WorksPhysics and Astronomy
Non-Abelian dynamics on a cube: improving quantum compilation through qudit-based simulations
Abstract
Recent developments in mapping lattice gauge theories relevant to the Standard Model onto digital quantum computers identify scalable paths with well-defined quantum compilation challenges toward the continuum. As an entry point to these challenges, we address the simulation of SU(2) lattice gauge theory. Using qudit registers to encode the digitized gauge field, we provide quantum resource estimates, in terms of elementary qudit gates, for arbitrarily high local gauge field truncations. We then demonstrate an end-to-end simulation of real-time, qutrit-digitized SU(2) dynamics on a cube. Through optimizing the simulation, we improved circuit decompositions for uniformly-controlled qudit rotations, an algorithmic primitive for general applications of quantum computing. The decompositions also apply to mixed-dimensional qudit systems, which we found advantageous for compiling lattice gauge theory simulations. Furthermore, we parallelize the evolution of opposite faces in anticipation of similar opportunities arising in three-dimensional lattice volumes. This work details an ambitious executable for future qudit hardware and attests to the value of codesign strategies between lattice gauge theory simulation and quantum compilation.
Introduction
Gauge field theories describe three of the four fundamental forces of our universe. Lattice gauge theory (LGT) provides a systematic regularization of such field theories amenable to computational approaches. In LGT, the field is discretized onto a spatial lattice [1, 2] in which bosonic (fermionic) content of the theory resides on links (sites). As a non-perturbative framework, lattice techniques are well-suited to address the complex emergent phenomena ubiquitous in gauge theories.
The expanse and density of a lattice reflects the accuracy of the associated discretized field representation. Reducing the lattice spacing (raising the ultraviolet truncation) and increasing the volume (lowering the infrared truncation) to systematically improve resolution of the associated discretized field representation. Reducing the lattice spacing (raising the ultraviolet truncation) and increasing the volume (lowering the infrared truncation) to systematically improve resolution of physical length scales recovers the continuum behavior of gauge theories. However, this requires large Hilbert spaces to accurately capture dynamical, highly entangled processes in nuclear and high-energy physics. As such, many experimentally-relevant observables remain intractable, even in the era of exascale computing [3, 4].
Quantum computers are expected to efficiently simulate LGTs [5, 6, 7, 8, 9]. Many representations of LGT on quantum computers are actively being developed [5, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39]. The computational resource consequences of each representation—e.g., circuit width and depth, ratio of physical to unphysical Hilbert space and sensitivity to noise, compilation efficiency and required classical preprocessing, and rapidity of convergence to the
continuum—vary dramatically. While innovative representations continue to arise, persistent features of scalable strategies involve mapping quantum registers to local quantum numbers of the gauge field and retaining a non-zero gauge-variant subspace within the computational Hilbert space of the quantum device.
The latticized gauge field is represented by an infinite-dimensional Hilbert space local to each link, making qudits a natural candidate for LGT simulations as they encode gauge fields at a higher truncation than qubits. Qudit-based quantum computing hardware has also advanced rapidly in recent years [40, 41, 42, 43, 44, 45, 46, 47, 48], incentivizing further exploration of qudit-based quantum simulations [24, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62]. Moreover, qudits show promise for scaling universal quantum computing in general applications [63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83], placing importance on understanding qudit circuit synthesis techniques and associated costs [84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103].
In this work, we design and compile a qudit-based quantum simulation of LGT on a cube. We utilize the representation developed in Refs. [17, 24, 104], i.e., a link-extended local basis storing only the total angular momentum at each link. This basis is scalable, has reduced proportion of unphysical subspace, requires fewer quantum registers, and offers a clear path for explicit quantum compilation into elementary gates. By engaging in the compilation, we mutually advance qudit-based simulation of LGT and fundamental qudit compilation techniques with the following contributions:
improved circuit decompositions for qudit state preparation and unitary synthesis,
optimized synthesis of the arbitrary-truncation (qudit) plaquette operator evolution,
full specification of an executable for qutrit-digitized gauge field simulation on a cube.
Our first contribution benefits qudit-based quantum computing in general. For instance, the improved decomposition of Uniformly-Controlled Qudit Rotations (Section III A) has direct applications to qudit state preparation [85, 86] and unitary synthesis [85, 86, 87, 88, 89, 92, 96]. Furthermore, the decomposition of Control-Restricted Qudit Rotations (Section III C) is useful in mixed-dimensional qudit settings. In particular, we find value in incorporating auxiliary qudits whose dimension scales only with lattice connectivity rather than gauge field truncation.
Time-evolution of the plaquette operator comprises the dominant cost of LGT simulations. As our second contribution, we efficiently synthesize the unitary evolution of the plaquette operator and provide precise quantum resource estimates in terms of elementary qudit gates. The synthesis procedures apply to any gauge field truncation, providing an explicit recipe for simulating LGT on quantum hardware.
Our third contribution serves as verification of the decompositions, and as an ambitious executable for future qutrit hardware [105]. Our use of a cube highlights parallelization opportunities that arise at the operator level when time evolution involves shared registers in higher-dimensional lattices.
Below, we begin with an explanation of the target LGT simulation and derive properties of the qudit plaquette operator in Section II. The improved qudit decomposition techniques are presented in Section III. These techniques are used to efficiently synthesize the qutrit cube simulation in Section IV, and to report explicit resource counts for unitary evolution of the qudit plaquette operator in § 1.4. We discuss future works in Section VI. In optimizing the target LGT simulation, we glean insights into quantum architectural design and compilation methods for general applications of qudit-based quantum computing, which is reminiscent of the role LGT played in driving HPC development [106, 107, 108]. To that end, this work is written from both LGT and quantum compilation perspectives, separately and together, with the goal of facilitating ongoing codesign.
Pure SU(2) Gauge Field on a Cube
Demonstrating techniques that extend naturally to larger lattices, we simulate SU(2) lattice gauge theory on a cube with no matter fields (Figure 1). We use the angular momentum basis and remove alignment variables by incorporating part of the local gauge symmetry [17]. The remaining Hilbert space is represented by a single quantum register on each gauge link. Registers are structured to represent a one-dimensional tower of states labeled by total angular momentum spanning half integers from 0 to the truncation . Absent matter (both static and dynamical), physical lattice configurations in this basis satisfy the triangle inequalities of angular momentum addition at each vertex, and continue to do so throughout the unitary dynamics. In our simulation, we use -level quantum systems (qudits) to encode the gauge field using one qudit at each link.
Figure 1. We use single qudits to encode the gauge field at each link. While the developed decomposition techniques apply to Trotterizations for many lattice configurations of plaquettes, the simulation demonstrated is a cube. Applying the plaquette operator to face A evolves the (teal-colored) plaquette link controlled on the neighboring links .
To capture the non-Abelian dynamics, we use the Hamiltonian of Yang-Mills lattice gauge theory [2], which is composed of an electric and magnetic term,
For the SU(2) gauge theory, the plaquette operator of the magnetic term is Hermitian, , and the local gauge-invariant Casimir operator is , where is the projection in the computational basis.
Truncated Plaquette Operator
The matrix elements of the plaquette operator generate transitions between states in our chosen basis of total angular momentum quantum number per link. The Hilbert space of the operator comprises a physical subspace, (where states contract to form a gauge singlet at each vertex), and an unphysical subspace, . The plaquette operator can be expressed as a direct sum, , reflecting the separation of these subspaces throughout time evolution.
Initializing our simulation in grants the freedom to assign arbitrary (Hermiticity-preserving) values to matrix elements of . This freedom can be leveraged to introduce a gauge-variant completion (GVC) that simplifies the eventual quantum circuit implementation [17]. We choose a GVC for with the form
where the Hermitian operators are restricted to adjacent subspaces due to the ladder connectivity of irreducible representations (irreps) of SU(2) dynamics generated by Equation (1). The physical transition matrix elements are obtained from Equation (11) [17, 104].
Insight into the structure of can be obtained by visualizing the flux line configurations of the three links at a vertex of the cube, as depicted in Figure 2. To connect all valid SU(2) irreps, the operator must contain a complete set of distinct , since there are choices for each in (indices suppressed for clarity).
Figure 2. Graphical representation of how affects the flux lines at a vertex via example for . The number of flux lines is , representing the link’s total angular momentum. The operator transitions the plaquette link flux lines between two physical configurations indicated by the arrows. The control values corresponding to physical configurations span from to . However, applying to the flux configuration with results in an unphysical state, and thus is not paired with in .
Gauge-invariance is preserved by conditional application of [109]. Applying to plaquette A in Figure 1 would apply to the control links , thus controlling the application of to the plaquette links with control word . The control value of a valid gauge-invariant transition at the vertex (Figure 2) satisfies ,
where since the gauge field is digitized to a -dimensional qudit.
To ensure the system is physical, a further constraint is imposed by Gauss’s law via a Gaussian surface that cuts through all four control links of . The set of projectors controlling becomes
where . This form highlights the symmetry inherent to , which we leverage to further optimize the classical preprocessing required to construct : there are ways to assign choices of to four links of a plaquette under symmetry, where is a doubly triangular number. Thus, we
| order | |||
| 0000 | 1 | 0000 0011 0101 0110 1001 1010 1100 1111 | 8 |
| 0001 | 4 | 1001 1012 1102 1111 2002 2011 2101 2112 | 8 |
| 0011 | 4 | 101_ 1021 1111 112_ 2011 202_ 211_ 2121 | 12 |
| 0101 | 2 | 1111 1122 1212 1221 2112 2121 2211 2222 | 8 |
| 0111 | 4 | 11__ 1111 12_1 121_ 21_1 211_ 22__ 2211 | 18 |
| 1111 | 1 | ____ __11 _1_1 _11_ 1__1 1_1_ 11__ 1111 | 41 |
Table 1. The terms in the qutrit plaquette operator. Under symmetry, there are representative ways to assign to each . These representative classes are paired with from the set defined in Equation 4. The control words are shown, using the symbol _ to denote cases where all possible combinations of control values 0 and 2 are present. To obtain the complete set of 16 , the indices are permuted using the transformations; “order” indicates the number of permutations resulting in a unique . The qutrit plaquette operator has 217 terms, which are explicitly shown with values in Equation (12).
only need to compute and for unique . Then, the full set of is obtained by permuting the indices .
Every pairing of operator and from the set of corresponding controls is associated with a non-zero physical transition matrix element of (verified numerically for ). As such, our construction provides a concise description of that is applicable for any gauge field truncation, where in the continuum limit. For our demonstrated simulation, we specialize to a qutrit truncation (). Table I enumerates the elements of the qutrit plaquette operator, denoted as . To support future calculations with increasing truncation, the arbitrary- techniques developed above are applied to construct in Appendix A.
Simulation Overview
We apply a first-order Suzuki-Trotter decomposition with time steps to simulate real-time evolution of the gauge field on a cube. A quantum circuit depicting the overarching structure of a Trotter step is presented in Figure 3. Each step consists of evolution under the Casimir operator, followed by successive evolutions of the plaquette operator over all six faces of the cube.
Figure 3. Quantum circuit for a single Trotter step in the time evolution of SU(2) gauge fields on a cube. The simulation alternates between the Casimir operator and plaquette operators. Evolution of the latter is parallelized in pairs of opposing faces. Each plaquette evolution is decomposed as a series of controlled rotations. While those characterized by the same rotation commute, an additional layer of Trotter error is introduced in ordering the by increasing integer representation of their subspaces, i.e., the binary strings in Table I. The structure of the elementary-gate level decomposition in the lower right corner is specific to the qutrit digitization of the gauge field. The boxed-“×” circuit elements are qudit operations (see Figure 4).
In the GVC of Equation (2), each plaquette operator evolution is decomposed into a series of controlled four-qudit rotations whose angles are proportional to physical transition matrix elements. Our decompositions enable interleaving of controlled operations, parallelizing for opposite faces of the cube that share control links, as depicted in the bottom right of Figure 3. This approach leads to significant savings in circuit depth. Similar parallelization opportunities arise for plaquette operators in genuine higher-dimensional lattices, whether organized by cubic components of a hypercubic lattice or residing in a broader class of spatial latticizations.
Decomposition Techniques
The following decomposition techniques are presented in a general manner as they can be leveraged for qudit compilation beyond the present application which inspired their design. We describe their application to efficiently synthesize the evolution of the plaquette operator for qutrit and qudit digitization of the gauge field in Section IV A and Section V, respectively.
We use the elementary qudit gate set of Ref. [89], where the Generalized Controlled X (GCX) gate (Figure 4) serves as our elementary two-qudit gate. All GCX gates are assigned the same cost in our cost model, regardless of control value or subspace. In the present combination of SU(2) gauge field representation and decomposition techniques, nearly all GCX gates apply to nearest-neighbor qudit levels, which may be experimentally preferable.
Figure 4. The two-qudit GCX gate applies on the target qudit, conditioned on the control qudit being in state . For qutrits, we use distinct shapes to represent three possible control values.
Uniformly-Controlled Qudit Rotations
The Hermitian operators , , and are the generators of two-level qudit rotation operators , [89]. The Uniformly-Controlled Qudit Rotations subroutine is the set of -controlled gates, , controlled on all possible control words.1
We generalize existing techniques [92, 110, 111, 112] to efficiently decompose by systematically constructing a circuit using the -ary reflected Gray code. Figure 5 shows our technique applied to . The index of the varying digit in the Gray code identifies the GCX control qudit; the control value is determined by the higher value of the varied digit. As an example, the transition in the Gray code places a GCX gate controlled on the second control qudit with control value 2.
Figure 5. (top) Uniformly-Controlled Qudit Rotations about the axis . Above, the number of control qudits is and the qudit dimension is . The gate label is shorthand for (likewise for ). (bottom) The efficient decomposition of . All gates are applied to the same subspace.
Referring to the notation in Figure 5, the rotation angles are transformed into according to
where is the -th digit of the -ary representation of the integer , and is the -th digit of the -ary reflected Gray code term at index . We remark that for qubits, evaluates to 1 only when both inputs are 1, and the sum over all bit indices reduces to a simple dot product of the binary and Gray code sequences for and . This yields exactly the inverse of the expression for in Equation (3) of Ref. [111]. For , Equation (6) is no longer simply a dot product as it depends on the order of the arguments.
This decomposition of requires single-qudit rotation gates, at most GCX gates, and at most one gate. For a full account of the circuit construction and resource count, as well as a ququart example, see Appendix B.
To the best of our knowledge, the most efficient known decomposition for requires GCX gates [89]. In the present application where is fixed by the lattice connectivity and increases with gauge field truncation, our decomposition is a improvement on the GCX gate count and depth in the limit that .
Subspace Gating
Let and be two qudit dimensions, . We implement a technique, Subspace Gating, that allows the use of -dimensional qudit decompositions as subroutines in -dimensional qudit circuits, provided a “gating subcircuit” is constructed. The gating subcircuit prevents a -dimensional subroutine from being applied when inputs to the subroutine are outside the -dimensional subspace. Essentially, Subspace Gating uses an auxiliary qudit to exclude spectator dimensions, leading to reduced gate counts.
Figure 6 shows an instance where Subspace Gating saves GCX gates. For comparison, we will temporarily use the number of -controlled rotation gates as a proxy for the number of GCX gates. The unitary can be expressed as an -controlled , yielding -controlled rotation gates via the decomposition in Section III A. By contrast, with Subspace Gating, decomposes into two -controlled rotation gates, independent of both and . The savings stem from use of the qudit subroutine (bottom of Figure 6) enabled by the gating subcircuit.
Figure 6. The unitary applies controlled by an -qudit register. (top) Without Subspace Gating, is decomposed as a sequence of -controlled gates with control words spanning . This is equivalent to where associated with control words in are 0. Decomposing with the technique in Section III A yields rotation gates, each controlled by the upper register, yielding a total of -controlled rotation gates. (bottom) With Subspace Gating, the gating subcircuit (teal) enables the decomposition of in the qubit subspace, i.e., the decomposition of . This requires two -controlled rotation gates regardless of . On the right, we show the small case of , and for concreteness.
Generally speaking, the gating subcircuit verifies an input is in the subroutine’s subspace using the (AND) gate before toggling the auxiliary qudit to activate the subroutine. In the special case of qutrits, it is more efficient to implement the (OR) gate version of the input verification gate (which is obtained using De Morgan’s law [113], described in Figure 15). An example is given in the qutrit specialization presented at the bottom right of Figure 6. In this case, the gate is implemented as a single control-on- GCX gate. No gates are required for the bottom input qutrit since the subroutine naturally has no effect when the bottom qutrit is in state , a subroutine-specific optimization we apply throughout this work. We provide input verification gates for arbitrary and in Appendix § 1.2.3.
Figure 15. The and version of the gating subcircuit. De Morgan’s law, , allows us to implement an gate using an and a gate. In the context of the gating subcircuit: the two gates flip the control value from 1 to 0, the gates become gates, and the inputs satisfying the Boolean condition are inverted from to where is the subroutine’s subspace.
As a caveat, we note there exists a better decomposition of that takes advantage of our specific values (see Figure 13), yielding -controlled rotation gates. Although improves upon , it is still worse than the 2 achievable through Subspace Gating.
Figure 13. An alternate decomposition to Figure 6 of for arbitrary and , illustrated for . Since contains an equal amount of positive and negative angles, we decompose controlled rotation gates in pairs where the angles have equal magnitude and opposite sign. The resulting circuit contains -controlled rotation gates.
Control-Restricted Qudit Rotations
Let be a sequence of control words ordered by increasing integer representation of their -ary strings. For described in Section III A, the control sequence is . We generalize where , i.e., does not contain all possible control words, under the condition that the control qudits are also restricted to . Qudit states may be restricted in circuits with mixed-dimensional qudits, artificially using Subspace Gating, or due to the properties of the quantum simulation. Figure 7 shows an example of where contains the even-numbered control words spanning two control qutrits.
Figure 7. A step-by-step decomposition of with . We begin with the Gray-code ordered (which in this particular case is equivalent to ) and generate the quantum circuit using the procedure described in Section III A. (top) Assuming the control qutrits are restricted to , observe that the top control qutrit being in state uniquely identifies the control word “11” and its corresponding rotation angle . The gates coloured in teal are not necessary, and we are free to relabel the Gray code term from “11” to “12”. Doing so saves one GCX gate. The relabelling changes the s to s, while the change is also reflected in such that the overall operation is unchanged. (middle) Since s are sums of s, the order in which their constituent rotations occurs is arbitrary. Thus, we are free to swap the order of “20” and “22”, as this saves another GCX gate. (bottom) The final decomposition of applies only one GCX gate between rotation gates.
We provide a decomposition of in Appendix D for arbitrary , , and . Our implementation generates two sequences of -ary strings, and , from . The sequences and are used to transform the angles according to Equation (6) (taking the shorter sequence length into account when constructing ) and generates the quantum circuit.
By keeping general, we sacrifice the guarantee that the matrix is invertible, although we conjecture our
particular implementation yields an invertible when
i.e., uniform control but non-uniform qudit dimension.
Although our implementation produces the decomposition in Figure 7 with an invertible , we encountered cases where this did not hold, specifically for larger comprising all even-numbered control words for a mixed-dimensional qudit system. An example is found in Section IV (Equation (9)). In that situation, manual corrections were applied (as described in Appendix § 2.6) to make invertible. Although the manual corrections may not be optimal, their existence suggests that there is an improved implementation which does yield an invertible under such control sequence restrictions. Finding this implementation, which we henceforth refer to as the “hypothesized decomposition” is the subject of future work.
Qutrit Cube Simulation
We apply the decomposition techniques described in Section III to simulate SU(2) gauge field on a cube, truncated to . First, we decompose the Trotterized unitary evolution of the plaquette operator digitized to qutrits. Then, we repeatedly apply the decomposed operator to perform the qutrit cube simulation.
Plaquette Operator Evolution
The Trotterized evolution of the qutrit plaquette operator is
where is shorthand for a -weighted sum over all projectors associated with the term (Equation 4 and column in Table I) and incorporates constants from Equation (1).
We take the following steps to decompose :
decompose into the circuit in Figure 8 with Subspace Gating (Section III B),
Figure 8. Decomposition of applied to plaquette B of the cube (equivalent to the bottom-right of Figure 3, without parallelization). The boxed “” is whereas the regular CNOT target represents . The circuit evolves the plaquette links while leaving the control links and auxiliary qutrit unchanged. The subroutine on the left calculates the subspace parity in the computational basis, so we name it the -parity subroutine. The subroutine in the middle is a sequence of four-qutrit controlled with control words corresponding to row 2 of Table I. An additional control is placed on the rotations by the gating subcircuit, which gates the subroutine by setting the auxiliary qutrit to if any of the first three input qutrits are in state . The subroutine naturally has no effect if the fourth input qutrit is in state , as all gates operating on that qutrit are in the subspace. Therefore, only the three-input gate is needed to subspace gate the plaquette links.
decompose the remaining Control-Restricted Qudit Rotations into elementary qutrit gates (Section III C),
apply the decomposed circuit 16 times, varying parameters for each , to implement .
In step 1, we decompose by Subspace Gating the plaquette links to enable the qubit decomposition of . We factor out Hadamard gates () and GCX gates to obtain the circuit in Figure 8. The subspaces of and are the same as those of assigned to the plaquette links, and the control values of the GCX gates are the larger of and . The gating subcircuit uses the three-input gate (Figure 19) to verify that the plaquette links are in the same subspace as . Lastly, we commute the gates through the and GCX gates to facilitate further cancellations. At this stage, all gates are elementary qutrit gates except for the contiguous -controlled rotations in the middle of the circuit.
Figure 19. The three-input qutrit gate, a special case of the three-input qudit gate shown in Figure 16, where an additional GCX gate is saved by applying the optimization in Figure 12 on the pair of GCX gates corresponding to the rightmost GCX gate in the qutrit Toffoli gate (Figure 17). This three-input qutrit gate uses 6 GCX gates, an gate and has depth 6.
In step 2, we decompose the Control-Restricted Qudit Rotations using the technique in Section III C. These rotations are controlled by five qutrits (four control links and one auxiliary qutrit from the gating subcircuit). The control words corresponding to the control links are constrained by from Equation 4, and the control value of the auxiliary qutrit is restricted to . Thus, the control sequence of is
In this case, the matrix is not invertible and requires manual correction (via additional GCX gates). The corrected version, , incurs 12 additional GCX gates and 12 additional depth (see Appendix § 2.6). Despite
| Subcircuits | Counts | GCX | Depth | |||
| 94 | 82 | 0 | 0 | 176 | ||
| gate | 6 | 0 | 1 | 0 | 6 | |
| -parity | 3 | 0 | 0 | 4 | 3 | |
| 112 | 82 | 2 | 8 | 194 | ||
| (Total) | 1792 | 1312 | 32 | 128 | 3104 |
Table 2. Quantum resource estimates for the Trotterized evolution of one qutrit plaquette operator . The Trotterized is a product of 16 unitaries of the form , denoted as in the second and last row. The decomposition uses one auxiliary qutrit, leading the operator to span a total of nine qutrits.
this, uses 94 GCX gates, far less than used for .
Finally, in step 3, we implement as per Equation (8) by applying circuits like that in Figure 8 for each of the 16 distinct subspaces, control values of the qutrit gates, and the angles are changed in accordance with Table I. The quantum resource estimates for implementing one are shown in Table II.
As a final remark, we developed an alternate decomposition of that leverages the constraint on control words, in addition to Equation (3). For clarity of presentation, we relegate this more detailed decomposition to Appendix § 2.7.
Cube Simulation
To simulate the real-time dynamics of SU(2) gauge fields on a cube, we repeatedly apply the decomposed as shown in Figure 3. Taking advantage of the parallelization opportunity described in Section II B, we stack two decompositions applied to opposite faces of the cube to halve the depth of the circuit, as shown in Figure 9. We reverse the gate order for one face and interleave the controls on the shared gauge links (bottom-right of Figure 3). Simultaneously evolving two faces requires a second auxiliary qutrit, increasing the qutrit count of the cube simulation with Subspace Gating to 14.
Figure 9. Decomposition of two , applied in parallel to opposite faces of the cube, using two auxiliary qutrits to apply Subspace Gating on the plaquette links. The subspaces and control values of all gates are contingent on the particular term.
Resources for one Trotter step of the cube simulation are shown in Table III. We implemented the simulation in PennyLane [114], verifying the correctness of our decompositions by comparing simulation observables (Figure 10) with exact numerical results. The simulation was executed on a workstation running Ubuntu 24.04 with an Intel Core i7-12700KF 3.6 GHz CPU, 96GB RAM, and MSI GeForce RTX 4090 GPU with 24GB memory. A custom PennyLane (branched from version 0.38) [115] with specialized qutrit operations was used, alongside GPU-enabled JAX (v-0.4.12) and just-in-time compilation. It took roughly 3 hours to generate data for Figure 10.
Figure 10. Expectation value of the electric energy contribution of plaquette A on the cube with coupling . The Trotterized results converge towards the exact time evolution, which was simulated in the global basis of physical states (see Appendix G). The icon on the top left, introduced in [112], denotes purely classical simulation.
| Operators | Counts | GCX | X | H | Depth | |
| two | ×3 | 3584 | 2624 | 64 | 256 | 3104 |
| ×1 | 0 | 24 | 0 | 0 | 2 | |
| Total | 10752 | 7896 | 192 | 768 | 9314 |
Table III. Resource estimates for one full Trotter step of the cube simulation (Figure 3). Evolution of two s (row 1) on opposite cube faces occurs in parallel. As such, the gate counts are double that of a single (last row of Table II) while the depth remains the same. Repeating the parallelized face evolution thrice to evolve all six faces, and incorporating the electric operator (row 2) gives the total resource estimates for one Trotter step (last row). The decomposition requires two auxiliary qutrits, and 14 qutrits in total.
PennyLane was also used to obtain quantum resource estimates directly from the implemented circuits. We found exact agreement on all values in Table III except depth, which was reduced from 9314 to 9284. We attribute this improvement to PennyLane parallelizing additional gates, e.g., at the boundaries between distinct . Compiler optimizations, such as cancelling adjacent self-inverse gates, merging adjacent rotation gates, and Hadamard-GCX rewrite rules (Appendix H) can be used to further reduce gate count and circuit depth.
Qudit Simulations and Resource Scaling
To simulate SU(2) non-Abelian gauge fields at arbitrary truncation level , we digitize to qudits, . The procedure for decomposing Trotterized unitary evolution of is similar to the qudit case in Section IV A, with key differences detailed below.
In step 1, we use the -version of the input verification gate, shown in Figure 14, instead of the V-version used for the qudit case. Since the gate contains gates operating in the same subspace as the qubit subroutine, we can no longer commute a gate through the -parity subroutine as we did for the V gate in Figure 8. Despite missing out on some cancellations, the gate is more efficient overall than the gate when . We further observe that the dimension of the auxiliary qudit is independent of gauge-field truncation level, scaling instead with the lattice connectivity. With the current connectivity, the auxiliary is a ququart, reflecting the Subspace Gating of four plaquette links.
Figure 14. The input verification circuit for qudits with checks whether the input is in the subspace. The half-filled circles represent adjacent GCX gates (with control values spanning ) applying the same unitary. This circuit construction supports input qudits as long as the auxiliary qudit has dimension . The circuit uses GCX (and has the same depth). Taking the qubit subspace as an example, , each circuit element becomes two GCX gates, one controlled on and the other on . The circuit writes to the auxiliary qudit if the input is in the chosen subspace of . For the plaquette operator with and , the circuit uses 10 GCX gates and has depth 10.
In step 2, we decompose the Control-Restricted Qudit Rotations with the control sequence
instead of Equation (9). Doing so forgoes the resource improvement from the naturally restricted Equation (9), effectively doubling the number of GCX and gates, but with the benefit of an invertible .
In step 3, we repeatedly apply the decomposed circuit times (varying parameters appropriately), to implement . We report resource estimates for the full in Table IV. We achieved an overall GCX gate count of , and circuit depth . Substituting and comparing with the qutrit resources in Table II, we find only expected discrepancies due to the different gating subcircuits and the decomposition.
| Subcircuits | Counts | GCX | X | H | Depth | |
| 1 | 0 | |||||
| gate | 10 | 0 | 0 | 0 | 10 | |
| -parity | 3 | 0 | 0 | 4 | 3 | |
| 1 | 8 | |||||
| (Total) | ||||||
| (Scaling) |
Table IV. Upper bound resource estimates for time evolution of one plaquette operator (last two rows). The organization of the table follows that of Table II. We use the decomposition of with (to avoid the need for manual corrections) and the gate (Figure 14) in our gating subcircuit. A single auxiliary qudit is used in the decomposition, leading the operator to span a total of nine qudits.
The resource estimates can be related directly to the plaquette operator GVC. In the following description, we consider access to the hypothesized decomposition of discussed in Section III C. With the aid of one auxiliary ququart, the presented decomposition uses GCX gates and gates with counts roughly equal to the number of combinations, constrained only by the qudit dimension and not by Equation 4. This connection originates from the matrix transforming of size to of size for each , where corresponds to the qudit version of Equation (9). These counts are half of those reported in Table IV, bringing the qudit resource estimates inline with those for qutrits (Table II). Table IV serves as a resource upper bound in the absence of the hypothesized decomposition of . The hypothesized decomposition precludes manual corrections in the qutrit case as well.
The alternate decomposition presented in Appendix § 2.7 achieves better scaling in gate count and depth at the cost of three additional auxiliary qudits. However, if we free the dimension of the auxiliary from the gauge-field truncation as we did above, then the decomposition can be done with one qu8it, an 8-dimensional qudit, reflecting the Subspace Gating of all eight gauge field links. The alternate decomposition uses GCX gates and gates with counts roughly double the number of non-zero transition matrix elements, as obtained by Equation 4. This scaling can be seen in the resource estimates presented in Table IX.
| , no gates on control links | ||||||
| Subcircuits | Counts | GCX | X | H | Depth | |
| 94 | 82 | 0 | 0 | 176 | ||
| 188 | 164 | 0 | 0 | 176 | ||
| Total Contribution | 188 | 164 | 0 | 0 | 176 | |
| , two-input on control links | ||||||
| Subcircuits | Counts | GCX | X | H | Depth | |
| 36 | 36 | 0 | 0 | 72 | ||
| Control | 4 | 0 | 0 | 0 | ||
| 80 | 72 | 0 | 0 | 80 | ||
| Total Contribution | 320 | 288 | 0 | 0 | 320 | |
| , three-input on control links | ||||||
| Subcircuits | Counts | GCX | X | H | Depth | |
| 24 | 24 | 0 | 0 | 48 | ||
| Control | 7 | 0 | 1 | 0 | ||
| 62 | 48 | 2 | 0 | 62 | ||
| Total Contribution | 248 | 192 | 8 | 0 | 248 | |
| , four-input on control links | ||||||
| Subcircuits | Counts | GCX | X | H | Depth | |
| 16 | 16 | 0 | 0 | 32 | ||
| Control | 9 | 0 | 1 | 0 | ||
| 50 | 32 | 2 | 0 | 46 | ||
| Total Contribution | 350 | 224 | 14 | 0 | 322 | |
| Common Subcircuits | ||||||
| Subcircuits | Counts | GCX | X | H | Depth | |
| -parity | 3 | 0 | 0 | 4 | ||
| gate | 2 | 0 | 1 | 0 | ||
| Plaquette | 6 | 0 | 1 | 0 | ||
| Total Contribution | 696 | 0 | 124 | 256 | 378 | |
| Total (two ) | 1802 | 868 | 146 | 256 | 1444 |
Table IX. Resource count for two applied in parallel to opposite faces of the cube. The with gates of the same input size (Table VIII) use the same resources. All gates on the control links use one additional GCX gate and depth to copy the output to another auxiliary qutrit. The total qutrit count is 17, where 12 are used to encode the gauge links and 5 are auxiliary.
Results and Discussion
In this work, we simulated pure SU(2) gauge fields in the total angular momentum basis, where each link is encoded by a qudit. We decomposed the Trotterized evolution of the plaquette operator for arbitrary gauge field truncation levels into elementary qudit gates and detailed a resource tradeoff between gate counts and dimensionality of the auxiliary qudit. The precise resource counts of these decompositions are presented in Table II, Table IV, and Table IX. In particular, when the auxiliary qudit is a qu8it, the gate counts scale linearly with the number of non-zero plaquette operator transition matrix elements, which we enumerate precisely in Equation 4 for the SU(2) gauge group. These are, to the best of our knowledge, the first quantum resource estimates at the elementary gate level for the evolution of the SU(2) plaquette operator in the scalable angular momentum representation. We present these gate-level resource estimates as an upper bound to be improved through future community investigations.
In the provided code [105], we automatically generate the SU(2) plaquette operator GVC for arbitrary truncation levels. Then, the current implementation, specialized to qutrits, compiles circuits directly from the qutrit plaquette operator GVC. Our code is extendable to qudits with appropriate enhancement of PennyLane (i.e., addition of qudit devices and operations). The provided code also implements the decompositions of Uniformly-Controlled Qudit Rotations and Control-Restricted Qudit Rotations for arbitrary . These decompositions, along with Subspace Gating, are inspired by optimizing the presented LGT simulation, but are useful in procedures common to most quantum computing applications. Quantifying the improvements afforded by our decompositions to general qudit compilation methods such as state preparation and unitary synthesis is the subject of future work.
We assemble the compiled qutrit circuits to simulate, in PennyLane, the real-time dynamics of qutrit-digitized SU(2) gauge fields on a cube, verifying the correctness of the circuitry and resource estimates. In assembling the circuitry, we utilize parallelization opportunities that arise in higher-dimensional lattice systems.
Our decompositions yield a quantum circuit that returns to the physical subspace after each term (and predictably transforms within ), enabling error-mitigation techniques to be employed at a sub-Trotter-step frequency. Such digital approaches for physical subspace verification and protection have been explored in recent work [116, 117, 118, 119, 120, 121, 122, 123]. Furthermore, we anticipate that compression techniques [124, 125] can be designed to improve the efficiency of the simulation while maintaining locality and thereby scalability of the LGT representation. These techniques become increasingly important to compensate for the diminishing ratio of physical states in the computational basis as the simulation volume and gauge field truncation are increased toward the continuum. Along with these opportunities, our qutrit cube simulation presents a compelling, physics-driven executable for near-future qutrit hardware.
The techniques presented can be readily extended to qudit compilation of the SU(3) gauge group [24], moving towards simulations of lattice quantum chromodynamics in three spatial dimensions. These techniques also form the foundation for a software package addressing quantum compilation challenges toward the continuum of LGT simulations.
Acknowledgments
This collaboration was funded by the NSERC Alliance International Catalyst Quantum Grant program (ALLRP 586483-23). JJ is funded in part by the NSERC CREATE in Quantum Computing Program, grant number 543245. NK acknowledges funding in part from the NSF STAQ Program (PHY-1818914). ODM acknowledges funding from NSERC, the Canada Research Chairs Program, and UBC. This work was conceived at the 2023 Quantum Computing, Quantum Simulation, Quantum Gravity and the Standard Model workshop at the InQubator for Quantum Simulation (IQuS) hosted by the Institute for Nuclear Theory (INT). IQuS is supported by U.S. Department of Energy, Office of Science, Office of Nuclear Physics, under Award Number DOE (NP) Award DE-SC0020970 via the program on Quantum Horizons: QIS Research and Innovation for Nuclear Science, and by the Department of Physics, and the College of Arts and Sciences at the University of Washington.
Truncated Plaquette Operator
In Section II A, we determined all with non-zero in our plaquette operator GVC via Equation 4. The values of can be calculated using [17, 104],
which is written in the language of physical flux configurations. To find the physical flux configurations associated with a , we begin by setting a plaquette link to hold total angular momentum flux lines, corresponding to assigned to the link. The remaining three plaquette links are determined by Gauss’s law, subject to the first link and fixed control values. We apply on the resulting physical flux configuration to find its partner configuration. These physical flux configuration pairs are input into Equation (11) to obtain .
From the programmatic construction, we provide the growth of various terms with increasing qudit dimension in Table V. As an example, we provide the ququart plaquette operator in Table VI. Additionally, we provide the full qutrit plaquette operator, including , in Equation (12) with a reordered Hilbert space to emphasize the organization by .
| # of | # of | # of | |
| 2 | 1 | 1 | 8 |
| 3 | 6 | 16 | 217 |
| 4 | 21 | 81 | 2346 |
| 5 | 55 | 256 | 14872 |
| 6 | 120 | 625 | 66950 |
| 7 | 231 | 1296 | 237981 |
| 8 | 406 | 2401 | 711828 |
| 9 | 666 | 4096 | 1866940 |
Table V. Scaling of various terms with qudit dimension : number of distinct under symmetry (denoted as ), total number of and the number of with non-zero .
| order | |||
| 0000 | 1 | 0000 0011 0101 0110 1001 1010 1100 1111 | 8 |
| 0001 | 4 | 1001 1012 1102 1111 2002 2011 2101 2112 | 8 |
| 0002 | 4 | 2002 2013 2103 2112 3003 3012 3102 3113 | 8 |
| 0011 | 4 | 101_ 102* 111* 112_ 201* 202_ 211_ 212* | 16 |
| 0012 | 8 | 201* 202_ 2112 212* 3012 302* 311* 3122 | 12 |
| 0022 | 4 | 202_ 203* 212_ 213_ 302* 303_ 312_ 313* | 16 |
| 0101 | 2 | 1111 1122 1212 1221 2112 2121 2211 2222 | 8 |
| 0102 | 4 | 2112 2123 2213 2222 3113 3122 3212 3223 | 8 |
| 0111 | 4 | 11_ 11** 12_ 12*_ 21_* 21_* 22_ 22** | 32 |
| 0112 | 8 | 21_* 212 22_2 22** 31_2 31** 32_ 32*2 | 24 |
| 0121 | 4 | 11** 1122 122 122 212 212 22* 2222 | 18 |
| 0122 | 8 | 21*_ 212* 22** 222_ 31** 312_ 32_* 322* | 24 |
| 0202 | 2 | 2222 2323 3232 3333 | 8 |
| 0212 | 4 | 22** 2222 232 232 322 322 33** 3322 | 18 |
| 0222 | 4 | 22_ 22** 23_* 23*_ 32_* 32_* 33_ 33** | 32 |
| 1111 | 1 | ---- --** -_ -_ ---* __* __* **** | 128 |
| 1112 | 4 | *_*2 *_2 _2 **** 2__2 2_ 2_ 2**2 | 72 |
| 1122 | 4 | __ _ 2_2 2_ 2_2 2_2_ 2*_ 22 | 72 |
| 1212 | 2 | **** 22 22 22 22 22 22 2222 | 41 |
| 1222 | 4 | __ 2_ 2_ 22 2_ 2*_ 22__ 22** | 72 |
| 2222 | 1 | ---- --** -_ -_ ---* __* __* **** | 128 |
Table VI. TABLE VI. The with non-zero values in the ququart plaquette operator. Under symmetry, there are 21 representative ways to assign to each in . These representative classes are paired with from the set defined in Equation 4. The control words “ijk l” are shown. We write _ (*) to denote cases where both the 0 and 2 (1 and 3) control values are present. To obtain the complete set of 81 , the indices are permuted using transformations; “order” indicates the number of permutations leading to unique assignments. The ququart plaquette operator has 2346 distinct with non-zero .
Uniformly-Controlled Qudit Rotations
The complete resource count for is
The dependence on and is due to the varying numbers of GCX gates that must be appended to the circuit to ensure no net rotation. For odd , we append GCX gates with control values from 1 to to each control qudit, which reduces to a single GCX gate and an gate using the optimization in Figure 12. When is even, all gates cancel, otherwise a single one remains. For even , we append GCX gates to the top control qudit and apply the same optimization when .
Figure 12. Applying an gate controlled on , is the same as not controlling on .
In Figure 11, we show the circuit decomposition of , which corresponds to uniform-control over two ququarts (, ). In the same figure, we also show the diagrammatic construction of a row in , which uniquely combines s to yield the corresponding to the control word indexing the row.
Figure 11. Circuit decomposition of , a Uniformly-Controlled Qudit Rotations subroutine for and . An open diamond represents control on . In the angle calculation, , all entries of are , and the sign of the element is determined by whether an odd or even number of indices satisfy the condition as per Equation (6). The depicted example is for the row , where in quaternary notation. To finish the circuit construction, we append three GCX gates controlled on , and to the top control ququart, and use the optimization in Figure 12 to reduce this to one GCX gate controlled on , and an gate.
Subspace Verification Circuits
The gating subcircuit accompanying a subroutine verifies the input is in the affected subspace of the subroutine and writes the result of the verification to an auxiliary qudit. For a circuit with an auxiliary qudit with dimension , we can verify the inputs using the gate in Figure 14.
In some cases, it is easier to verify instead that inputs are outside the subroutine’s subspace. An example is verifying inputs to qubit subroutines in qutrit circuits. This version of the verification is done using gates (through application of De Morgan’s law shown in Figure 15).
We provide the qudit gate construction for arbitrary input size in Figure 16, subject to . When , the qudit gate requires additional auxiliary qudits as shown in Figure 18. An optimized version of the qudit gate, the three-input qutrit gate (Figure 19), is used to synthesize the evolution of the qutrit plaquette operator.

Figure 16. The qudit gate writes to the auxiliary qudit if any of the input qudits is in state . The gate can be configured to detect states other than by changing the control value of the GCX gates. When , the circuit above features the left completion and uses GCX gates and depth. When , the right completion is used, utilizing GCX gates and depth by cancelling 2 GCX gates corresponding to the leftmost GCX gate in the qudit Toffoli gate (as constructed in Figure 17). As an example, the two-input qudit gate (left completion) uses 3 GCX gates, and has depth 3.
Figure 17. The qudit Toffoli gate, where . The circuit uses GCX gates and depth.

Figure 18. (top) We construct larger gates from smaller gates using the (combine) gate to combine gate outputs. Inputs to the gate are assumed to be in the subspace . The result is written on the qudit indicated by the notch. This construction is useful when , as those cases are not addressed by the circuits in Figure 16. (bottom) For example, the four-input qutrit gate is composed of two two-input qutrit gates and a gate, amounting to a total resource count of two auxiliary qutrits, 8 GCX gates, an gate, and depth 6. This gate is used in the fine-grained SU(2) decomposition presented in Figure 21.
The -ary and Gray Sequencers
The following details our decomposition of , following the intuition described in Section III C. Two algorithms, the -ary sequencer and the Gray sequencer, are used to generate and respectively. They are then normalized to yield the final sequences and .
We use Figure 20 to aid in describing the sequencers.

Figure 20. The -ary (light grey) and Gray (teal) trees corresponding to before branch relabelling. Notice the reflected branches of the Gray tree at the marked position, which corresponds to reflecting branches at every second parent node with multiple child nodes. The Gray sequencer relabels branch “1” which is the sole child of its parent branch to the label of the nearest left branch, which is “2”. Therefore, the Gray sequence is . In tandem, the -ary sequencer relabels the corresponding branch to “0”, and so the -ary sequence is . Since starts with the 00 control word, the sequence normalizer is not needed in this instance.
The -ary sequencer generates a tree, the -ary tree, directly from , where the branches are ordered by increasing digit values. The Gray sequencer generates a tree, the Gray tree, through modified breadth-first traversal of the -ary tree. While traversing each layer, the Gray sequencer reflects branches at every second parent node that has multiple child nodes. Parent nodes with a single child node adopt the label of the nearest branch on the left (on the same layer). In tandem, the -ary sequencer relabels the corresponding branch in the -ary tree to 0. The unnormalized sequences, and , correspond to the labels of the path taken from the root node to the leaf nodes in the -ary and Gray trees, respectively.
Finally, the sequence normalizer replaces the first control words of and with the control word by making appropriate changes to the subsequent control words, yielding and . This final step is not needed for proper circuit functionality. Rather, it avoids GCX gates that would otherwise have been placed before the first rotation gate.
From Singular to Invertible
In the absence of the hypothesized decomposition of Control-Restricted Qudit Rotations, we show how to manually correct a non-invertible generated from Equation (9) by inserting GCX gates. The matrix has dimensions (since ), but the rank of generated using the systematic techniques of Section III A is only 76. The manual corrections are shown for this specific case, although the general strategy can be broadly applied.
To increase the rank of , we conjugate a subset of the rotation gates with GCX gates, which effects sign flips on particular matrix elements. For example, suppose we conjugate the rotation gate corresponding to column 3 of in Figure 11 (rotation gate at ), and control the GCX pair on the bottom control qutrit with control value 1. The modified has the sign flipped for the matrix elements corresponding to the column indexed by in the Gray code and the row indexed by , i.e., all rows indexed by a -ary string that ends in 1.
We apply a search-based technique to find appropriate corrections to a non-invertible . Beginning with the lowest indexed column in the nullspace of , we perform a brute force search for the control qudit and control value, applying the correction and updating the nullspace if the rank of increases. If no suitable controlled operation is found, we increment the nullspace column index and repeat. We do this until the modified has full rank.
For the decomposition in Section IV A, after applying the six corrections in Table VII to the original , the modified matrix has full rank. Since each correction consists of two GCX gates and two additional depth, the modified decomposition uses 94 GCX gates, 82 gates, and has depth 176.
| Column | Control qudit | Control value |
| 3 | ||
| 9 | ||
| 27 | ||
| 48 | ||
| 59 | ||
| 65 |
Table VII. The six corrections needed to make the matrix invertible for the decomposition of with from Equation (9). Each correction is applied by conjugating the rotation gate corresponding to the (zero-indexed) column number of with a pair of GCX gates. In accordance with Equation (9), we order the control register such that the auxiliary qutrit is first, followed by the four control link qutrits. For clarity, we adopt the qutrit labels in Figure 9.
Alternate Decomposition
In Figure 21, we present an alternative decomposition of . Beginning from Figure 9, we apply Subspace Gating to the control links in addition to the plaquette links, enabling decompositions with shorter control sequences. The control sequences take advantage of the small size of set in Equation (3) in addition to the

Figure 21. Alternative decomposition of Figure 9 with Subspace Gating applied to the control links in addition to the plaquette links. The input subspace information on the control links is computed, with the output copied onto two auxiliary qutrits. The gate (top of Figure 18) combines the input subspace information of the control links with the plaquette links to inform the gating of the subroutine for opposite plaquette faces independently, maintaining the parallelized structure of the decomposition. The notch on the and gates indicate the qutrit holding the combined input subspace information. The input size of the gates depend on the particular term (Table VIII), with the circuit above showing the case for , , and as well as their transformations under symmetry, covering 7 of the 16 .
constraint.
For example, consider (row 1 of Table VIII), which applies controlled operations based on the control sequence
| order | gates | control links | |
| 1 | Four-input | ||
| 4 | Four-input | ||
| 4 | Three-input | ||
| 2 | Four-input | ||
| 4 | Two-input | ||
| 1 | No Subspace Gating |
Table VIII. The 16 separated into 4 groups by the input size of the applied gates. These gates are used in place of the four-input gates in Figure 21. The control links under Subspace Gating permute in tandem with the subspace and with the face of the cube applied (below, we apply to face A).
In this case, the control links are controlling only on a four qubit subspace. By Subspace Gating the control links, we can use the subroutine with the above, which uses 16 GCX gates and 16 gates, instead of with in Equation (9), which uses 94 GCX gates and 82 gates.
The set of control links covered by the gating subcircuit depends on the term. In Table VIII, we show the control links covered by the gates for each term. The required gates are the two-input and three-input versions from Figure 16, as well as the four-input gate in Figure 18. We note that the four-input gate can be implemented with Figure 16 using a single auxiliary ququart. For hardware implementations, it may be advantageous to invest in mixed-dimensional auxiliary qudits to reduce circuit width and depth.
We found that the decompositions worked best (invertible and fewer gates) for where has a large integer representation. Thus, we used decompositions for , permuting the order of the control qutrits to obtain the remaining . We use the modified decomposition from Appendix E to implement the case. We expect this dependence on ordering to be an artifact of the present implementation.
Table IX shows the resource count for two applied in parallel to opposite faces of the cube. Compared to the first row of Table III, we save GCX gates and depth is reduced to , which is around half the resources at the cost of three additional auxiliary qutrits (on a purely qutrit platform).
Simulation Results
Simulation Results
The global basis used to obtain the exact simulation results in Figure 10 (and shown in Table X) comprises only physical states of the cube. The observable is the expectation value of the electric operator of plaquette
| Time () | |
| 0.02 | 0.0059995205375 |
| 0.12 | 0.2013777323202 |
| 0.22 | 0.4735643212012 |
| 0.32 | 0.5530191097325 |
| 0.42 | 0.4390914030334 |
| 0.52 | 0.2780532551655 |
| 0.62 | 0.2269820889009 |
| 0.72 | 0.3062084922296 |
| 0.82 | 0.4407325362572 |
| 0.92 | 0.4749953963899 |
Table X. Exact simulation using the global basis of physical states on a cube. It took 16 seconds to generate this table.
We report the expectation values of the cube simulation for Trotter steps 1 and 2 in Table XI. We note the first timestep takes significantly longer to execute. This is due to the use of just-in-time (JIT) compilation with JAX, which is applied to multiple subroutines in the cube simulation. The first time a JIT-compiled function executes, a compiled binary is generated and cached for future use, yielding a significant speed boost.
| Time () | Runtime (seconds) | |
| 0.02 | 0.00607722 | 367.325 |
| 0.12 | 0.23791493 | 42.091 |
| 0.22 | 0.57592832 | 42.295 |
| 0.32 | 0.74714905 | 42.006 |
| 0.42 | 0.58758336 | 42.043 |
| 0.52 | 0.34950834 | 42.102 |
| 0.62 | 0.00206689 | 42.068 |
| 0.72 | 0.16572029 | 42.031 |
| 0.82 | 0.44481068 | 42.074 |
| 0.92 | 0.67250186 | 42.310 |
| 0.02 | 0.00603121 | 80.451 |
| 0.12 | 0.21797391 | 80.381 |
| 0.22 | 0.48899010 | 80.128 |
| 0.32 | 0.56093830 | 80.621 |
| 0.42 | 0.52352385 | 84.377 |
| 0.52 | 0.46890193 | 84.309 |
| 0.62 | 0.43751637 | 78.712 |
| 0.72 | 0.42842750 | 78.745 |
| 0.82 | 0.40681914 | 78.721 |
| 0.92 | 0.41583451 | 78.728 |
Table XI. The measured observable is the expectation value of for face , shown in Figure 10. The runtime of the first Trotter step is significantly longer due to overhead incurred by just-in-time (JIT) compilation, performed by JAX. After that, the runtime scales linearly with the number of Trotter steps . As an example, the average runtime for is 278 seconds. We use a time step of 0.05 but show every second expectation value for conciseness.
Hadamard Gate Removal
In the decomposition of the plaquette operator unitary evolution, the subspaces of and GCX gates originate from the subspace of , which are neighboring basis states in the SU(2) gauge theory. Thus, for a qutrit digitization of the gauge field, only configurations in Figure 22 appear. As can be seen in Figure 8, the gates on the left can be propagated through the circuit until they cancel with those on the right, transforming and GCX GCZ along the way. Therefore, we can remove all gates from the circuit, and by extension, the entire cube simulation.
Figure 22. All gates can be removed by flipping the GCX gate and changing the control value and subspace.
References
- [1]K. G. Wilson, Confinement of quarks, Phys. Rev. D 10, 2445 (1974).DOI
- [2]J. Kogut and L. Susskind, Hamiltonian formulation of Wilson’s lattice gauge theories, Phys. Rev. D 11, 395 (1975).DOI
- [3]B. Joó, C. Jung, N. H. Christ, W. Detmold, R. Edwards, M. Savage, and P. Shanahan (USQCD), Status and Future Perspectives for Lattice Gauge Theory Calculations to the Exascale and Beyond, Eur. Phys. J. A 55, 199 (2019), arXiv:1904.09725 [hep-lat].DOI
- [4]J. Carlson, M. J. Savage, R. Gerber, K. Antypas, D. Bard, R. Coffey, E. Dart, S. Dosanjh, J. Hack, I. Monga, et al., Nuclear Physics Exascale Requirements Review: An Office of Science Review sponsored jointly by Advanced Scientific Computing Research and Nuclear Physics, June 15 - 17, 2016, Gathersburg, Maryland, Tech. Rep. (Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States), 2022).
- [5]T. Byrnes and Y. Yamamoto, Simulating lattice gauge theories on a quantum computer, Phys. Rev. A 73, 022328 (2006), arXiv:quant-ph/0510027.DOI
- [6]M. C. Bańuls et al., Simulating Lattice Gauge Theories within Quantum Technologies, Eur. Phys. J. D 74, 165 (2020), arXiv:1911.00003 [quant-ph].DOI
- [7]M. Aidelsburger et al., Cold atoms meet lattice gauge theory, Phil. Trans. Roy. Soc. Lond. A 380, 20210064 (2021), arXiv:2106.03063 [cond-mat.quant-gas].DOI
- [8]C. W. Bauer, Z. Davoudi, A. B. Balantekin, T. Bhattacharya, M. Carena, W. A. de Jong, P. Draper, A. El-Khadra, N. Gemelke, M. Hanada, D. Kharzeev, H. Lamm, Y.-Y. Li, J. Liu, M. Lukin, Y. Meurice, C. Monroe, B. Nachman, G. Pagano, J. Preskill, E. Rinaldi, A. Roggero, D. J. Santiago, M. J. Savage, I. Siddiqi, G. Siopsis, D. Van Zanten, N. Wiebe, Y. Yamauchi, K. Yeter-Aydeniz, and S. Zorzetti, Quantum Simulation for High-Energy Physics, PRX Quantum 4, 027001 (2023).DOI
- [9]C. W. Bauer, Z. Davoudi, N. Klco, and M. J. Savage, Quantum simulation of fundamental particles and forces, Nature Rev. Phys. 5, 420 (2023), arXiv:2404.06298 [hep-ph].
- [10]E. Zohar, J. I. Cirac, and B. Reznik, Cold-Atom Quantum Simulator for SU(2) Yang-Mills Lattice Gauge Theory, Phys. Rev. Lett. 110, 125304 (2013), arXiv:1211.2241 [quant-ph].DOI
- [11]D. Banerjee, M. Bögli, M. Dalmonte, E. Rico, P. Stebler, U. J. Wiese, and P. Zoller, Atomic Quantum Simulation of U(N) and SU(N) Non-Abelian Lattice Gauge Theories, Phys. Rev. Lett. 110, 125303 (2013), arXiv:1211.2242 [cond-mat.quant-gas].DOI
- [12]E. Zohar, J. I. Cirac, and B. Reznik, Quantum simulations of gauge theories with ultracold atoms: local gauge invariance from angular momentum conservation, Phys. Rev. A 88, 023617 (2013), arXiv:1303.5040 [quant-ph].DOI
- [13]E. Zohar and M. Burrello, Formulation of lattice gauge theories for quantum simulations, Phys. Rev. D 91, 054506 (2015), arXiv:1409.3085 [quant-ph].DOI
- [14]E. Zohar, J. I. Cirac, and B. Reznik, Quantum Simulations of Lattice Gauge Theories using Ultracold Atoms in Optical Lattices, Rept. Prog. Phys. 79, 014401 (2016), arXiv:1503.02312 [quant-ph].DOI
- [15]E. Zohar and J. I. Cirac, Eliminating fermionic matter fields in lattice gauge theories, Phys. Rev. B 98, 075119 (2018), arXiv:1805.05347 [quant-ph].DOI
- [16]I. Raychowdhury and J. R. Stryker, Loop, string, and hadron dynamics in SU(2) Hamiltonian lattice gauge theories, Phys. Rev. D 101, 114502 (2020), arXiv:1912.06133 [hep-lat].DOI
- [17]N. Klco, J. R. Stryker, and M. J. Savage, SU(2) non-Abelian gauge field theory in one dimension on digital quantum computers, Phys. Rev. D 101, 074512 (2020), arXiv:1908.06935 [quant-ph].DOI
- [18]H. Lamm, S. Lawrence, and Y. Yamauchi (NuQS), General Methods for Digital Quantum Simulation of Gauge Theories, Phys. Rev. D 100, 034518 (2019), arXiv:1903.08807 [hep-lat].DOI
- [19]Z. Davoudi, I. Raychowdhury, and A. Shaw, Search for efficient formulations for Hamiltonian simulation of non-Abelian lattice gauge theories, Phys. Rev. D 104, 074505 (2021), arXiv:2009.11802 [hep-lat].DOI
- [20]A. J. Buser, H. Gharibyan, M. Hanada, M. Honda, and J. Liu, Quantum simulation of gauge theory via orbifold lattice, JHEP 09, 034, arXiv:2011.06576 [hep-th].DOI
- [21]Y. Ji, H. Lamm, and S. Zhu (NuQS), Gluon Field Digitization via Group Space Decimation for Quantum Computers, Phys. Rev. D 102, 114513 (2020), arXiv:2005.14221 [hep-lat].DOI
- [22]M. Khreshchuk, W. M. Kirby, G. Goldstein, H. Beauchemin, and P. J. Love, Quantum simulation of quantum field theory in the light-front formulation, Phys. Rev. A 105, 032418 (2022), arXiv:2002.04016 [quant-ph].
- [23]C. W. Bauer, M. Freytsis, and B. Nachman, Simulating Collider Physics on Quantum Computers Using Effective Field Theories, Phys. Rev. Lett. 127, 212001 (2021), arXiv:2102.05044 [hep-ph].DOI
- [24]A. Ciavarella, N. Klco, and M. J. Savage, Trailhead for quantum simulation of SU(3) Yang-Mills lattice gauge theory in the local multiplet basis, Phys. Rev. D 103, 094501 (2021), arXiv:2101.10227 [quant-ph].DOI
- [25]U.-J. Wiese, From quantum link models to D-theory: a resource efficient framework for the quantum simulation and computation of gauge theories, Phil. Trans. A. Math. Phys. Eng. Sci. 380, 20210068 (2021), arXiv:2107.09335 [hep-lat].DOI
- [26]J. F. Haase, L. Dellantonio, A. Celi, D. Paulson, A. Kan, K. Jansen, and C. A. Muschik, A resource efficient approach for quantum and classical simulations of gauge theories in particle physics, Quantum 5, 393 (2021), arXiv:2006.14160 [quant-ph].DOI
- [27]S. Kadam, I. Raychowdhury, and J. R. Stryker, Loop-string-hadron formulation of an SU(3) gauge theory with dynamical quarks, Phys. Rev. D 107, 094513 (2023), arXiv:2212.04490 [hep-lat].DOI
- [28]Z. Davoudi, A. F. Shaw, and J. R. Stryker, General quantum algorithms for Hamiltonian simulation with applications to a non-Abelian lattice gauge theory, Quantum 7, 1213 (2023), arXiv:2212.14030 [hep-lat].DOI
- [29]G. Pardo, T. Greenberg, A. Fortinsky, N. Katz, and E. Zohar, Resource-efficient quantum simulation of lattice gauge theories in arbitrary dimensions: Solving for Gauss’s law and fermion elimination, Phys. Rev. Res. 5, 023077 (2023), arXiv:2206.00685 [quant-ph].DOI
- [30]H. Liu, T. Bhattacharya, S. Chandrasekharan, and R. Gupta, Phases of 2D massless QCD with qubit regularization, Phys. Rev. D 111, 094511 (2025), arXiv:2312.17734 [hep-lat].arxiv.org/abs/2312.17734
- [31]T. V. Zache, D. González-Cuadra, and P. Zoller, Quantum and Classical Spin-Network Algorithms for q-Deformed Kogut-Susskind Gauge Theories, Phys. Rev. Lett. 131, 171902 (2023), arXiv:2304.02527 [quant-ph].DOI
- [32]A. N. Ciavarella, Quantum simulation of lattice QCD with improved Hamiltonians, Phys. Rev. D 108, 094513 (2023), arXiv:2307.05593 [hep-lat].arxiv.org/abs/2307.05593
- [33]I. D’Andrea, C. W. Bauer, D. M. Grabowska, and M. Freytsis, New basis for Hamiltonian SU(2) simulations, Phys. Rev. D 109, 074501 (2024), arXiv:2307.11829 [hep-ph].arxiv.org/abs/2307.11829
- [34]A. N. Ciavarella and C. W. Bauer, Quantum Simulation of SU(3) Lattice Yang-Mills Theory at Leading Order in Large-Nc Expansion, Phys. Rev. Lett. 133, 111901 (2024), arXiv:2402.10265 [hep-ph].arxiv.org/abs/2402.10265
- [35]D. M. Grabowska, C. F. Kane, and C. W. Bauer, A Fully Gauge-Fixed SU(2) Hamiltonian for Quantum Simulations, (2024), arXiv:2409.10610 [quant-ph].arxiv.org/abs/2409.10610
- [36]A. H. Z. Kavaki and R. Lewis, From square plaquettes to triamond lattices for SU(2) gauge theory, Commun. Phys. 7, 208 (2024), arXiv:2401.14570 [hep-lat].DOI
- [37]J. C. Halimeh, M. Hanada, S. Matsuura, F. Nori, E. Rinaldi, and A. Schäfer, A universal framework for the quantum simulation of Yang-Mills theory, (2024), arXiv:2411.13161 [quant-ph].arxiv.org/abs/2411.13161
- [38]A. N. Ciavarella, I. M. Burbano, and C. W. Bauer, Efficient Truncations of SU(Nc) Lattice Gauge Theory for Quantum Simulation, (2025), arXiv:2503.11888 [hep-lat].arxiv.org/abs/2503.11888
- [39]P. Balaji, C. Conefrey-Shinozaki, P. Draper, J. K. El-haderi, D. Gupta, L. Hidalgo, A. Lytle, and E. Rinaldi, Quantum Circuits for SU(3) Lattice Gauge Theory, (2025), arXiv:2503.08866 [hep-lat].arxiv.org/abs/2503.08866
- [40]H.-H. Lu, Z. Hu, M. S. Alshaykh, A. J. Moore, Y. Wang, P. Imaný, A. M. Weiner, and S. Kais, Quantum Phase Estimation with Time-Frequency Qudits in a Single Photon, Adv. Quantum Technol. 3, 1900074 (2020).DOI
- [41]A. Cervera-Lierta, M. Krenn, A. Aspuru-Guzik, and A. Galda, Experimental high-dimensional Greenberger-Horne-Zeilinger entanglement with superconducting transmon qutrits, Phys. Rev. Applied 17, 024062 (2022), arXiv:2104.05627 [quant-ph].
- [42]M. Ringbauer, M. Meth, L. Postler, R. Stricker, R. Blatt, P. Schindler, and T. Monz, A universal qudit quantum processor with trapped ions, Nature Phys. 18, 1053 (2022), arXiv:2109.06903 [quant-ph].DOI
- [43]A. Morvan, V. V. Ramasesh, M. S. Blok, J. M. Kreikebaum, K. O’Brien, L. Chen, B. K. Mitchell, R. K. Naik, D. I. Santiago, and I. Siddiqi, Qutrit Randomized Benchmarking, Phys. Rev. Lett. 126, 210504 (2021).DOI
- [44]P. Hrmo, B. Wilhelm, L. Gerster, M. W. van Mourik, M. Huber, R. Blatt, P. Schindler, T. Monz, and M. Ringbauer, Native qudit entanglement in a trapped ion quantum processor, Nature Commun. 14, 2242 (2023), arXiv:2206.04104 [quant-ph].DOI
- [45]N. Goss et al., High-fidelity qutrit entangling gates for superconducting circuits, Nature Commun. 13, 7481 (2022), [Erratum: Nature Commun. 14, 4256 (2023)], arXiv:2206.07216 [quant-ph].DOI
- [46]L. B. Nguyen, N. Goss, K. Siva, Y. Kim, E. Younis, B. Qing, A. Hashim, D. I. Santiago, and I. Siddiqi, Empowering a qudit-based quantum processor by traversing the dual bosonic ladder, Nature Commun. 15, 7117 (2024), arXiv:2312.17741 [quant-ph].
- [47]L. M. Seifert, Z. Li, T. Roy, D. I. Schuster, F. T. Chong, and J. M. Baker, Exploring ququart computation on a transmon using optimal control, Phys. Rev. A 108, 062609 (2023), arXiv:2304.11159 [quant-ph].DOI
- [48]P. J. Low, B. White, and C. Senko, Control and Readout of a 13-level Trapped Ion Qudit, (2023), arXiv:2306.03340 [quant-ph].arxiv.org/abs/2306.03340
- [49]C. Senko, P. Richerme, J. Smith, A. Lee, I. Cohen, A. Retzker, and C. Monroe, Realization of a Quantum Integer-Spin Chain with Controllable Interactions, Phys. Rev. X 5, 021026 (2015).
- [50]M. S. Blok, V. V. Ramasesh, T. Schuster, K. O’Brien, J. M. Kreikebaum, D. Dahlen, A. Morvan, B. Yoshida, N. Y. Yao, and I. Siddiqi, Quantum Information Scrambling on a Superconducting Qutrit Processor, Phys. Rev. X 11, 021010 (2021), arXiv:2003.03307 [quant-ph].DOI
- [51]D. M. Kurkcuoglu, M. S. Alam, J. A. Job, A. C. Y. Li, A. Macridin, G. N. Perdue, and S. Providence, Quantum simulation of ϕ⁴ theories in qudit systems, (2021), arXiv:2108.13357 [quant-ph].arxiv.org/abs/2108.13357
- [52]D. González-Cuadra, T. V. Zache, J. Carrasco, B. Kraus, and P. Zoller, Hardware Efficient Quantum Simulation of Non-Abelian Gauge Theories with Qudits on Rydberg Platforms, Phys. Rev. Lett. 129, 160501 (2022), arXiv:2203.15541 [quant-ph].
- [53]T. V. Zache, D. González-Cuadra, and P. Zoller, Fermion-qudit quantum processors for simulating lattice gauge theories with matter, Quantum 7, 1140 (2023), arXiv:2303.08683 [quant-ph].DOI
- [54]P. P. Popov, M. Meth, M. Lewenstein, P. Hauke, M. Ringbauer, E. Zohar, and V. Kasper, Variational quantum simulation of U(1) lattice gauge theories with qudit systems, Phys. Rev. Res. 6, 013202 (2024), arXiv:2307.15173 [quant-ph].DOI
- [55]J. C. Halimeh, M. Aidelsburger, F. Grusdt, P. Hauke, and B. Yang, Cold-atom quantum simulators of gauge theories, Nature Phys. 21, 25 (2025), arXiv:2310.12201 [cond-mat.quant-gas].arxiv.org/abs/2310.12201
- [56]A. Vezvaee, N. Earnest-Noble, and K. Najafi, Quantum simulation of Fermi-Hubbard model based on transmon qudit interaction, (2024), arXiv:2402.01243 [quant-ph].arxiv.org/abs/2402.01243
- [57]M. Illa, C. E. P. Robin, and M. J. Savage, Quantum simulations of SO(5) many-fermion systems using qudits, Phys. Rev. C 108, 064306 (2023), arXiv:2305.11941 [quant-ph].DOI
- [58]M. Meth et al., Simulating two-dimensional lattice gauge theories on a qudit quantum computer, Nature Phys. 21, 570 (2025), arXiv:2310.12110 [quant-ph].
- [59]F. Turro, I. A. Chernyshev, R. Bhaskar, and M. Illa, Qutrit and qubit circuits for three-flavor collective neutrino oscillations, Phys. Rev. D 111, 043038 (2025), arXiv:2407.13914 [quant-ph].DOI
- [60]M. Illa, C. E. P. Robin, and M. J. Savage, Qubits for quantum simulations of lattice quantum chromodynamics, Phys. Rev. D 110, 014507 (2024), arXiv:2403.14537 [quant-ph].
- [61]G. Calajò, G. Magnifico, C. Edmunds, M. Ringbauer, S. Montangero, and P. Silvi, Digital Quantum Simulation of a (1+1)D SU(2) Lattice Gauge Theory with Ion Qudits, PRX Quantum 5, 040309 (2024), arXiv:2402.07987 [quant-ph].
- [62]D. M. Kürkçüoglu, H. Lamm, and A. Maestri, Qudit Gate Decomposition Dependence for Lattice Gauge Theories, (2024), arXiv:2410.16414 [quant-ph].arxiv.org/abs/2410.16414
- [63]D. Gottesman, Fault tolerant quantum computation with higher dimensional systems, Chaos Solitons Fractals 10, 1749 (1999), arXiv:quant-ph/9802007.DOI
- [64]E. T. Campbell, Enhanced Fault-Tolerant Quantum Computing in d-Level Systems, Phys. Rev. Lett. 113, 230501 (2014).DOI
- [65]T. Inada, W. Jang, Y. Iiyama, K. Terashi, R. Sawada, J. Tanaka, and S. Asai, Measurement-Free Ultrafast Quantum Error Correction by Using Multi-Controlled Gates in Higher-Dimensional State Space, (2021), arXiv:2109.00086 [quant-ph].arxiv.org/abs/2109.00086
- [66]A. Cowtan, Qudit lattice surgery, (2022), arXiv:2204.13228 [quant-ph].arxiv.org/abs/2204.13228
- [67]B. L. Brock, S. Singh, A. Eickbusch, V. V. Sivak, A. Z. Ding, L. Frunzio, S. M. Girvin, and M. H. Devoret, Quantum error correction of qudits beyond break-even, Nature 641, 612 (2025), arXiv:2409.15065 [quant-ph].DOI
- [68]A. Muthukrishnan and C. R. Stroud, Multivalued logic gates for quantum computation, Phys. Rev. A 62, 052309 (2000).DOI
- [69]A. S. Nikolaeva, E. O. Kiktenko, and A. K. Fedorov, Efficient realization of quantum algorithms with qudits, EPJ Quant. Technol. 11, 43 (2024), arXiv:2111.04384 [quant-ph].DOI
- [70]B. P. Lanyon, M. Barbieri, M. P. Almeida, T. Jennewein, T. C. Ralph, K. J. Resch, G. J. Pryde, J. L. O’Brien, A. Gilchrist, and A. G. White, Simplifying quantum logic using higher-dimensional Hilbert spaces, Nature Physics 5, 134 (2009).
- [71]W.-D. Li, Y.-J. Gu, K. Liu, Y.-H. Lee, and Y.-Z. Zhang, Efficient universal quantum computation with auxiliary Hilbert space, Phys. Rev. A 88, 034303 (2013).DOI
- [72]W.-Q. Liu and H.-R. Wei, Optimal synthesis of the Fredkin gate in a multilevel system, New Journal of Physics 22, 063026 (2020).DOI
- [73]J. M. Baker, C. Duckering, P. Gokhale, N. C. Brown, K. R. Brown, and F. T. Chong, Improved Quantum Circuits via Intermediate Qutrits, ACM Transactions on Quantum Computing 1, 10.1145/3406309 (2020).DOI
- [74]A. Litteken, J. M. Baker, and F. T. Chong, Communication Trade Offs in Intermediate Qudit Circuits, in 2022 IEEE 52nd International Symposium on Multiple-Valued Logic (ISMVL) (2022) pp. 43–49.DOI
- [75]A. Saha, R. Majumdar, D. Saha, A. Chakrabarti, and S. Sur-Kolay, Asymptotically improved circuit for a d-ary Grover’s algorithm with advanced decomposition of the n-qudit Toffoli gate, Phys. Rev. A 105, 062453 (2022), arXiv:2012.04447 [quant-ph].DOI
- [76]R. Majumdar, A. Saha, A. Chakrabarti, and S. Sur-Kolay, Intermediate qudit-assisted Toffoli gate decomposition with quantum error correction, Quant. Inf. Proc. 23, 42 (2024).DOI
- [77]L. Lysaght, T. Goubault, P. Sinnott, S. Mansfield, and P.-E. Emeriau, Quantum circuit compression using qudit logic on qudits, (2024), arXiv:2411.03878 [quant-ph].
- [78]Y. Wang, Z. Hu, B. C. Sanders, and S. Kais, Qudits and High-Dimensional Quantum Computing, Frontiers in Physics 8, 10.3389/fphy.2020.589504 (2020).DOI
- [79]L. E. Fischer, D. Miller, F. Tacchino, P. K. Barkoutsos, D. J. Egger, and I. Tavernelli, Ancilla-free implementation of generalized measurements for qubits embedded in a qudit space, Phys. Rev. Res. 4, 033027 (2022).DOI
- [80]Y. Deller, S. Schmitt, M. Lewenstein, S. Lenk, M. Federer, F. Jendrejewski, P. Hauke, and V. Kasper, Quantum approximate optimization algorithm for qudit systems, Phys. Rev. A 107, 062410 (2023).DOI
- [81]G. Bottrill, M. Pandey, and O. Di Matteo, Exploring the Potential of Qutrits for Quantum Optimization of Graph Coloring, in 2023 International Conference on Quantum Computing and Engineering (2023) arXiv:2308.08050 [quant-ph].arxiv.org/abs/2308.08050
- [82]M. Zahidy et al., Practical high-dimensional quantum key distribution protocol over deployed multicore fiber, Nature Commun. 15, 1651 (2024).DOI
- [83]C. Popp, T. C. Sutter, and B. C. Hiesmayr, A Novel Stabilizer-based Entanglement Distillation Protocol for Qudits, (2024), arXiv:2408.02383 [quant-ph].DOI
- [84]D. L. Zhou, B. Zeng, Z. Xu, and C. P. Sun, Quantum computation based on d-level cluster state, Phys. Rev. A 68, 062303 (2003).DOI
- [85]S. S. Bullock, D. P. O’Leary, and G. K. Brennen, Asymptotically Optimal Quantum Circuits for d-Level Systems, Phys. Rev. Lett. 94, 230502 (2005).pubmed.ncbi.nlm.nih.gov/16090450
- [86]G. K. Brennen, S. S. Bullock, and D. P. O’Leary, Efficient Circuits for Exact-Universal Computations with Qudits, (2005), arXiv:quant-ph/0509161 [quant-ph].DOI
- [87]F. S. Khan and M. Perkowski, Synthesis of multi-qudit hybrid and d-valued quantum logic circuits by decomposition, Theoretical Computer Science 367, 336–346 (2006).DOI
- [88]Y. Nakajima, Y. Kawano, H. Sekigawa, M. Nakanishi, S. Yamashita, and Y. Nakashima, Synthesis of quantum circuits for d-level systems by using cosine-sine decomposition, Quantum Information & Computation 9, 423 (2009).
- [89]Y.-M. Di and H.-R. Wei, Synthesis of multivalued quantum logic circuits by elementary gates, Phys. Rev. A 87, 012325 (2013).DOI
- [90]M.-X. Luo, X.-B. Chen, Y.-X. Yang, and X. Wang, Geometry of Quantum Computation with Qudits, Scientific Reports 4, 4044 (2014).DOI
- [91]M. Luo and X. Wang, Universal quantum computation with qudits, Science China Physics, Mechanics & Astronomy 57, 1712 (2014).DOI
- [92]Y.-M. Di and H.-R. Wei, Optimal synthesis of multivalued quantum circuits, Phys. Rev. A 92, 062317 (2015).DOI
- [93]W. Zi, Q. Li, and X. Sun, Optimal Synthesis of Multi-Controlled Qudit Gates, (2023), arXiv:2303.12979 [quant-ph].arxiv.org/abs/2303.12979
- [94]G.-L. Jiang, W.-Q. Liu, and H.-R. Wei, Optimal synthesis of general multi-qudit quantum computation, (2023), arXiv:2310.11996 [quant-ph].arxiv.org/abs/2310.11996
- [95]K. Mato, S. Hillmich, and R. Wille, Mixed-dimensional qudit state preparation using edge-weighted decision diagrams, in Proceedings of the 61st ACM/IEEE Design Automation Conference (2024) pp. 1–6.DOI
- [96]S. Yang, L. Xu, G. Tian, and X. Sun, Quantum circuit synthesis with qudit phase gadget method, (2025), arXiv:2504.12710 [quant-ph].arxiv.org/abs/2504.12710
- [97]S. Prakash, A. Jain, B. Kapur, and S. Seth, Normal form for single-qutrit Clifford+T operators and synthesis of single-qutrit gates, Phys. Rev. A 98, 032304 (2018).DOI
- [98]L. Yeh and J. van de Wetering, Constructing All Qutrit Controlled Clifford+T gates in Clifford+T, in Reversible Computation: 14th International Conference, RC 2022, Urbino, Italy, July 5–6, 2022, Proceedings (Springer-Verlag, Berlin, Heidelberg, 2022) p. 28–50.
- [99]J. van de Wetering and L. Yeh, Building Qutrit Diagonal Gates from Phase Gadgets, EPTCS 394, 46 (2023), arXiv:2204.13681 [quant-ph].DOI
- [100]A. R. Kalra, M. Saikia, D. Valluri, S. Winnick, and J. Yard, Multi-qudit exact synthesis, (2024), arXiv:2405.08147 [quant-ph].arxiv.org/abs/2405.08147
- [101]A. N. Glaudell, N. J. Ross, J. van de Wetering, and L. Yeh, Exact Synthesis of Multiqudit Clifford-Cyclotomic Circuits, EPTCS 406, 44 (2024), arXiv:2405.08136 [quant-ph].DOI
- [102]A. R. Kalra, M. Mosca, and D. Valluri, Synthesis and Arithmetization of Single Qutrit Circuits, Quantum 9, 1647 (2025), arXiv:2311.08696 [quant-ph].DOI
- [103]E. J. Gustafson, H. Lamm, D. Liu, E. M. Murairi, and S. Zhu, Synthesis of Single Qutrit Circuits from Clifford+R, (2025), arXiv:2503.20203 [quant-ph].DOI
- [104]S. A Rahman, R. Lewis, E. Mendicelli, and S. Powell, SU(2) lattice gauge theory on a quantum annealer, Phys. Rev. D 104, 034501 (2021).
- [105]https://github.com/QSAR-UBC/QuditOpsViaLGT.
- [106]E. Marinari, The APE Computer and Lattice Gauge Theories, in Lattice Gauge Theory: A Challenge in Large-Scale Computing, edited by B. Bunk, K. H. Mütter, and K. Schilling (Springer US, Boston, MA, 1986) pp. 295–304.DOI
- [107]P. Boyle, D. Chen, N. Christ, M. Clark, S. Cohen, Z. Dong, A. Gara, B. Joo, C. Jung, L. Levkova, X. Liao, G. Liu, R. Mawhinney, S. Ohta, K. Petrov, T. Wettig, A. Yamaguchi, and C. Cristian, QCDOC: A 10 Teraflops Computer for Tightly-Coupled Calculations, in SC ’04: Proceedings of the 2004 ACM/IEEE Conference on Supercomputing (2004) pp. 40–40.DOI
- [108]N. H. Christ, QCDSP and QCDOC Computers, in Encyclopedia of Parallel Computing, edited by D. Padua (Springer US, Boston, MA, 2011) pp. 1668–1677.
- [109]A. Ciavarella, N. Klco, and M. J. Savage, Some Conceptual Aspects of Operator Design for Quantum Simulations of Non-Abelian Lattice Gauge Theories (2022) arXiv:2203.11988 [quant-ph].arxiv.org/abs/2203.11988
- [110]V. Bergholm, J. J. Vartiainen, M. Möttönen, and M. M. Salomaa, Quantum circuits with uniformly controlled one-qubit gates, Physical Review A 71, 10.1103/physreva.71.052330 (2005).DOI
- [111]M. Möttönen, J. J. Vartiainen, V. Bergholm, and M. M. Salomaa, Transformation of quantum states using uniformly controlled rotations, Quantum Info. Comput. 5, 467–473 (2005).arxiv.org/abs/quant-ph/0407010
- [112]N. Klco and M. J. Savage, Minimally entangled state preparation of localized wave functions on quantum computers, Physical Review A 102, 10.1103/physreva.102.012612 (2020).DOI
- [113]A. De Morgan, Formal logic: or, the calculus of inference, necessary and probable (Taylor and Walton, 1847).
- [114]V. Bergholm et al., PennyLane: Automatic differentiation of hybrid quantum-classical computations (2022), arXiv:1811.04968 [quant-ph].
- [115]https://github.com/QSAR-UBC/pennylane_more_qutrit_ops.
- [116]J. R. Stryker, Oracles for Gauss’s law on digital quantum computers, Phys. Rev. A 99, 042301 (2019), arXiv:1812.01617 [quant-ph].DOI
- [117]M. C. Tran, Y. Su, D. Carney, and J. M. Taylor, Faster Digital Quantum Simulation by Symmetry Protection, PRX Quantum 2, 010323 (2021), arXiv:2006.16248 [quant-ph].DOI
- [118]H. Lamm, S. Lawrence, and Y. Yamauchi (NuQS), Suppressing Coherent Gauge Drift in Quantum Simulations, (2020), arXiv:2005.12688 [quant-ph].arxiv.org/abs/2005.12688
- [119]N. H. Nguyen, M. C. Tran, Y. Zhu, A. M. Green, C. H. Alderete, Z. Davoudi, and N. M. Linke, Digital Quantum Simulation of the Schwinger Model and Symmetry Protection with Trapped Ions, PRX Quantum 3, 020324 (2022), arXiv:2112.14262 [quant-ph].DOI
- [120]E. Ballini, J. Mildenberger, M. M. Wauters, and P. Hauke, Symmetry verification for noisy quantum simulations of non-Abelian lattice gauge theories, (2024), arXiv:2412.07844 [quant-ph].DOI
- [121]C. Ball and T. D. Cohen, Zero effect suppression of gauge drift in quantum simulations, Phys. Rev. A 110, 022417 (2024), arXiv:2405.09462 [hep-lat].arxiv.org/abs/2405.09462
- [122]M. M. Wauters, E. Ballini, A. Biella, and P. Hauke, Symmetry-protection Zero phase transition in monitored lattice gauge theories, Phys. Rev. B 111, 094315 (2025), arXiv:2405.18504 [quant-ph].arxiv.org/abs/2405.18504
- [123]M. Carena, H. Lamm, Y.-Y. Li, and W. Liu, Quantum error thresholds for gauge-redundant digitizations of lattice field theories, Phys. Rev. D 110, 054516 (2024), arXiv:2402.16780 [hep-lat].arxiv.org/abs/2402.16780
- [124]S. Bravyi, J. M. Gambetta, A. Mezzacapo, and K. Temme, Tapering off qubits to simulate fermionic Hamiltonians, (2017), arXiv:1701.08213 [quant-ph].arxiv.org/abs/1701.08213
- [125]O. Di Matteo, A. McCoy, P. Gysbers, T. Miyagi, R. M. Woloshyn, and P. Navrátil, Improving Hamiltonian encodings with the Gray code, Phys. Rev. A 103, 042405 (2021), arXiv:2008.05012 [quant-ph].DOI

















