Parameterized Quantum Fourier Circuit

• The topic is defined by circuits that integrate Fourier-phase structures via fixed phase angles or trainable variational parameters.
• Research encompasses exact QFT decompositions, Fourier-series synthesis for function approximation, and connectivity-aware optimizations.
• Applications include algorithmic speed-ups in QFT routines, scalable hardware-specific designs, and variational quantum algorithm enhancements.

A parameterized quantum Fourier circuit is a quantum-circuit construction in which Fourier structure is realized through explicit gate parameters, symmetry-constrained phases, or parameter-dependent circuit templates. In the current literature, the expression does not denote a single canonical object. It can refer to an exact QFT realization whose phase angles are fixed by binary-fraction data, a Fourier-like variational template used to approximate structured unitaries such as $e^A$ for antisymmetric $A$, or a circuit that synthesizes a Fourier series rather than performing the basis-changing QFT itself (Romero et al., 11 Jul 2025, Daskin, 2 May 2025, Li et al., 2021).

1. Scope of the term in the literature

The literature associates the phrase with several technically distinct constructions. The common element is the use of Fourier-phase structure as the organizing principle of the circuit, but the role of the parameters varies substantially.

Usage	Core circuit form	Role of parameters
Exact QFT realization	Product of single-qubit phase states	Fourier phases such as $\varphi=\pi/2^m$
Fourier-like eigendecomposition	$P^\dagger D^\dagger F^\dagger \Lambda F D P$	Trainable layers $P,D,F,\Lambda$
Fourier-series synthesis	Controlled blocks $U_1,\dots,U_N$	Fourier coefficients and phase data
Connectivity-aware QFT	Cascades along graph paths	Hardware graph and covering paths

In the exact-QFT setting, the parameters are not variational degrees of freedom but prescribed Fourier angles. In the variational setting, the QFT appears as a structured subcircuit inside a larger parameterized ansatz. In the Fourier-series setting, the circuit is “Fourier” because it reconstructs a finite Fourier expansion in measurement statistics, not because it changes basis in the standard QFT sense. This suggests that “parameterized quantum Fourier circuit” is best treated as a family resemblance term rather than a uniquely defined circuit class (Romero et al., 11 Jul 2025, Daskin, 2 May 2025, Li et al., 2021, Khadiev et al., 10 Oct 2025).

2. Exact phase-parameterized realizations of the QFT

One explicit meaning is an exact QFT circuit whose parameters are the phases appearing in the standard product decomposition of Fourier basis states. For computational basis input $\ket{j}$, the factorized form is written as

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$

or equivalently

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$

Each factor

$$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$

is prepared by

$A$0

with

$A$1

The relevant phase parameters are explicitly

$A$2

so the circuit is parameterized by exact Fourier phases rather than by trainable angles (Romero et al., 11 Jul 2025).

This formulation motivates a recursive circuit assembled from Hadamards, phase gates, controlled phases, and swaps. The paper also introduces the pattern

$A$3

defined as “apply controlled-phase $A$4 and then swap.” The same phase-gate organization is used for an inverse-QFT subroutine in an alternative HHL implementation, where $A$5 applies bit reversal followed by reversed Hadamards and controlled phase rotations $A$6. The paper states that the new QFT code improves the standard one by about $A$7 seconds, and that the HHL example improves by about $A$8 seconds, while also noting that it does not provide a detailed asymptotic gate-count derivation or depth proof (Romero et al., 11 Jul 2025).

A higher-dimensional generalization appears in multilevel-qudit arithmetic. For $A$9 qudits of dimension $\varphi=\pi/2^m$0, with $\varphi=\pi/2^m$1, the QFT factorizes as

$\varphi=\pi/2^m$2

and the generalized controlled rotation is

$\varphi=\pi/2^m$3

Here the parameterization depends on the qudit dimension $\varphi=\pi/2^m$4, the register size $\varphi=\pi/2^m$5, and the phase increments $\varphi=\pi/2^m$6 (Pavlidis et al., 2017).

3. Fourier-like parameterized ansätze for structured operators

A second meaning arises in variational simulation, where the QFT is embedded into a structured eigendecomposition template. For a real antisymmetric matrix $\varphi=\pi/2^m$7, one paper approximates

$\varphi=\pi/2^m$8

using

$\varphi=\pi/2^m$9

The construction is based on a canonical antisymmetric matrix $P^\dagger D^\dagger F^\dagger \Lambda F D P$0 whose eigenvectors are geometric progressions and therefore form a phase-shifted QFT basis. The eigenbasis matrix is written as

$P^\dagger D^\dagger F^\dagger \Lambda F D P$1

where $P^\dagger D^\dagger F^\dagger \Lambda F D P$2 is the standard quantum Fourier transform matrix and

$P^\dagger D^\dagger F^\dagger \Lambda F D P$3

The conceptual point is that the diagonalizing basis is not the plain QFT basis but a phase-shifted version of it (Daskin, 2 May 2025).

The phase ramp is implemented as a tensor product of $P^\dagger D^\dagger F^\dagger \Lambda F D P$4-rotations,

$P^\dagger D^\dagger F^\dagger \Lambda F D P$5

with

$P^\dagger D^\dagger F^\dagger \Lambda F D P$6

The spectral layer is similarly encoded by

$P^\dagger D^\dagger F^\dagger \Lambda F D P$7

The resulting circuit uses $P^\dagger D^\dagger F^\dagger \Lambda F D P$8-rotations both to implement the phase-shifted Fourier basis and to encode the eigenspectrum. The role of $P^\dagger D^\dagger F^\dagger \Lambda F D P$9 is to implement row and column permutations and sign flips, with controlled $P,D,F,\Lambda$0, $P,D,F,\Lambda$1, and $P,D,F,\Lambda$2 gates suggested for that layer (Daskin, 2 May 2025).

In this framework, the QFT block is exact for the canonical structured core, while parameterization is used to adapt the circuit to a general antisymmetric matrix. Because the major blocks are separated rather than synthesized as a generic unitary, the total circuit size scales as $P,D,F,\Lambda$3 gates. Optimization is described through objectives such as

$P,D,F,\Lambda$4

or, when the target unitary is known,

$P,D,F,\Lambda$5

This makes the circuit “Fourier-like” and parameterized simultaneously: Fourier structure supplies the eigendecomposition template, while trainable parameters fit the spectral data of $P,D,F,\Lambda$6 (Daskin, 2 May 2025).

4. Fourier-series synthesis circuits beyond basis-changing QFT

Another use of the terminology concerns circuits that synthesize a Fourier expansion of a periodic function. In this setting, the target is

$P,D,F,\Lambda$7

with $P,D,F,\Lambda$8 adopted for convenience. The circuit contains $P,D,F,\Lambda$9 data qubits and $U_1,\dots,U_N$0 auxiliary qubits with

$U_1,\dots,U_N$1

and begins from

$U_1,\dots,U_N$2

A preprocessing module $U_1,\dots,U_N$3 prepares the control amplitudes

$U_1,\dots,U_N$4

after which $U_1,\dots,U_N$5 controlled blocks $U_1,\dots,U_N$6 encode the Fourier components (Li et al., 2021).

The circuit is parameterized by $U_1,\dots,U_N$7, chosen so that each block contributes

$U_1,\dots,U_N$8

and the measured output on the final qubit reproduces the scaled Fourier series: $U_1,\dots,U_N$9 The paper emphasizes that this is not the standard QFT. Its Fourier character lies in the synthetic realization of a Fourier series through parameterized controlled rotations, rather than a computational-basis-to-Fourier-basis transform. The overall time complexity is stated as

$\ket{j}$0

or

$\ket{j}$1

when measurement precision $\ket{j}$2 is included. The square-wave example sets $\ket{j}$3, omitting the corresponding blocks $\ket{j}$4 (Li et al., 2021).

This usage broadens the meaning of a parameterized quantum Fourier circuit. The circuit is Fourier-structured because Fourier coefficients and phases are encoded directly into gate parameters, yet the operational output is a measurement-derived function value rather than a QFT state (Li et al., 2021).

5. Connectivity-aware parameterization of QFT circuits

Parameterization also appears at the architecture level, where the QFT circuit depends on a hardware connectivity graph

$\ket{j}$5

In this formulation, the problem is to realize the QFT with minimal CNOT count under restricted two-qubit connectivity. The organizing subproblem is a covering path $\ket{j}$6 such that every vertex is either on the path or adjacent to a vertex on the path. Its cost is defined as

$\ket{j}$7

The interpretation given in the paper is that path steps correspond to movement of the active target qubit, while vertices in $\ket{j}$8 can supply a controlled-phase interaction without extending the path (Khadiev et al., 10 Oct 2025).

The algorithmic core is the $\ket{j}$9-covering path problem. An exact dynamic program uses

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$0

with

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$1

where $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$2. Its stated complexity is

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$3

For larger graphs, the paper gives an approximation based on a connected dominating set and a path-covering construction, yielding a

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$4

with time complexity

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$5

Here the circuit is parameterized not by phase variables but by graph-dependent paths and cascade orderings (Khadiev et al., 10 Oct 2025).

At the gate level, the paper uses the facts that a controlled-phase gate $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$6 can be implemented with 2 CNOTs and 3 $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$7 gates, and that a pair $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$8 plus $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \Bigl(\ket{0}+e^{2\pi i[0.b_0]}\ket{1}\Bigr) \Bigl(\ket{0}+e^{2\pi i[0.b_1b_0]}\ket{1}\Bigr) \cdots \Bigl(\ket{0}+e^{2\pi i[0.b_{n-1}\cdots b_0]}\ket{1}\Bigr),$$9 can be represented using 3 CNOTs. The resulting theorem states

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$0

A corollary gives

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$1

For the linear nearest-neighbor architecture, the paper reports

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$2

while for the 16-qubit “sun” and 27-qubit “two joint suns” architectures it reports 324 and 957 CNOTs, respectively. The same section states that these latter results improve by about $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$3 over an earlier arbitrary-graph construction that gives 342 and 1009 (Khadiev et al., 10 Oct 2025).

6. Physical embodiments, related architectures, and terminological boundaries

Substrate-specific implementations introduce a further sense of parameterization. In electric-circuit realizations, telegrapher wires and scattering widgets are used to synthesize the phase-shift family needed for QFT decompositions. The generalized phase gate $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$4 is produced by a multi-bridge mixer with

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$5

combined with path-delay sections implementing

$${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$6

This supplies the ladder of phases used in standard QFT circuits, including the explicit decompositions for $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$7 and $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$8. The implementation is parameterized by the target phase $${\rm QFT}\ket{j} = \frac{1}{\sqrt{2^n}} \prod_{m=0}^{n-1}\left(\ket{0}+\omega^{2^m j}\ket{1}\right), \qquad j=0,\dots,2^n-1.$$9, the number of bridges, the inductance ratio, and the chosen reflectionless momentum $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$0 (Ezawa, 2019).

A different physical realization uses a three-qubit circulant Hamiltonian. Because circulant matrices are diagonalized by Fourier modes, tuning the Hamiltonian to circulant form yields explicit three-qubit Fourier basis states with

$$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$1

The scheme performs an adiabatic mapping from computational basis states to Fourier modes, and the paper stresses that the eigenvectors do not depend on the magnitudes of $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$2, $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$3, $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$4, and $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$5 provided circulant symmetry is maintained. With sinusoidal modulation, it reports a gate fidelity of about $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$6 at $$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$7 ms for one parameter set, and introduces a counter-driving Hamiltonian

$$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$8

to accelerate the process (Yachi et al., 2021).

Other works sharpen the terminological boundary. In QFT-based PDE solvers, the circuit pattern is

$$\frac{1}{\sqrt{2}}\left(\ket{0}+e^{i\varphi}\ket{1}\right)$$9

with parameters supplied by the PDE, the time step, truncation order, and smoothness assumptions. The paper explicitly notes that these are parameter-dependent Fourier-space constructions rather than parameterized quantum circuits meant for optimization (Lubasch et al., 22 May 2025). Conversely, “Quantum circuit for the fast Fourier transform” treats the reversible implementation of the classical FFT on basis-encoded registers and states that it is “essentially different” from the QFT, so it belongs to a different circuit family despite its Fourier content (Asaka et al., 2019).

A recurrent misconception is therefore that every parameterized Fourier circuit is either a variational QFT or a standard QFT decomposition. The literature does not support that uniform reading. Some constructions are exact and phase-prescriptive, some are genuinely variational, some are architecture-adaptive, and some are Fourier-synthesis devices whose output is a probability or expectation rather than a QFT state (Romero et al., 11 Jul 2025, Daskin, 2 May 2025, Li et al., 2021, Lubasch et al., 22 May 2025).