Spectral Structure of the Naive Dirac Operator on the Face-Centered Cubic Lattice

Spectral Structure of the Naive Dirac Operator

on the Face-Centered Cubic Lattice

Raghu Kulkarni

SSMTheory Group, IDrive Inc, Calabasas, CA

raghu@idrive.com

March 2026

Abstract

We construct the naive Dirac operator on the Face-Centered Cubic (FCC) lattice

(

K = 12

) in three spatial dimensions and map its complete zero-mode structure.

The spectrum is naturally stratied into three topological classes: a singlet at

, an

isolated quartet at the four

-points, and continuous nodal lines along the Brillouin

zone boundary connecting

and

. No other interior zeros were found in a

numerical search (all minimizers lie adjacent to the known zero loci); completeness

is conjectured but not proved, and this is the paper's principal open problem. Com-

puting the Jacobian determinant at each isolated zero assigns topological indices

(not

eigenvalues; there is no

in 3D)

= +1

and

= −1

(per point). A

numerical winding-number computation yields integer winding

for each



nodal line segment; with 3 nodal loops (one per spatial axis, separate connected

components of the zero set), the total boundary chirality is

boundary

= +3

, and the

Nielsen-Ninomiya sum

+1 − 4 + 3 = 0

is satised. The

-point quartet transforms

under the chiral octahedral group

∼

and decomposes as

4 → 1 ⊕ 3

; Schur's

lemma constrains any

-invariant mass matrix to split singlet and triplet, with an

explicit anisotropic Wilson term producing

∆M = 4/a

. The stratied spectrum

with distinct Wilson masses (

12/a

16/a

) and distinct chiralities across three

topological classes has no parallel on hypercubic lattices, where all

= 8

doublers

are geometrically equivalent with identical gapping scales.

Keywords:

FCC lattice, lattice fermions, fermion doubling, Nielsen-Ninomiya the-

orem, Brillouin zone

1 Introduction

On a standard hypercubic lattice in

spatial dimensions, the naive discretization of the

Dirac equation produces

fermion doublers [1, 2], all geometrically equivalent. Wilson

fermions [3] and staggered fermions [4, 5] suppress or reduce these artifacts, but at the

cost of chiral symmetry or fermion content.

We work in 3 spatial dimensions throughout. The primary application of lattice

fermionsQCDrequires 4D, but the 3D case is of independent interest for two reasons:

it is the natural setting for condensed matter applications (Weyl semimetals, frustrated

magnets, topological insulators on non-cubic lattices), and it provides a tractable testing

ground for non-hypercubic lattice constructions that may generalize to 4D. The FCC

lattice exists in any dimension, and the spectral analysis could in principle be attempted

on a 4D FCC analogue (the

root lattice, with 24 nearest neighbors and a 24-cell

Brillouin zone), though that case introduces substantially greater complexity that we do

not analyze here.

The Face-Centered Cubic (FCC) lattice oers a dierent starting point. It is the dens-

est sphere packing in 3D (the Kepler conjecture, proved by Hales [6]), has coordination

number

K = 12

, and possesses the full octahedral point group symmetry

. Gauge elds

on the Body-Centered Cubic (BCC) lattice have been studied [9], and FCC-type lattices

appear in condensed matter contexts (frustrated magnets, Weyl semimetals). However,

to our knowledge, the zero-mode structure of the naive Dirac operator on the FCC lattice

has not been systematically characterized in the lattice eld theory literature.

In this paper we classify all zero modes of the naive Dirac operator on the FCC

lattice (Section 3), compute chiralities at isolated zeros (Section 4), independently verify

the Nielsen-Ninomiya theorem via winding-number computation (Section 5), and analyze

the representation-theoretic structure of the

-point quartet (Section 6).

2 The FCC Lattice

2.1 Real space

The FCC lattice is generated by the 12 nearest-neighbor vectors:

N = {(±1, ±1, 0), (±1, 0, ±1), (0, ±1, ±1)}

(1)

in units of

a/2

, where

is the cubic lattice constant. The convex hull of these 12 points is

the cuboctahedron [7]: 12 vertices, 24 edges, 14 faces. The symmetry group is

(order

48).

2.2 Reciprocal space

The reciprocal lattice is BCC. The rst Brillouin zone (FBZ) is a truncated octahedron

[8], with high-symmetry points:

Γ = (0, 0, 0)

L = (±π/2, ±π/2, ±π/2)

(8 points;

and

−k

identied

→

4 independent)

X = (π, 0, 0), (0, π, 0), (0, 0, π)

(3 independent)

W = (π, π/2, 0)

and permutations (12 points, 6 independent) (2)

3 The Naive Dirac Operator

3.1 Construction

We work in the integer basis dened by Eq. (1), where the neighbor vectors

have integer

components (in units of

a/2

). The momenta

are correspondingly dimensionless, with

the Brillouin zone boundaries at multiples of

In 3-dimensional Euclidean space (the signature relevant for lattice eld theory in

imaginary time), the minimal spinor representation is

2 × 2

(Pauli matrices

). There

(

)

Brillouin zone:

(singlet),

(quartet),

(nodal lines)

Figure 1: Projection of the FCC Brillouin zone (truncated octahedron) showing zero-

mode locations:

(green), 4

-points (red squares), and



nodal lines (blue dashed).

is no

operator in 3D Euclidean space; the chirality we compute is the topological

index

= sgn(det J)

the local degree of the map

f : T

→ R

not a

eigenvalue.

Readers familiar with 4D lattice QCD should note this distinction: in 4D, chirality refers

to the eigenvalue of

= γ

; in 3D, no such operator exists. The topological

index classies zero modes by the orientation of the

-eld in their neighborhood. The

relevant no-go theorem in

spatial dimensions states that the total degree of the map

f = f /|f | : T

\ f

−1

(0) → S

D−1

must vanish [1, 2, 10]. For

D = 3

, this constrains

= 0

, where

= sgn(det J)

at isolated zeros and

equals the transverse winding

number for nodal-line components. The extension from isolated zeros to codimension-2

nodal lines is standard in the condensed matter literature on Weyl nodal-line semimetals

(see e.g. [11] for the topological classication); in the lattice eld theory context, the

relevant mathematical fact is that the degree of

decomposes as a sum over all connected

components of

−1

(0)

, with each component contributing its linking number.

We dene the kinetic vector eld:

(k) =

n∈N

sin(k · n), µ = 1, 2, 3

(3)

where

k · n =

. The naive Dirac operator is:

D(k) = iγ · f (k) = i

µ=1

(k)

(4)

where

are

2 × 2

Pauli matrices. Zero modes occur wherever

|f (k)| = 0

3.2 Zero-mode classication

Numerical minimization of

|f (k)|

across the full Brillouin zone, combined with analytic

verication at high-symmetry points, yields three topological classes.

Proposition 1

(Zero-mode stratication)

The naive Dirac operator on the FCC lattice

has zero modes at:

Γ = (0, 0, 0)

: a single isolated zero (the fundamental mode).

L = (±π/2, ±π/2, ±π/2)

: four isolated zeros (with antipodal identication).



boundary lines: continuous families of zeros connecting

-points to

-points

on the zone boundary.

Proof.

sin(k · n) = 0

for all

, giving a zero trivially.

L = (π/2, π/2, π/2)

: the 12 scalar products

L · n

take values in

{0, ±π}

. Since

sin(0) = sin(±π) = 0

, every term vanishes. The same holds at all 4 independent

-points.

For the boundary line

k(t) = (π, t, 0)

connecting

X = (π, 0, 0)

W = (π, π/2, 0)

the FCC neighbor set has inversion symmetry (

n ∈ N ⇒ −n ∈ N

). When

= π

, the

terms from

and

−n

cancel:

sin(k ·n) + sin(k ·(−n)) = 0

because

sin(π + x) = −sin(x)

Analogous lines exist for all symmetry-equivalent



paths.

Completeness: the function

|f (k)|

is a sum of squares of trigonometric polynomials

and therefore real-analytic on the torus

. Its zero set is an analytic variety whose

isolated components are rigid. A

grid search over the full Brillouin zone, proled as

a function of exclusion distance

from all known zero loci, gives:

d min |f |

Minimizer location Dist. to nearest zero

0.10 4.5 × 10

−4

(0, −0.94π, −0.94π) 0.20

0.20 0.005 (−0.91π, −0.91π, 0) 0.30

0.30 0.027 (−0.88π, 0.88π, 0) 0.39

0.40 0.099 (0, −0.84π, −0.84π) 0.49

0.50 0.28 (−0.81π, −0.81π, 0) 0.59

In every case, the minimizer sits at distance

≈ d

from the nearest known zero (the



boundary nodal lines)it is the closest allowed point to the known zero set, not a

hidden interior minimum. The minimum

|f |

increases monotonically with

. Under

symmetry, any additional isolated zero generates an orbit of

≥ 6

points; the grid spacing

(

∆k = 0.098

) would detect at least one orbit member.

We conjecture the classication is complete (as stated in the abstract, this is the

paper's principal open problem). The least-constrained region is the BZ interior with

| ∈ (0.3π, 0.7π)

, where

|f |

≥ 0.027

. We note that the plateau argument does not by

itself rule out a hidden minimum at

d < 0.1

; a rigorous exclusion would require either

interval arithmetic on the explicit trigonometric form of Eq. (3), or a Morse-theoretic

argument bounding the number of critical points of

|f |

Unlike the hypercubic lattice, where all

= 8

doublers are geometrically equivalent

zone corners, the FCC spectrum is naturally stratied into three qualitatively distinct

classes.

Point Coordinates

|f |

Wilson

Type

Γ (0, 0, 0)

0 0 Isolated singlet

L (

)

12/a

Isolated (

×4

)

X (π, 0, 0)

16/a

Nodal line

W (π,

, 0)

16/a

Nodal line

test

(

3π

, 0) 1.37 ∼ 15.7/a

Not a zero

†

Table 1: Zero modes of the naive Dirac operator on the FCC lattice.

†

This test point

veries that the numerical search correctly distinguishes zeros from near-zeros; its

|f |

1.37

is far above zero.

Wilson

is the mass from a Wilson term

(r/a)

(1 − cos k · n)

with

r = 1

. At

: all cosines equal 1, so

M = 0

. At

: six neighbors give

cos = 1

, six

give

cos = −1

, so

M = (1/a)(6 × 0 + 6 × 2) = 12/a

. At

X = (π, 0, 0)

: eight neighbors

give

cos = −1

, four give

cos = 1

, so

M = (1/a)(8 × 2 + 4 × 0) = 16/a

4 Topological Index at Isolated Zeros

The topological index

at an isolated zero

is the sign of the Jacobian determinant:

µν

(k) =

∂f

∂k

n∈N

cos(k · n)

(5)

Proposition 2

(

-point index)

Γ = (0, 0, 0)

µν

(Γ) =

µν

(6)

This is positive denite with eigenvalues

{8/a, 8/a, 8/a}

and

det J > 0

, giving

= +1

Proof.

k = 0

cos(k · n) = 1

for all

. The sum

over the 12 FCC neighbors

evaluates to

8δ

µν

(each o-diagonal entry cancels by symmetry; each diagonal entry sums

to 8).

Proposition 3

(

-point index)

L = (π/2, π/2, π/2)

, the Jacobian matrix is:

µν

(L) =





0 −4 −4

−4 0 −4

−4 −4 0





(7)

with eigenvalues

{+4, +4, −8}/a

and

det J = −128/a

< 0

, giving

= −1

Proof.

, the 12 scalar products

L · n

take values in

{0, ±π}

. Six neighbors (with

L · n = 0

) contribute

cos = +1

; six (with

L · n = ±π

) contribute

cos = −1

The diagonal entries vanish: for each

, the

contributions from

cos = +1

and

cos = −1

neighbors cancel exactly. The o-diagonal entry

= (1/a)

cos(L ·n)

the 4 neighbors with

= 0

are

(±1, ±1, 0)

, all with

L ·n = ±π

(hence

cos = −1

), and

= ±1

. The products

cos(L · n)

sum to

−4/a

. By symmetry, all o-diagonal

entries equal

−4/a

The matrix

J = −(4/a)(E − I)

where

is the all-ones matrix. Its eigenvalues are:

−(4/a)(3 − 1) = −8/a

(eigenvector

(1, 1, 1)/

√

) and

−(4/a)(0 − 1) = +4/a

(two-fold

degenerate). The determinant is

(+4)(+4)(−8)/a

= −128/a

< 0

With 4 independent

-points:

total

= 4 ×(−1) = −4

5 Boundary Chirality and Nielsen-Ninomiya

5.1 Winding-number computation

To determine

boundary

independently of the Nielsen-Ninomiya theorem, we compute the

winding number of the transverse kinetic eld around each



nodal line segment.

For a nodal line along the

direction (e.g.,

k = (π, t, 0)

), we parametrize a small

circle of radius

ϵ = 0.01

in the transverse

, k

)

plane. Since the nodal line runs along

, the component

is longitudinal and does not enter the transverse winding; only the

transverse components

, f

)

are relevant. We compute these at 2000 points and extract

the winding:

w =

2π

d arg(f

+ if

)

(8)

Segment Transverse plane Winding

(π, t, 0) (k

, k

) +1.0000

(consistent with

)

(π, 0, t) (k

, k

) +1.0000

(consistent with

)

(t, π, 0) (k

, k

) +1.0000

(consistent with

)

(0, π, t) (k

, k

) +0.9971

(consistent with

)

(t, 0, π) (k

, k

) +0.9971

(consistent with

)

(0, t, π) (k

, k

) +0.9971

(consistent with

)

Table 2: Winding number of the transverse kinetic eld around each



nodal line

segment, computed at the segment midpoint with

ϵ = 0.01

and 2000 angular samples. All

six are consistent with integer winding

. The rst three deviate by

< 10

−4

while the

last three deviate by

∼ 0.003

; this reects a dierence in convergence rate. The segments

centered at

(π, . . .)

have their transverse circle well inside the BZ, while those centered

(0, π, . . .)

intersect the BZ boundary identication surface, where

has a kink in its

periodic extension. This kink is a coordinate artifact (not a discontinuity of

) but

slows angular convergence. At

N = 4000

0.9985

; at

N = 8000

0.9993

. The

1/N

scaling

conrms discretization error.

5.2 Loop counting

The nodal lines form closed circuits around each square face of the truncated octahedron

BZ. The square face at

= π

has 4



segments forming one closed loop that

lies entirely within the plane

= π

. There are 6 square faces, but opposite faces are

identied by BZ periodicity: the reciprocal lattice vector

G = (2π, 0, 0)

maps the face at

= +π

= −π

. This leaves 3 loopsone per spatial axis.

These 3 loops are

separate connected components

of the zero set

−1

(0)

: the loop at

= π

lies in the plane

= π

, the loop at

= π

in the plane

= π

, and the loop at

= π

in the plane

= π

. These three planes are disjoint within the BZ. To see this: the

truncated octahedron BZ of the FCC lattice is bounded by the constraint

|+ |k

| ≤ π

for each pair

i = j

(in the integer-basis conventions of Eq. (1); see [8], Ch. 9). A point

with

| = π

and

| = π

would require

| + |k

| = 2π > π

, violating this bound.

Therefore no point inside the BZ lies on two of the three planes simultaneously, and the

three loops are genuinely separate connected components of the zero set.

For a codimension-2 zero set (a 1D nodal line in 3D), the chirality contribution of

each connected component equals the winding number of

⊥

on any small circle linking

that component. This winding number is constant along the loop provided no other

zeros enter the tubular neighborhood: if

|f (k)| > 0

for all

within distance

of the loop

(excluding the loop itself), then the winding is a topological invariant independent of

where on the loop the linking circle is placed. The completeness search (Proposition 1)

establishes

|f |

≥ 4.5 ×10

−4

at Euclidean distance

d ≥ 0.1

(in the dimensionless

-space

units of Eq. (1)) from all known zeros, including the BZ boundary region where the nodal

loops reside. This conrms a zero-free tubular neighborhood of radius

ϵ = 0.1

around

each loop. The measured winding at every tested segment is

(Table 2). Therefore:

boundary

= 3 ×(+1) = +3

(9)

5.3 Nielsen-Ninomiya verication

With

= +1

(Section 4),

total

= −4

(Section 4), and

boundary

= +3

(computed above):

total

= +1 −4 + 3 = 0 ✓

(10)

This is a consistency check, with the caveat that the topological invariance of the

winding (Section 5.1) depends on the zero-free tubular neighborhood established by the

completeness search, which is itself a conjecture (Proposition 1). Precisely:

assuming the

completeness conjecture

, the winding is a topological invariant,

boundary

= +3

, and the

Nielsen-Ninomiya sum vanishes.

Caveat.

The per-segment winding is numerical (not analytic), and the identication of

3 independent loops relies on the connected-component structure of the zero set. A fully

rigorous treatment would require an analytic proof that each nodal loop has transverse

winding exactly

6 Representation Theory of the

-Point Quartet

The 4 independent

-points are the centers of the 4 pairs of hexagonal faces of the

truncated octahedron BZ. They transform under the chiral octahedral group

O ⊂ O

(the 24-element rotation group of the cube, isomorphic to

). The 4 body diagonals of

the cube are permuted by

, giving a faithful action on the quartet. We write

∼

throughout.

Observation.

The permutation representation of

objects decomposes as

n →

1 ⊕ (n −1)

. For

4 → 1 ⊕ 3

(11)

where

is the trivial representation and

is the standard irreducible representation. This

is standard; the content specic to the FCC lattice is the consequences for the fermion

propagator.

6.1 Consequences for the propagator

Any

-invariant mass matrix

on the quartet, by Schur's lemma, takes the form

M =

+ m

(projections onto singlet and triplet). The singlet and triplet generically

acquire dierent masses; within the triplet, all three modes are exactly degenerate as long

is unbroken.

A modied Wilson term assigns

′

= 1.5

to the 6 neighbors with

·n| = π

(aligned

with the

(1, 1, 1)

diagonal) and

r = 1

to the other 6. This breaks

∼

→ S

. At

= (π/2, π/2, π/2)

: the 6 enhanced neighbors all have

cos(k · n) = −1

, contributing

6 × 1.5 × 2 = 18/a

. The other 6 have

cos = +1

, contributing 0. So

M(L

) = 18/a

= (π/2, π/2, −π/2)

: only 4 enhanced neighbors give

cos = −1

(contributing

4×1.5 ×2 = 12/a

) and 2 standard neighbors also give

cos = −1

(contributing

2×1 ×2 =

4/a

); the net is

M(L

) = 14/a

. By

symmetry,

M(L

) = M(L

) = 14/a

. The splitting

∆M = 4/a

is cleanly resolved.

6.2 Properties of the quartet

The

-point quartet is distinguished from the boundary nodal lines by three properties:

Isolation:

The

-points are the only zero modes in the BZ interior apart from

Denite chirality:

Each carries

χ = −1

, computed from the Jacobian (Section 4).

Gapping:

A standard Wilson term gives

-points mass

12/a

while

remains

massless (Table 1). The boundary lines acquire

16/a

7 Comparison with Other Lattices

Lattice

Doublers

decomp.

Simple cubic 6

= 8

equiv. corners

4 → 1 ⊕ 3

†

BCC 8 Multiple; no isolated quartet No natural 4-orbit

∗

FCC 12

1(Γ) + 4(L) +

lines

4 → 1 ⊕ 3

Table 3: Doubler structure on dierent 3D lattices.

†

The 8 simple cubic doublers reduce

to 4 under antipodal identication, giving

4 → 1 ⊕ 3

under

formally identical to

FCC. The distinction is quantitative: on SC, all 4 receive identical Wilson mass

6r/a

and

identical Jacobian eigenvalue structure; on FCC, the 4

-points have

M = 12/a

(distinct

from

and boundary at

16/a

) and a dierent Jacobian signature (

{+4, +4, −8}/a

vs.

{8, 8, 8}/a

). The FCC spectrum is stratied; the SC spectrum is not.

∗

The

BCC Brillouin zone is a rhombic dodecahedron with 12 face centers. Under

, these

form a single orbit; under antipodal identication they reduce to 6. The permutation

representation of

decomposes as

6 → 1 ⊕ 5

, giving no triplet; the group-theoretic

structure diers qualitatively from the FCC quartet. See [9] for BCC gauge theory.

Among the common 3D lattices (simple cubic, BCC, FCC), only FCC produces a

stratied spectrum with an isolated quartet, denite chiralities, and gappable boundary

modes. We have not exhaustively surveyed all possible 3D lattices.

8 Discussion

We outline directions for future work, noting that these are open questions rather than

predictions of the current analysis:

Minimum fermion content.

A Wilson term with

r = 1

gives the

-modes mass

12/a

and boundary modes

16/a

, leaving

massless (Table 1). In the continuum limit, the

heavy modes decouple, leaving a single massless speciesthe same outcome as hypercubic

Wilson fermions, achieved with a dierent doubler structure.

Retaining the quartet.

The anisotropic Wilson term in Section 6.1 demonstrates that

the singlettriplet splitting is real and computable. However, all

-modes remain at

the cuto scale (

∼ 1/a

). Retaining them at physical masses while gapping boundary

lines would require a momentum-dependent Wilson term. One could seek a momentum-

dependent Wilson term

w(k)

that is large at

and small at

. The function

w(k) =

cos

)

satises this:

w(X) = 3

(since

cos

π + cos

0 + cos

0 = 3

) and

w(L) = 0

(since

cos

(π/2) = 0

). In position space,

cos

) = (1 + cos 2k

)/2

, which corresponds

to on-site and next-nearest-neighbor couplingsa nite-range, local action. However, we

have not checked whether the resulting lattice action satises reection positivity, which

is the key constraint for a well-dened Euclidean eld theory. We ag this as a direction

for future investigation, not a concrete proposal.

Boundary nodal lines.

The continuous nodal lines carrying

boundary

= +3

are a

regularization challenge with no hypercubic analogue. Domain-wall fermions [12] oer

a possible approach: if the FCC lattice can be embedded as the boundary of a higher-

dimensional lattice with a domain-wall mass termwhich is nontrivial, since the transfer

matrix structure on non-hypercubic bulk lattices has not been studiedthe

mode might

appear as a chiral zero mode on the wall. Overlap fermions would be another natural

approach. The key open question is whether any method can gap the nodal lines while

preserving the

-quartet.

9 Conclusion

The FCC lattice has a richer Dirac spectral structure than hypercubic lattices. The naive

Dirac operator yields a stratied zero-mode spectrumsinglet, isolated quartet, bound-

ary lines (completeness conjectured from numerical search)with chiralities

(com-

puted),

−4

(computed), and

(winding-number computation), satisfying the Nielsen-

Ninomiya theorem:

+1 −4 + 3 = 0

. The quartet decomposes under

∼

4 → 1 ⊕3

with an explicit anisotropic Wilson term producing a

∆M = 4/a

singlettriplet splitting.

References

[1] Nielsen H. B., Ninomiya M., No-go theorem for regularizing chiral fermions,

Nucl.

Phys. B

185

, 20 (1981).

[2] Nielsen H. B., Ninomiya M., Absence of neutrinos on a lattice: (I). Proof by homo-

topy theory,

Phys. Lett. B

105

, 219 (1981).

[3] Wilson K. G., Connement of Quarks,

Phys. Rev. D

, 2445 (1974).

[4] Susskind L., Lattice fermions,

Phys. Rev. D

, 3031 (1977).

[5] Kogut J., Susskind L., Hamiltonian formulation of Wilson's lattice gauge theories,

Phys. Rev. D

, 395 (1975).

[6] Hales T. C., A proof of the Kepler conjecture,

Ann. Math.

162

, 1065 (2005).

[7] Coxeter H. S. M.,

Regular Polytopes

, Dover (1973).

[8] Ashcroft N. W., Mermin N. D.,

Solid State Physics

, Saunders College (1976).

[9] Celmaster W., Gauge theories on the body-centered cubic lattice,

Phys. Rev. D

, 2955 (1982).

[10] Karsten L. H., Smit J., Lattice fermions: species doubling, chiral invariance, and

the triangle anomaly,

Nucl. Phys. B

183

, 103 (1981).

[11] Fang C., Weng H., Dai X., Fang Z., Topological nodal line semimetals,

Chin. Phys.

, 117106 (2016).

[12] Kaplan D. B., A method for simulating chiral fermions on the lattice,

Phys. Lett.

288

, 342 (1992).