames, and ice e sa.
3
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
2.2.3. Complex p ojec i e p-designs and uni o mi y
Uni o mi y is only p ope ly de ined in he dis ibu ion sense, o n→∞. In his limi , i is use ul o iew
uni o mi y as he p ope y o a ame ha in eg a ing a unc ion o e he Haa measu e o he co esponding
mani old is he same as a e aging he unc ion o e he ame elemen s. Fo a ini e ame, an a e age o e
he ame elemen s will only p o ide an app oxima ion o he in eg al. In his con ex , a ame is said o be a
complex p ojec i e p-design i a e aging a polynomial o deg ee po less o e he ame elemen s is
equi alen o in eg a ing he polynomial o e he Haa measu e o he co esponding mani old [5]. In
layman’s e ms, ini e ames canno be uni o m, bu he ame ec o s can be su icien ly e enly dis ibu ed
on he mani old ha a e aging simple enough unc ions o e hem is equi alen o a e aging o e he ull
mani old.
The p- ame po en ial obeys he Welch bound
FPp(Φd,n)⩾n2
(d+p−1
p)−n,(8)
wi h equali y when he ame is a complex p ojec i e p-design [26,27]. Thus, he uni o mi y o a ini e ame
can be quan i ied by compu ing he la ges p o which i sa u a es he Welch bound. This also implies ha
minimize s o he p- ame po en ial end o be p ojec i e p-designs.
2.2.4. One-designs a e igh
The ame po en ial was in oduced by Bennede o and Fickus, o iginally wi h p=1, o he s udy o ini e
no malized igh ames [28]. They p o ed ha FP1has se e al con enien p ope ies: i s global minima a e
igh ames and sa u a e he Welch bound FP1(¯
Φd,n[p=1]) = n2/d−n, and all i s local minima a e
degene a e, i.e. hey a e global minima and hus igh ames and one-designs. This ells us ha we can see
igh ames as a pa icula ins ance o MOFs as de ined in p oposi ion 1. In ac , Benede o and Fickus
al eady e e o a se o ec o s ha minimize he ame po en ial FP1as maximally o hogonal. Thus, hey
equa e MOFs wi h igh ames. As a gued abo e, i is na u al o elax his equa ion and conside igh ames
as jus one case o many MOFs. Jus like FP1leads o MOFs ha a e igh , o he weigh ing unc ions a e
equally alid and endow he co esponding MOFs wi h o he use ul p ope ies, such as uni o mi y o low
cohe ence. We men ion in passing ha ano he consequence o Benede o and Fickus’s wo k is ha inding
igh ames is an easy ask. A local op imiza ion om an a bi a y ini ial condi ion su ices o cons uc igh
ames nume ically, al hough cons uc i e analy ical me hods ha e also been de ised [29,30]. O he
weigh ing unc ions do no lead o such an easily minimizable ene gy unc ion, and hus ob aining he
co esponding MOFs poses a nume ical challenge.
We a e now in he posi ion o men ion ha igh ames a e no , jus by i ue o being igh , uni o m.
This is o be expec ed since hey a e he minimize s o FP1and hus, gene ally, jus p ojec i e one-designs, he
lowes ank o uni o mi y o ini e ames. No e also ha he e mus exis a igh ame o e e y degene a e
global minima o FP1, so a igh ame is no a unique objec o a gi en dand n, whe eas we iden i y a
uni o m ame as a highly-symme ic unique a angemen , modulo o hogonal ans o ma ions. In ac , o
a gi en con igu a ion o dand n igh ames o m ei he a mani old o a disjoin union o mani olds [31].
2.2.5. Cohe ence: he p →∞and s →∞limi s
Fo s→∞, we can w i e
lim
s→∞ (REs)1/s=m1/smax
i=j
1
D(|ϕi⟩,|ϕj⟩),(9)
whe e mis he mul iplici y o he la ges o e lap. The e o e, only he maximum o e lap(s) (minimum
dis ance(s)) con ibu e o he ene gy and se s ha only di e in submaximal o e laps will be degene a e. In
his se ing, he MOF will be gi en by
¯
Φd,n[s→∞] = a gmin{Φd,n}{max
i=j|⟨ϕi|ϕj⟩|}.(10)
No e ha he same MOF is ob ained by minimizing he p- ame po en ial in he limi p→∞, he wo
weigh ing unc ions a e equi alen in his limi . Equa ion (10) indica es ha ¯
Φd,n[s→∞]is he se ha
minimizes he maximum o e lap o , equi alen ly, ha maximizes he minimum ( ace) dis ance (6) be ween
ec o s. P oblems o his so a e ypically e e ed o as packing p oblems [32]. I is also impo an o iden i y
he cohe ence µas he quan i y ha is minimized in equa ion (10)
µ(Φd,n) = max
i=j|⟨ϕi|ϕj⟩|.(11)
4
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
The minimize s o cohe ence a e called G assmannian ames [10]. Low cohe ence and ‘ igh ness’ a e dual
p ope ies in e ms o he encoding capabili ies o a ame. A igh ame allows ‘pe ec econs uc ion’ o
he encoded in o ma ion, whe eas a G assmannian ame maximizes ‘ obus ness agains e o ’ [33]. In
gene al, i is impossible o op imize bo h simul aneously. This is e iden om he ac ha each is associa ed
wi h minimizing a di e en ene gy unc ion, FP1 o igh ness and RE∞o FP∞ o low cohe ence.
Simila ly o he Welch bound o he p- ame po en ial, he e exis bounds o he cohe ence. Namely,
•Bukh–Cox bound [34]
µ(Φd,n)⩾(n−d)2
n[1+ (n−d−1)√n−d+1]−(n−d)2i n>d.(12)
•Welch–Rankin bound [26,35]
µ(Φd,n)⩾√n−d
d(n−1)i n>d.(13)
•O hoplex bound [36]
µ(Φd,n)⩾1
√di n>d2.(14)
•Le ens ein bound [37]
µ(Φd,n)⩾√2n−d(d+1)
(n−d)(d+1)i n>d2.(15)
F ames ha sa u a e he maximum applicable bound a e op imal packings, also known as op imal
G assmannian ames.
2.2.6. Cohe ence and degene acy
To unde s and he geome ic p ope ies o G assmannian ames, le us discuss he pa icula case o d=2,
n=5. Because ec o s in d=2 equi alen unde global phase changes o m CP1, which is di eomo phic o
he 2-sphe e S2, hey can be isualized on he Bloch sphe e. Fu he mo e, o d=2 he ace dis ance
be ween ec o s (6) is equal o he Euclidean dis ance be ween hei co esponding poin s on he Bloch
sphe e. Fo la ge d he nice isualiza ion is los , al hough some in ui ion abou he opology o CPd−1can
s ill be ob ained [38]. Using he Bloch sphe e, we can see ha he e is a con inuum o degene a e ames ha
minimize he cohe ence, {¯
Φ2,5[p→∞]}, which a e illus a ed in igu e 1. They a e degene a e because hei
ene gy is de e mined by he la ges o e lap, which is he o e lap be ween he ec o s a he Equa o o he
Bloch sphe e and he ec o s a he No h and Sou h poles. The h ee ec o s a ound he Equa o can be
edis ibu ed be ween o ming igh angles be ween hemsel es, and hus hei o e laps among hemsel es
being equal o hei o e lap wi h he No h- and Sou h-pole ec o s, and o ming an equila e al iangle.
Howe e , his edis ibu ion only dec eases he submaximal o e laps, and does no lowe he cohe ence.
Al e na i ely, one may e alua e he ene gy o hese se s unde , e.g. he 6- ame po en ial. Because o p=6 all
o e laps con ibu e o he ene gy, he degene acy is b oken and he se in which he ec o s a he Equa o
o m an equila e al iangle is lowes in 6- ame po en ial. In ac , his se is p ecisely he se ha minimizes
he 6- ame po en ial and is he e o e he MOF o his weigh ing unc ion, ¯
Φ2,5[p=6]. No e ha , as we jus
explained, ¯
Φ2,5[p=6]∈{¯
Φ2,5[p→∞]}. Because he Bloch sphe e g an s us a g aphical in e p e a ion o he
di e en se s, i is easy o see in igu e 1 ha he ec o s in ¯
Φ2,5[p=6]a e mo e uni o mly dis ibu ed on he
sphe e han he ec o s o he o he se s in {¯
Φ2,5[p→∞]}. The phenomenon ha we jus desc ibed o d=2
and n=5 is also obse ed o o he alues o dand nbecause i is a consequence o he ac ha he
cohe ence dis ega ds submaximal o e laps (o he examples appea in sec ion 5).
No e ha ou a gumen in he p e ious pa ag aph is cen e ed on he appea ance o sys ema ic
degene acies in he cohe ence. We a e dis ega ding degene acies a ising om he o hogonal symme y
in insic o MOFs.
2.2.7. Uni e sal MOFs
We ha e iden i ied igh ness and low cohe ence as associa ed wi h di e en weigh ing unc ions and hus
gene ally incompa ible in a MOF. We ha e also discussed he concep o complex p ojec i e p-designs and
he uni o mi y o ames. Uni o mi y is o be expec ed o MOFs ha a e he minimize s o an ene gy ha
5

Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
Figu e 1. Degene acy o he cohe ence. Bloch-sphe e ep esen a ion ( op) and o e laps among ec o s (bo om) o he di e en
G assmannian ames o d=2 and n=5. Column (a) shows he dis ibu ion o ec o s ob ained by sub ac ing one ec o om
he equa o om ¯
Φ2,6[s→ ∞]. The ames depic ed in columns (b) and (c) a e ob ained by con inuously sp eading he
emaining h ee ec o s a he equa o un il hey o m an equila e al iangle. No e ha , o each jin all h ee panels, he e a e i e
do s in he plo , each co esponding o he o e lap o ec o jwi h e e y ec o in he ame (including i sel ). Visually, he e
appea s o be only h ee do s o each jbecause he e a e only h ee dis inc o e laps and so h ee o he do s coincide pe ec ly.
a o s sho - ange in e ac ions, wi hou ocusing only on nea es -neighbo in e ac ions like cohe ence does.
Thus nei he igh no G assmannian ames a e, gene ally, maximally uni o m gi en hei size. The e can
exis , howe e , o ce ain alues o dand n, uni e sal MOFs, i.e. ames ha a e simul aneously igh ,
minimally cohe en and maximally uni o m gi en hei size. ETFs a e he mos p ominen example [10].
ETFs a e igh ames whe e e e y ec o has he same o e lap wi h e e y o he ec o in he ame. Thei
exis ence is es ic ed o n⩽d2[39]. ETFs wi h n=d2a e known in quan um mechanics as SIC-POVMs.
The cha ac e is ics ha gi e hem hei name make SIC-POVMs use ul o pe ec econs uc ion o inpu
s a es om measu emen da a [4–9]. Thei exis ence o any dimension, dwas conjec u ed by Zaune [40].
SIC-POVMs a e also complex p ojec i e wo-designs [4].
The ac ha no ETFs exis o n>d2 ells us ha d2ma ks he h eshold beyond which he ame ec o s
s a o c owd he ec o space. Since he ame canno be equiangled while minimizing he ene gy, some
ec o s will necessa ily ha e bo h sho - and long- ange neighbo s. This gi es ise o he dynamics a o ded
by he compe i ion be ween sho - and long- ange in e ac ions ha we discussed be o e. I is o n>d2 ha
igh ness, low-cohe ence and uni o mi y s a o be incompa ible. Ne e heless, he e do exis highly
symme ic ames o n>d2. A no able example a e se s o MUB [12,13]. A pai o o hono mal bases
{|ϕi⟩}d
i=1,{|ψj⟩}d
j=1o Cda e mu ually unbiased i
|⟨ϕi|ψj⟩|2=1
d.(16)
The maximum numbe o MUB M(d)obeys M(d)⩽d+1, wi h equali y only i dis an in ege powe o a
p ime numbe . Thus, i dis an in ege powe o a p ime i is possible o cons uc highly symme ic ames
o d(d+1) ec o s, by combining he d+1 MUB, whe e e e y ec o has only nea es -neighbo s wi h
o e lap 1/√d( he ec o s o he o he bases) and nex -nea es -neighbo s wi h null o e lap ( he o he ec o s
o he same basis). Despi e no being ETFs, ames buil om d+1 se s o MUB a e G assmannian and igh
[17] and complex p ojec i e wo-designs [27].
3. Rela ed nume ical op imiza ion p oblems
F om he de ini ion o MOFs one can see ha inding hem cons i u es a mul i a ia e minimiza ion p oblem.
Finding he global minima o a non-con ex mul i a ia e unc ion is a complex ask, wi hou gua an ee o
success in mos cases. A no able excep ion is he inding o igh ames, FP1, whe e he local minima a e all
degene a e and hus global minima a e commonplace [28]. In he emaining cases, inding MOFs can be
ackled wi h nume ical op imiza ion. In his sec ion, we discuss he ela ion be ween he p oblem o inding
MOFs and o he es ablished nume ical p oblems.
6
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
3.1. The Thomson p oblem
The Thomson p oblem aims o de e mine he minimum ene gy a angemen o Nelec ons cons ained o
he su ace o he uni sphe e and subjec o elec os a ic epulsion [16,41]. Thus, i ij =∥ i− j∥is he
dis ance be ween each pai o elec ons, he Thomson p oblem o Nelec ons is equi alen o inding he
a angemen { i|∥ i∥=1}N
i=1 ha minimizes
E=∑
i=j
1
ij
.(17)
Iden i ying he elec on coo dina es on he uni sphe e wi h he coo dina es o complex ec o s o d=2 on
he Bloch sphe e, i is clea ha he Thomson p oblem is equi alen o he p oblem o inding MOFs in d=2
by minimizing he Riesz one-ene gy.
The Thomson p oblem is a challenging nume ical minimiza ion p oblem, as he numbe o local minima
is obse ed o g ow exponen ially wi h n. In ac , i has been used as a benchma k o global op imiza ion
algo i hms. The e is ag eemen be ween all nume ical and heo e ical me hods on wha is he global
minimum o a angemen s o n⪅100 and pu a i ely global minima ha e also been ob ained o n⪅1000
[42,43]. O he di e en me hods employed, a gene ic algo i hm combined wi h local op imiza ion a each
gene a ion has p o en pa icula ly e ec i e, eaching a angemen s o n⩽200 [20]. The gene ic algo i hm
ha we p esen in he nex sec ion can be seen as a gene aliza ion o he gene ic algo i hm employed on he
Thomson p oblem [44].
3.2. Op imal packings in complex p ojec i e space
The p oblem o inding op imal packings in complex p ojec i e space consis s o inding a se o lines
h ough he o igin o Cd ha a e geome ically as sp ead apa as possible, i.e. wi h maximum angle
dis ances be ween hem. In p ac ice, he lines a e ep esen ed by uni ec o s, such ha a se is ep esen ed by
a uni no m ame and he geome ic sp ead is quan i ied by he cohe ence (11) o he ame. Since he
cohe ence is in a ian unde changes in he global phase o he ec o s, we can iden i y equi alence classes
o all uni ec o s di e ing by a global phase, so he se o lines in Cdis ul ima ely equi alen o a se o
elemen s o complex p ojec i e space CPd−1. Finding op imal packings in complex p ojec i e space is hus
equi alen o inding MOFs o FP∞o RE∞, i.e. G assmannian ames.
The e exis a numbe o heo e ical and nume ical esul s on G assmannian ames, which a e nea ly
summa ized in he e iew by Jaspe e al [17]. In he e iew, he au ho s ad e ise a websi e by he name o
Game o Sloanes (in hono o Neil Sloane, who hos s a websi e wi h he bes known packings in Euclidean
space) ha hos s an open compe i ion o ind pu a i ely op imal esul s o packings in complex p ojec i e
space [45].
The nume ical me hod used o ob ain he bes known packings lis ed on he Game o Sloanes consis s on
sequen ially applying wo algo i hms: local op imiza ion on he G assmannian mani old [46,47] ollowed by
al e na ing p ojec ion [48]. As i s name implies, he algo i hm based on local op imiza ion on he
G assmannian mani old is speci ic o his p oblem. On he con a y, al e na ing p ojec ion is a he gene al
and can be used in a a ie y o se ings. In any case i does no wo k ha well on i s own, mainly because i
bene i s om an adequa e ini ial condi ion, like he one ou pu ed by he local op imiza ion. Fo his eason,
i is in e es ing o see how a comple ely gene al app oach like a gene ic algo i hm a es agains hese
specialized echniques. To his ega d, i mus be no ed ha al hough he sea ch o a G assmannian ame is
oo ed in he minimiza ion o he cohe ence (11), in p ac ice, o ob ain a smoo h unc ion, he cohe ence is
de ined as he limi
µ(Φ) ≡lim
p→∞ (FPp(Φ))1
2p(18)
Then, a ela i ely small alue o pis used ini ially, and se e al local op imiza ions a e pe o med sequen ially.
A e each local op imiza ion, he alue o pis inc eased and a new local op imiza ion begins. This p ocess is
epea ed un il con e gence is eached [47]. Due o he ini e alues o pused in p ac ice, he ac ual nume ical
minimiza ion is equi alen o he p oblem o inding MOFs o he p- ame po en ial wi h ini e p. Howe e ,
in his case, pac s as a nume ical hype pa ame e ha is uned o op imize esul s wi h espec o a igu e o
me i : he cohe ence (11).
3.3. Op imal e e ence s a es in quan um classi ie s
The concep o maximally o hogonal s a es is in oduced in he con ex o a single-qubi classi ie o e e o
a se o h ee o mo e s a es in he Bloch sphe e ha would op imally se e as e e ence s a es o each class in
he classi ica ion p ocess [18]. In an n-pa i e classi ica ion one would ha e o selec ns a es o he qubi as
7
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
e e ence s a es, assigning each o hem o one o he classes. Then, he aining p ocess would s i e o lea n
o assign o each da a en y a s a e on he Bloch sphe e ha was closes , i.e. wi h he la ges o e lap, o he
e e ence s a e o he co esponding class. This p ocess is acili a ed i he e e ence s a es a e e enly
dis ibu ed in he s a e space o he qubi . This is so ha he basin o s a es ha a e closes o a gi en e e ence
s a e is o equal size o all e e ence s a es, p e en ing de aul biases in he classi ie ha would ha e o be
o e come du ing he lea ning p ocess. In sec ion 2we discussed ha uni o mi y is a p ope y only o some
MOFs ha a e he minimize s o an adequa e ene gy unc ion. Howe e , as we discussed in ha same
sec ion, o ames wi h ew elemen s, he e exis uni e sal MOFs ha achie e igh ness, minimal cohe ence
and uni o mi y. Consequen ly, in a qubi , and o small se s, coming up wi h se s o maximally o hogonal
s a es is ela i ely s aigh o wa d, wi h some imagina ion i can e en be done g aphically. This is he case in
e e ence [18] whe e only se s o up o six maximally o hogonal s a es a e conside ed and he challenge o
ob aining la ge se s is no discussed. The idea is u he de eloped in e e ence [19], whe e hey s udy a
single-qudi classi ie as he na u al gene aliza ion o he single-qubi classi ie . In o de o gene a e la ge se s
o op imal e e ence s a es o a qudi i was necessa y o o malize he concep o maximal o hogonali y,
whe e he nuanced dis inc ion be ween maximum o hogonali y and maximum uni o mi y becomes
appa en , and o de elop he gene ic algo i hm ha we p esen in his pape . The equi alence be ween
maximally o hogonal s a es o a qudi and MOFs is clea om he ac ha he coo dina es o all possible
no malized pu e s a es o a d-dimensional qudi o m he complex p ojec i e space CPd−1.
4. The gene ic algo i hm
A gene ic algo i hm is a gene al op imiza ion me hod inspi ed by Da winian na u al selec ion. A popula ion
o candida e solu ions o he op imiza ion p oblem is e ol ed ac oss ime in successi e gene a ions. A each
gene a ion, he i es indi iduals among he popula ion a e selec ed and h ough ecombina ion, mu a ion
and su i al, hey gene a e he nex popula ion. The p ocess is i e a ed un il con e gence is eached. The
p ocess o ecombina ion a emp s o s ochas ically combine he good cha ac e is ics o one indi idual wi h
hose o ano he , gi ing ise o an o e -all be e indi idual. The p ocess o mu a ion in oduces noise,
allowing a andom explo a ion o he s a e space. The p ocess o su i al in oduces de e minism in he
algo i hm. Because he i es indi iduals a e allowed o su i e, i ensu es a s eady low owa d be e
solu ions, wi hou s ochas ic se backs. The p oblem o inding MOFs lends i sel o be implemen ed as a
gene ic algo i hm because he ecombina ion p ocess can be implemen ed s aigh o wa dly by combining
subse s o ec o s o he wo pa en ames.
In ou pa icula se ing, o a gi en dand n, a popula ion, Pd,n={Φk
d,n}N
k=1, will be a collec ion o
pu a i e MOFs, he indi iduals, wi h N he size o he popula ion. The ec o s in each ame a e no malized,
⟨ϕk
i|ϕk
i⟩=1∀i,k, and encoded as a coo dina e ec o in he o hono mal basis o Cd. To elimina e
degene acies due o symme y, we ix he i s ec o o each ame o be |ϕk
1⟩= (1,0,...,0)∀k. In all o he
ec o s, he global phase is elimina ed |ϕk
i=1⟩= (zk,1,...,zk,d), wi h zk,1∈Rand zk,l=1∈C.
Gene ic algo i hms a e ypically o mula ed as a maximiza ion p oblem, whe e he i ness Fis he
unc ion o maximize. He e, we de ine i ness as he nega i e ene gy Riesz s-ene gy: F(Φ) = −REs(Φ). A
each gene a ion, we so he indi iduals by i ness and selec a numbe o hem o se e as pa en s o he nex
gene a ion. I is common in gene ic algo i hms o selec he pa en s s ochas ically, wi h a p obabili y
p opo ional o hei i ness, o delay he con e gence p ocess and e icien ly explo e he pa ame e space.
He e we employ a udimen a y de e minis ic al e na i e. We ix a i ness gap, ∆F, and selec ou pa en s in
dec easing o de o i ness, s a ing om he i es indi idual and ensu ing ha he nex pa en has a i ness
a leas ∆Fsmalle han he p e ious pa en . The i ness gap is de e mined as a ac ion o he i ness o he
i es indi idual in ha gene a ion ∆F=αdi Fmax. The di e si y a io, αdi , is a hype pa ame e o he
algo i hm ha we se o αdi =0.1.
Recombina ion is implemen ed as a andom c osso e o ec o s om wo indi iduals. A i ial
ecombina ion could consis on gene a ing a andom in ege ibe ween 1 and n o c ea e a child wi h he i
i s ec o s o pa en A and he las n−i ec o s o m pa en B. Howe e , he ec o s wi hin a ame a e, a
p io i, no so ed in any pa icula ashion, i.e. hei o de does no e lec any geome ical s uc u e in he
mani old ha pa ame izes he ec o s. The e o e, he subse o ec o s inhe i ed om each pa en would be
andom, consis ing o ec o s om pa en A ha could be a bi a ily close o ec o s om pa en B, and hus
h ow away mos o he o hogonali y gained o e he cou se o he op imiza ion p ocess. Because o his, a
i ial ecombina ion like he one jus desc ibed would, mos likely, c ea e a child wi h a i ness much lowe
han ha o i s pa en s. To p e en his, we so he ec o s o each pa en by dis ance o he i s ec o o
each se (which is ixed and he same o all se s). A e he so , he o de o he ec o s in a se does e lec
some s uc u al p ope ies. Namely, he i s ec o o he se ma ks a e e ence poin in he mani old ha is
common o all se s. A e so ing, we apply he andom ecombina ion by d awing he i s i ec o s om
8
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
pa en A, which a e expec ed o be ela i ely close o he e e ence poin , and he las n−i ec o s om
pa en B, which a e expec ed o be ela i ely a om he e e ence poin . To be e unde s and his p ocess,
le us conside he case d=2. Now he mani old is he Bloch sphe e and he e e ence poin is gi en by he
ec o (1, 0), which ma ks he no h pole. Selec ing he i s i ec o s om pa en A co esponds o selec ing
ec o s co e ing a sphe ical cap a ound he no h pole wi h an app oxima e heigh de e mined by he a io
i/n. Then, selec ing he n−i ec o s om pa en B co esponds o selec ing ec o s co e ing a sphe ical cap
a ound he sou h pole wi h an app oxima e heigh de e mined by he a io (n−i)/n. Al hough no hing
gua an ees ha he wo caps do no o e lap o , con a ily, lea e blank a sphe ical s ip in be ween hem, a
leas he geome ical s uc u e wi hin each cap is p ese ed and he i ness o he child will only depend on
he seam be ween he wo caps. A simila ecombina ion s a egy is used in e e ence [20].
Following he same easoning, we implemen mu a ions by so ing a se by dis ance o he i s ec o ,
gene a ing a andom in ege ibe ween 1 and nand a andom phase θ∈[0,2π)and o a ing he las n−i
ec o s o he so ed se by he ela i e phase, i.e. (z1,z2,...,zd)→(z1,z2eiθ,...,zdeiθ). Fo d=2, his
amoun s o o a ing he las n−i ec o s o he se an angle θa ound he no h pole.
In addi ion o he usual ope a ions o a gene ic algo i hm: selec ion o he i es , ecombina ion, and
mu a ion, we also implemen local op imiza ion. A each i e a ion, a e he ou pa en s a e selec ed and
be o e ecombina ion and mu a ion, hey a e each subjec ed o a local op imiza ion p ocess. To implemen
he op imiza ion, we encode a se as a (n−1)×d×2 dimensional ec o whe e he i s elemen is he eal
pa o he second elemen o he i s ec o , he second elemen is he imagina y pa o he second elemen
o he i s ec o , he hi d elemen is he eal pa o he hi d elemen o he i s ec o and so on, i.e.
Φd,n→(ℜ(z2,1),ℑ(z2,1),...,ℑ(z2,d),ℜ(z3,1),...). Due o he necessi y o en o ce no maliza ion on he
ec o s composing each se h oughou he op imiza ion p ocess, his is a case o mul i a ia e cons ained
op imiza ion which we ackle using sequen ial leas -squa es p og amming. The local op imiza ion uns un il
ei he con e gence o a maximum numbe o i e a ions, which cons i u es ano he hype pa ame e , a e
eached. F om a gene ic pe spec i e, applying a local op imiza ion o he pa en s ha is la e inhe i ed by he
nex gene a ion in oduces a Lama ckian aspec o an o he wise Da winian e olu ion p ocess. Gene ic
algo i hms based on his hyb id app oach a e ypically e med meme ic algo i hms [49].
In summa y, he algo i hm uns as ollows:
1. Selec ene gy unc ion, i.e. selec he i ness unc ion as he nega i e o he ene gy unc ion.
2. Gene a e ini ial popula ion. We gene a e Nse s o nno malized andom ec o s d awn om he Haa
dis ibu ion.
3. Selec he ou ‘ i es ’ indi iduals as he pa en s ( espec ing he selec ed i ness gap, ∆F).
4. Apply local op imiza ion o he pa en s.
5. Gene a e wel e child en by ecombining e e y pa en wi h e e y o he pa en (no e ha each pai o
pa en s ma es wice, exchanging he oles o pa en s A and B in each case).
6. Gene a e ou o he child en by mu a ing each pa en once.
7. Fo m a new popula ion wi h he ou pa en s and he six een child en and e alua e he i ness o he
i es indi idual.
8. I he maximum numbe o i e a ions o he con e gence c i e ion a e eached, hal , o he wise go o s ep
3 wi h he new popula ion.
To induce he hal ing o he algo i hm, one can ix a maximum numbe o i e a ions o a con e gence
c i e ion based on he a e o change o he i ness ac oss gene a ions.
We a bi a ily ix he numbe o pa en s o ou , he numbe o ecombina ions o wel e, he numbe o
mu a ions o ou and consequen ly he popula ion size o wen y (inhe i ed om e e ence [20]), hese
numbe s could be le as hype pa ame e s.
Mo e nume ical de ails can be ound in appendix A
5. Resul s and discussion
5.1. The Thomson p oblem
As p esen ed in sec ion 3.1, o d=2 and using he Riesz one-ene gy as he weigh ing unc ion, he p oblem
o inding MOFs is equi alen o he Thomson p oblem. The e o e, we use he Thomson p oblem as a
benchma k o he GA. Figu e 2(a) shows he ela i e ene gy di e ence, ϵ el =|REGA
1−RE e
1|/|RE e
1|,
be ween he ames p oduced by he GA o he Thomson p oblem and he bes known a angemen s
epo ed in e e ence [50] o con igu a ions o up o n=100 poin s. The GA p oduces op imal o
nea -op imal ames wi h a ela i e e o unde ∼0.1% o n⩽100.
9
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
Table A1. Run imes o he gene ic algo i hm o di e en con igu a ions. The numbe o gene a ions was capped a 15.
d n Run ime Gene a ions Time/gene a ion
2 6 0.6s 5 0.12s
2 24 99.9s 15 6.6s
4 16 28.7s 7 4.1s
4 40 9min6s 9 1min7s
Figu e 6. Dependence o he mesh no m wi h he numbe o Haa ec o s NH o d=4 and di e en alues o n.
A.2. Compu ing he mesh no m and he sha e o he ec o space
The mesh no m and he sha e o he ec o space can only be compu ed app oxima ely by pe o ming a
ini e sampling o NH ec o s om he Haa dis ibu ion o each dimension d. In igu e 6we show how he
mesh no m depends on NH o di e en con igu a ions. We ind ha a alue o NH=107is su icien o
esol e changes in he mesh no m ⩾10−2, which is su icien o compa e, e.g. he ames ob ained by
op imizing di e en me ics in igu e 4.
Appendix B. Visualizing uni o mi y: sha e o he ec o space
In sec ion 5.2.1 we de ined he no malized sha e o he ec o space βjas a way o isualize he uni o mi y o
a ame. By de ini ion, he sha e and he mesh no m a e ela ed. The mesh no m measu es he adius o he
la ges gap le in he ec o space by ec o s o he ame. The sha e measu es he po ion o he ec o space
ha is closes o each ec o o he ame. The ela ion s ems om he ac ha he ec o s o he ame
su ounding he la ges gap will ha e a la ge sha e. The e o e, he mesh no m is posi i ely co ela ed wi h
he la ges alues o he sha e. This ela ion is simila o he ela ion be ween cohe ence and o e laps, whe e
cohe ence is he maximum o e lap. No e also ha he sha e o he ec o space is he me ic o op imize
when looking o op imal e e ence s a es o a quan um classi ie (see sec ion 3.3). A non-uni o m
dis ibu ion o he sha es is associa ed wi h de aul biases o he quan um classi ie .
In igu es 7and 8we isualize he uni o mi y o he ames ob ained wi h he GA o d=4. Figu e 7is
complemen a y o igu e 3. In igu e 7we show he ames ob ained o d=4 and n∈[4,43]. Each panel
displays he sha e o he ec o space βjo each ec o |ϕk⟩∈ ¯
Φ4,n[s=8]. In addi ion, he s anda d de ia ion
o he sha es is indica ed wi hin each panel. Uni o m ames ha e βj≈1 o all ec o s and a small s anda d
de ia ion. I is in e es ing o no e ha all ames up o he SIC-POVM o n=16 a e ela i ely uni o m.
App eciable non-uni o mi ies only appea beyond his poin . Ne e heless, highly symme ic ames such as
he se o MUB o he h ee-design a n=40 a e again ela i ely uni o m.
Figu e 8is complemen a y o igu e 4. In igu e 8we plo he sha e o he ec o space o h ee di e en
con igu a ions, o d=4 and n=23,40,43. Fo each con igu a ion we plo he ame wi h lowes cohe ence,
highes mesh no m and lowes looseness (closes o being igh ). Fo n=40, gi en he high symme y o he
h ee-design, he e a e ba ely any di e ences be ween he h ee cases. In con as , o n=23 and n=43 he e
a e clea di e ences in he sha es be ween he ames ha op imize each me ic. F ames ha minimize
cohe ence end o display sha es g ouped in wo o mo e dis inc alues. On he o he hand, ames ha
16

Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
Figu e 7. Maximally o hogonal ames o d=4 and s=8. Each panel shows he no malized sha e o he ec o space βj(21) o
each ec o he bes MOF ou o en uns o he GA o a ce ain size n. The ho izon al black line is a guide o he eye ma king he
alue 1. The panel i le indica es whe he i is a high-symme y ame such as an ETF, maximal se o MUB o SIC-POVM. The
s anda d de ia ion o he sha e σβis indica ed wi hin each panel.
Figu e 8. Gene ic algo i hm s ochas ic esul s. No malized sha e o he ec o space βj(21) o he maximally o hogonal ames
o d=4 and s=8 and n=23,40,43. Each column shows he esul o a di e en un o he GA ha op imizes a di e en me ic.
The ho izon al black line is a guide o he eye ma king he alue 1. The s anda d de ia ion o he sha e σβis indica ed wi hin each
panel.
17
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
maximize he mesh no m o minimize looseness ha e mo e e enly dis ibu ed sha es. This is e lec ed in he
s anda d de ia ion o he sha e, which is highe o he ames ha minimize cohe ence.
We no e ha o se e al con igu a ions in igu es 7and 8 he dis ibu ion o sha es is mos ly concen a ed
a ound 1, wi h one o a ew ou lie s wi h a alue signi ican ly di e en han 1. In hese cases, he s anda d
de ia ion may no e lec he ac ual uni o mi y o he ame and i could be in e es ing o inco po a e
es ima o s o ku osis o mul imodali y o he dis ibu ion o sha es.
ORCID iDs
Sebas ián Roca-Je a 0000-0001-5948-4263
Juan Román-Roche 0000-0003-2995-6615
Re e ences
[1] Du in R J and Schae e A C 1952 A class o nonha monic Fou ie se ies T ans. Am. Ma h. Soc. 72 341–66
[2] Daubechies I 1992 Ten Lec u es on Wa ele s (Socie y o Indus ial and Applied Ma hema ics)
[3] G Ku yniok and P G Casazza ed 2013 Fini e F ames: Theo y and Applica ions (Bi khäuse Bos on)
[4] Renes J M, Blume-Kohou R, Sco A J and Ca es C M 2004 Symme ic in o ma ionally comple e quan um measu emen s J. Ma h.
Phys. 45 2171–80
[5] Sco A J 2006 Tigh in o ma ionally comple e quan um measu emen s J. Phys. A: Ma h. Gen. 39 13507–30
[6] Słomczy´
nski W and Szymusiak A 2015 Highly symme ic POVMs and hei in o ma ional powe Quan um In . P ocess. 15 565–606
[7] Yoshida M and Kimu a G 2022 Cons uc ion o gene al symme ic-in o ma ionally-comple e–posi i e-ope a o - alued measu es
by using a comple e o hogonal basis Phys. Re . A 106 022408
[8] Gu ¸˘
a M, Kahn J, Kueng R and T opp J A 2020 Fas s a e omog aphy wi h op imal e o bounds J. Phys. A: Ma h. Theo . 53 204001
[9] Jameson C, Qin Z, Golda A, Wakin M B, Zhu Z and Gong Z 2024 Op imal quan um s a e omog aphy wi h local in o ma ionally
comple e measu emen s (a Xi :2408.07115) [quan -ph]
[10] S ohme T and Hea h R W 2003 G assmannian ames wi h applica ions o coding and communica ion Appl. Compu . Ha mon.
Anal. 14 257
[11] Fuchs C, Hoang M and S acey B 2017 The SIC ques ion: his o y and s a e o play Axioms 621
[12] Schwinge J 1960 Uni a y ope a o bases P oc. Na l. Acad. Sci. 46 570–9
[13] I ono ic I D 1981 Geome ical desc ip ion o quan al s a e de e mina ion J. Phys. A: Ma h. Gen. 14 3241–5
[14] Delsa e P, Goe hals J M and Seidel J J 1977 Sphe ical codes and designs Geom. Dedica a 6363
[15] Hogga S 1982 -designs in p ojec i e spaces Eu . J. Comb. 3233–54
[16] Thomson J 1904 On he s uc u e o he a om: an in es iga ion o he s abili y and pe iods o oscilla ion o a numbe o co puscles
a anged a equal in e als a ound he ci cum e ence o a ci cle; wi h applica ion o he esul s o he heo y o a omic s uc u e
Lond. Edinb. Dubl. Phil. Mag. 7237
[17] Jaspe J, King E J and Mixon D G 2019 Game o Sloanes: bes known packings in complex p ojec i e space P oc. SPIE
11138 111381E
[18] Pé ez-Salinas A, Ce e a-Lie a A, Gil-Fus e E and La o e J I 2020 Da a e-uploading o a uni e sal quan um classi ie Quan um
4226
[19] Roca-Je a S, Román-Roche J and Zueco D 2024 Qudi machine lea ning Mach. Lea n.: Sci. Technol. 5015057
[20] Mo is J R, Dea en D M and Ho K M 1996 Gene ic-algo i hm ene gy minimiza ion o poin cha ges on a sphe e Phys. Re . B
53 R1740
[21] Wald on S F 2018 An In oduc ion o Fini e Tigh F ames (Sp inge )
[22] Bilyk D, Glazy in A, Ma zke R, Pa k J and Vlasiuk O 2022 Op imal measu es o p- ame ene gies on sphe es Re . Ma . Ibe oam.
38 1129–60
[23] Chen X, Ha din D P and Sa E B 2020 On he sea ch o igh ames o low cohe ence J. Ma h. Anal. Appl. 27 2
[24] Dooho skoy A and Landko N 1972 Founda ions o Mode n Po en ial Theo y G undleh en de Ma hema ischen Wissenscha en
(Sp inge ) (a ailable a : h ps://books.google.es/books?id =ZUpxMQEACAAJ)
[25] Ha din D P and Sa E B 2004 Disc e izing mani olds ia minimum ene gy poin s No ices Ame . Ma h. Soc. 51 1186–94 (a ailable
a : h ps://www.ams.o g/jou nals/no ices/200410/200410FullIssue.pd )
[26] Welch L 1974 Lowe bounds on he maximum c oss co ela ion o signals (co esp.) IEEE T ans. In . Theo y 20 397
[27] Klappenecke A and Ro ele M 2005 Mu ually unbiased bases a e complex p ojec i e 2-designs P oc. In . Symp. on In o ma ion
Theo y, 2005. ISIT 2005. pp 1740–4
[28] Benede o J J and Fickus M 2003 Fini e no malized igh ames Ad . Compu . Ma h. 18 357–85
[29] Casazza P G, Fickus M and Mixon D G 2012 Au o- uning uni no m ames Appl. Compu . Ha mon. Anal. 32 1–15
[30] Cahill J, Fickus M, Mixon D G, Po ee M J and S awn N 2013 Cons uc ing ini e ames o a gi en spec um and se o leng hs
Appl. Compu . Ha mon. Anal. 35 52–73
[31] Dykema K and S awn N 2003 Mani old s uc u e o spaces o sphe ical igh ames (a Xi :ma h/0307367) [ma h.FA]
[32] Weai e D and As e T 2008 The Pu sui o Pe ec Packing (CRC P ess)
[33] Haas J I and Casazza P G 2017 On he s uc u es o g assmannian ames 2017 In . Con . on Sampling Theo y and Applica ions
(SampTA) (IEEE)
[34] Bukh B and Cox C 2020 Nea ly o hogonal ec o s and small an ipodal sphe ical codes Is . J. Ma h. 238 359–88
[35] Rankin R A 1955 The closes packing o sphe ical caps in n dimensions P oc. Glasg. Ma h. Assn. 2139–44
[36] Conway J H, Ha din R H and Sloane N J A 1996 Packing lines, planes, e c.: packings in G assmannian spaces Exp. Ma h. 5139
[37] Le ensh ein V 1982 Bounds on he maximal ca dinali y o a code wi h bounded modulus o he inne p oduc So ie Ma h. Dokl.
25 526–31
[38] Beng sson I and Zyczkowski K 2006 Geome y o Quan um S a es: An In oduc ion o Quan um En anglemen (Camb idge
Uni e si y P ess)
18
Mach. Lea n.: Sci. Technol. 6(2025) 035022 S Roca-Je a and J Román-Roche
[39] Sus ik M A, T opp J A, Dhillon I S and Hea h R W 2007 On he exis ence o equiangula igh ames Linea Algeb a Appl.
426 619–35
[40] Zaune G 2011 Quan um designs: ounda ions o a noncommu a i e design heo y In . J. Quan um In . 09 445–507
[41] Sa E B and Kuijlaa s A B J 1997 Dis ibu ing many poin s on a sphe e Ma h. In ell. 19 5–11
[42] Al schule E L and Pé ez-Ga ido A 2005 Global minimum o Thomson’s p oblem o cha ges on a sphe e Phys. Re . E 71 047703
[43] Wales D J and Ulke S 2006 S uc u e and dynamics o sphe ical c ys als cha ac e ized o he homson p oblem Phys. Re . B
74 212101
[44] E be T and Hockney G M 1995 Commen on ‘me hod o cons ained global op imiza ion’ Phys. Re . Le . 74 1482
[45] Jaspe J, King E J and Mixon D G Game o Sloanes (a ailable a : www.ma h.colos a e.edu/∼king/Gameo Sloanes.h ml)
[46] Edelman A, A ias T A and Smi h S T 1998 The geome y o algo i hms wi h o hogonali y cons ain s SIAM J. Ma ix Anal. Appl.
20 303–53
[47] Med a A and Da idson T N 2014 Flexible codebook design o limi ed eedback sys ems ia sequen ial smoo h op imiza ion on he
G assmannian mani old IEEE T ans. Signal P ocess. 62 1305
[48] T opp J, Dhillon I, Hea h R and S ohme T 2005 Designing s uc u ed igh ames ia an al e na ing p ojec ion me hod IEEE
T ans. In . Theo y 51 188
[49] Mosca o P 2000 On e olu ion, sea ch, op imiza ion, gene ic algo i hms and ma ial a s - owa ds meme ic algo i hms Cal ech
Concu en Compu a ion P og am
[50] Wikipedia 2024 Thomson p oblem—Wikipedia, he ee encyclopedia (a ailable a : h p://en.wikipedia.o g/w/index.
php? i le=Thomson%20p oblem&oldid=1227607089) (Accessed 20 June 2024)
[51] Sebas ián Roca-Je a J R-R 2025 (a ailable a : h ps://gi hub.com/Theo y-and-Simula ion-a -INMA/GA o MOF_public)
19