Qubitized Hamiltonians in Quantum Algorithms

• Qubitized Hamiltonians are Hamiltonians encoded on qubits using Pauli decompositions, enabling quantum algorithms to extract eigenphases directly.
• They employ techniques like block-encoding, LCU-based walk operators, and diagonal mappings from classical functions to streamline quantum simulation.
• Their applications span quantum chemistry, lattice gauge theories, and optimization, although challenges include computational complexity and encoding trade-offs.

Qubitized Hamiltonians are Hamiltonians encoded on qubits in a form compatible with quantum algorithms built from Pauli decompositions, block-encodings, and controlled signal operations. In the broad sense used across several recent works, they are Hamiltonians written as sums of tensor products of Pauli operators acting on qubits; in the narrower Low–Chuang sense, they are Hamiltonians embedded into a larger unitary so that their eigenvalues appear as eigenphases of a walk or signal operator. A further specialized usage defines qubitized Hamiltonians as Hermitian matrices with spectrum in $\{-1,0,+1\}$ and zero trace, so that in an appropriate basis they act as a single-qubit Pauli on a logical two-dimensional subspace. A complementary foundational route derives qubit Hamiltonians directly from classical functions by mapping Boolean and real functions to diagonal Pauli-$Z$ operators (Gunderman et al., 2023, Steudtner et al., 2019, Ollive et al., 18 Mar 2026, Hadfield, 2018).

1. Definitions and scope

The broadest formulation begins with a Pauli expansion

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$

or equivalently

$H=\sum_{i=1}^m c_iP_i,$

with real coefficients and $n$-qubit Pauli strings. In this sense, a qubitized Hamiltonian is simply a Hamiltonian already expressed on qubits in a Pauli basis, which is the standard representation used after mappings such as Jordan–Wigner or Bravyi–Kitaev, or after direct spin encodings (Gunderman et al., 2023, Bu et al., 26 Jan 2026).

In the qubitization framework of Hamiltonian simulation, the central object is instead a unitary query built from an LCU decomposition

$H=\sum_j w_j U_j$

or

$H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$

together with state preparation and controlled selection subroutines. The qubitized walk operator acts on a two-dimensional invariant subspace associated with each eigenstate of $H$, and its eigenvalues are

$$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$

so phase estimation yields $\pm \arccos(E)$ rather than simulating $Z$0 directly (Steudtner et al., 2019, Berry et al., 2019).

A distinct, paper-specific definition treats qubitized Hamiltonians as Hermitian matrices $Z$1 with eigenvalues in $Z$2 and $Z$3. In that usage, $Z$4 is written as

$Z$5

where $Z$6 is a logical Pauli $Z$7 on $Z$8 and zero on the orthogonal subspace. This definition emphasizes “single-qubit behavior when expressed in the appropriate basis” and is used to organize Hamiltonian simulation and block-encoding primitives for structured matrices (Ollive et al., 18 Mar 2026).

These usages are compatible rather than contradictory. A plausible implication is that “qubitized Hamiltonian” names a family of representations rather than one canonical construction: Pauli-sum encodings, block-encoded signal operators, and logical two-level reductions all describe the same underlying objective of translating problem Hamiltonians into qubit-native primitives.

2. Diagonal encodings of classical functions

A foundational construction maps a classical function $Z$9 to the unique diagonal Hamiltonian $$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$0 satisfying

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$1

Using Boolean Fourier analysis,

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$2

and the identity

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$3

one obtains

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$4

Thus every real function on the Boolean cube has a unique diagonal Pauli-$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$5 representation, and Boolean predicates become projectors or projector sums in the computational basis (Hadfield, 2018).

This representation directly supplies cost Hamiltonians for QAOA, problem Hamiltonians for quantum annealing, and diagonal or controlled Hamiltonians whose exponentials act as phase oracles. The same work gives explicit composition rules: $$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$6

$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$7

These formulas turn logical structure into algebraic structure and permit systematic construction from clauses such as AND, OR, XOR, implication, majority, not-all-equal, and $$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$8-in-$$\mathcal H=\sum_{i=1}^{N}\omega_i P_i,\qquad P_i\in \{I,X,Y,Z\}^{\otimes n},$$9 predicates (Hadfield, 2018).

The same source draws a sharp complexity distinction. For compact Boolean descriptions such as CNF formulas, computing even the identity coefficient

$H=\sum_{i=1}^m c_iP_i,$0

is #P-hard. By contrast, for pseudo-Boolean objectives of the form

$H=\sum_{i=1}^m c_iP_i,$1

with each $H=\sum_{i=1}^m c_iP_i,$2 local, the Hamiltonian is simply

$H=\sum_{i=1}^m c_iP_i,$3

and remains efficiently representable. This distinction is central to the practical use of qubitized Hamiltonians in optimization: exact projector encodings of global SAT predicates are generically intractable, while sums of local clause Hamiltonians are not (Hadfield, 2018).

3. Block-encoding, walk operators, and controlled Hamiltonians

In LCU-based qubitization, PREPARE constructs

$H=\sum_{i=1}^m c_iP_i,$4

and SELECT applies

$H=\sum_{i=1}^m c_iP_i,$5

The resulting signal operator block-encodes $H=\sum_{i=1}^m c_iP_i,$6, and the walk operator built from SELECT and a reflection about $H=\sum_{i=1}^m c_iP_i,$7 is the object on which phase estimation is performed (Berry et al., 2019).

A related presentation writes the target Hamiltonian as

$H=\sum_{i=1}^m c_iP_i,$8

with $H=\sum_{i=1}^m c_iP_i,$9 Pauli strings. In that formulation, qubitization avoids Trotter error because it does not approximate $n$0 by a product formula; instead it constructs exact unitaries $n$1, $n$2, and $n$3 whose joint action maps the eigenvalues of $n$4 to phases $n$5 (Steudtner et al., 2019).

Diagonal function Hamiltonians fit naturally into this framework. If $n$6 is the predicate Hamiltonian for a Boolean function $n$7, then controlled evolution is implemented by

$n$8

and computation of $n$9 into an ancilla is realized by

$H=\sum_j w_j U_j$0

These are exactly the kinds of controlled unitaries and oracles that are repackaged in block-encoding and qubitization-based constructions (Hadfield, 2018).

A more recent parallel framework, Hamiltonian Decoded Quantum Interferometry, also starts from a general Pauli Hamiltonian

$H=\sum_j w_j U_j$1

but prepares

$H=\sum_j w_j U_j$2

for a polynomial $H=\sum_j w_j U_j$3, using a controlled-Pauli layer, Bell interferometry, and a decoding oracle for the symplectic code of $H=\sum_j w_j U_j$4. The authors explicitly note that the construction is not phrased in block-encoding terminology, but it plays a similar role: a spectral polynomial transformation is realized on a maximally entangled reference state rather than through repeated signal-operator applications (Bu et al., 26 Jan 2026).

4. Compression, symmetry reduction, and alternative encodings

For Pauli Hamiltonians, qubitization is not only a matter of encoding $H=\sum_j w_j U_j$5 but also of minimizing the qubit register on which $H=\sum_j w_j U_j$6 acts. One systematic approach starts from

$H=\sum_j w_j U_j$7

and removes first all redundant qubits that are acted on by identities only, then all qubits fixed by conserved Pauli charges. The reduced Hamiltonian decomposes as

$H=\sum_j w_j U_j$8

with sector Hamiltonians

$H=\sum_j w_j U_j$9

The corresponding optimal qubit count is

$H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$0

with at most $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$1 subproblems, where $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$2 is the commutation matrix of a generating subset of Pauli terms (Gunderman et al., 2023).

This reduction is exact within the class of Clifford conjugations and Pauli symmetries. It preserves spectrum and sector dynamics, and it is explicitly motivated as a preprocessing step for Trotter simulation, VQE, and Low–Chuang qubitization alike. The same formalism is applied to chemical molecules, lattice gauge theories, the Hubbard model, and the Kitaev model, where conserved charges can remove nontrivial fractions of the original qubit register (Gunderman et al., 2023).

A different encoding trade-off appears in first-quantized rotor models. For lattices of dipolar planar rotors, a binary encoding uses

$H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$3

qubits per truncated rotor level $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$4, so the kinetic term becomes a quadratic Ising-like operator in $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$5, while the interaction terms become Pauli strings of weight up to $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$6. The unary encoding instead uses $H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$7 qubits per rotor with a one-hot physical subspace, making the kinetic term purely diagonal and 1-local and the interactions constant-weight four-body terms. This is a direct example of how different qubitized Hamiltonians can represent the same physical model with different qubit counts, Pauli weights, and circuit-depth consequences (Moeed et al., 23 Jul 2025).

These two lines of work address different bottlenecks. Conserved-charge reduction lowers the number of logical qubits for a fixed Pauli Hamiltonian, whereas binary versus unary encodings alter the Pauli structure itself. A plausible implication is that qubitized Hamiltonian design is jointly an algebraic and architectural problem.

5. Structured matrices, chemistry, and polynomial spectral transforms

Quantum chemistry provides a canonical large-scale application. For an arbitrary-basis electronic structure Hamiltonian, low-rank factorization rewrites the Coulomb tensor as

$H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$8

leading to LCU 1-norms

$H=\sum_{k=1}^{\Lambda}\alpha_k\,p^k_{\mathcal S},$9

and $H$0. With QROAM-based PREPARE and controlled Jordan–Wigner SELECT, the paper reports a variant with $H$1 T complexity and estimates about seven hundred times less surface code spacetime volume than prior quantum algorithms for FeMoco, despite using a larger and more accurate active space (Berry et al., 2019).

Another line of work reformulates structured matrices themselves as sums of qubitized Hamiltonians. In that usage, Toeplitz, circulant, Hankel, anti-circulant, grid, and outer-product matrices are decomposed into blocks that, after basis change, reduce to projector-controlled one-qubit gates. Each term can then be simulated by a single logical rotation $H$2, or block-encoded by writing the term as a difference of two reflections. This provides a uniform LCH-to-LCU conversion for many adjacency matrices associated with PDE discretizations, sparse matrices, and graph problems (Ollive et al., 18 Mar 2026).

Hamiltonian Decoded Quantum Interferometry addresses a different task: preparation of Gibbs-like states and Hamiltonian optimization for general Pauli Hamiltonians. For polynomial degree $H$3, the framework uses a single call to a weight-$H$4 decoding oracle and prepares $H$5, with robustness bound

$H$6

For Gibbs-state preparation, the paper imports the degree condition

$H$7

to obtain trace-distance approximation to $H$8 (Bu et al., 26 Jan 2026).

These examples illustrate two complementary roles of qubitized Hamiltonians. In chemistry and structured linear algebra they are optimized queries for simulation and phase estimation; in HDQI they are inputs to a one-shot spectral transformation scheme based on decoding rather than repeated walk-operator use.

6. Implementations, misconceptions, and limitations

The broad sense of qubitized Hamiltonians includes physically programmed qubit Hamiltonians with explicit control over graph structure and local fields. In a Rydberg-atom simulator, the microscopic Hamiltonian

$H$9

is recast as

$$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$0

with

$$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$1

The platform realizes connected three-, four-, and six-qubit graphs in two and three dimensions and reports spectroscopic resolution $$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$2, providing a direct experimental testbed for small graph-based Ising Hamiltonians (Kim et al., 2020).

Several common misconceptions are addressed by the literature. One is that qubitization refers only to the Low–Chuang walk-operator framework; in fact, the term is also used more broadly for any Hamiltonian encoded on qubits in a controlled Pauli form, and some foundational function-to-Hamiltonian constructions predate the later terminology (Hadfield, 2018). Another is that exact Pauli expansions of compact Boolean predicates should be easy because the predicates are concisely described; the opposite is true in general, since even the identity coefficient of the exact diagonal Hamiltonian can be #P-hard to compute (Hadfield, 2018). A third is that reducing qubit count is always free; conserved-charge compression is optimal only within Clifford transformations and Pauli symmetries, and unary encodings can lower Pauli weight only by increasing the number of qubits (Gunderman et al., 2023, Moeed et al., 23 Jul 2025).

The main limitations are correspondingly structural. Arbitrary Boolean functions can require $$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$3; arbitrary compact SAT predicates remain #P-hard to Fourier-expand; Low–Chuang qubitization trades exactness for additional ancillas and nontrivial PREPARE and SELECT logic; Toffoli-free unary schemes remove multi-controlled gates at the cost of $$\lambda_\pm(E)=E\pm i\sqrt{1-E^2}=e^{\pm i\arccos E},$$4 ancilla qubits; HDQI is efficient only when decoding or anticommutation structure is favorable; and physically programmed Ising Hamiltonians remain limited by coherence, blockade imperfections, and calibration errors (Hadfield, 2018, Steudtner et al., 2019, Bu et al., 26 Jan 2026, Kim et al., 2020).

Taken together, these results establish qubitized Hamiltonians as a unifying language for diagonal predicate encodings, Pauli-sum simulation inputs, block-encoded walk operators, symmetry-reduced effective models, and physically programmed spin Hamiltonians. The unifying principle is not a single algorithmic primitive but a shared requirement: the Hamiltonian must be expressed on qubits in a form whose algebra, symmetries, and control structure can be directly exploited by quantum hardware or by higher-level quantum algorithms.