--- title: Quantum Fourier Transform with Measurement url: https://www.emergentmind.com/topics/quantum-fourier-transform-with-measurement-qft-m type: topic --- # Quantum Fourier Transform with Measurement Searching arXiv for the cited QFT+M papers and related work to ground the article. Quantum Fourier Transform with Measurement (QFT+M) denotes a family of procedures in which a quantum Fourier transform is used specifically as a prelude to measurement, so that Fourier-domain structure is extracted as classical data rather than preserved as a coherent state for subsequent unitary processing. In the qubit setting, this includes the semiclassical Griffiths–Niu reformulation of the QFT when the output is immediately measured; in photonic interferometry, it includes implementing the Fourier unitary on path, polarization, or orbital-angular-momentum modes and then performing photon counting in the output basis; in period-finding settings, it is the measurement stage that converts a Fourier spectrum into an informative integer outcome. Across these settings, QFT+M is characterized by the same operational motif: the QFT maps phase, periodicity, or modal structure into a measurement distribution whose peaks, suppressions, or bitwise outcomes encode the quantity of interest [2403.09514], [1708.00296], [1010.0033]. ## 1. Formal definition and operational scope For an \(N=2^n\)-dimensional qubit register, the quantum Fourier transform is the unitary \[ F_N\ket{x}=\frac{1}{\sqrt{N}}\sum_{k=0}^{N-1} e^{\frac{2\pi i xk}{N}}\ket{k}. \] If the input state is \[ \ket{\psi}=\sum_{x=0}^{N-1}\alpha_x\ket{x}, \] then after the QFT the amplitude of \(\ket{k}\) is the discrete Fourier coefficient \[ \frac{1}{\sqrt{N}}\sum_{x=0}^{N-1}\alpha_x e^{\frac{2\pi i xk}{N}}, \] and a computational-basis measurement yields \[ p(k)=\frac{1}{N}\left|\sum_{x=0}^{N-1}\alpha_x e^{\frac{2\pi i xk}{N}}\right|^2. \] QFT+M is precisely this “apply \(F_N\), then measure” pattern [2502.10840]. In photonic linear optics, the same structure is expressed at the mode level. The \(N\)-mode QFT acts on creation operators by \[ \hat b_j^\dagger=\sum_{k=0}^{N-1}(F_N)_{jk}\hat a_k^\dagger, \qquad (F_N)_{jk}=\frac{1}{\sqrt{N}}e^{2\pi i jk/N}, \] and the measurement stage is projective detection in the output Fock basis of the spatial or polarization modes. In that setting, the “\(+M\)” refers to photon counting after the Fourier interferometer, or after an inserted phase operation and an \(F_N^\dagger\) analysis stage in metrology [1708.00296]. The scope of QFT+M is narrower than that of the fully coherent unitary QFT. A central limitation is that semiclassical replacement of controlled phases by measurement and feed-forward is valid only when the QFT output is immediately measured; if the coherent Fourier-domain state must be consumed by later unitary subroutines, the full unitary QFT is still required [2403.09514]. This distinction recurs throughout the literature: some constructions exploit measurement only at the end, some exploit mid-circuit measurements to reduce entangling-gate cost, and some use measurement-assisted internal gadgets while preserving a coherent final output. ## 2. Semiclassical and dynamic-circuit realizations on qubits The canonical algorithmic form of QFT+M is the Griffiths–Niu semiclassical trick. When the QFT is followed immediately by computational-basis measurement, the controlled two-qubit phase gates of the standard decomposition can be removed and replaced by a sequence of single-qubit rotations conditioned on already measured lower-significance bits. Measuring qubits from least significant bit to most significant bit, the phase applied to qubit \(j\) is \[ \phi_j = 2\pi(0.b_{j-1}b_{j-2}\dots b_0) = 2\pi\sum_{m=0}^{j-1}\frac{b_m}{2^{\,j-m}}, \] followed by \(H_j\) and measurement of qubit \(j\). With the running fractional phase \[ s_{j+1}=\frac{s_j}{2}+\frac{b_j}{2}, \] the classical update cost is \(O(1)\) per qubit [2403.09514]. This reformulation changes the resource profile fundamentally. A standard unitary QFT uses \(O(n^2)\) two-qubit controlled-phase gates, plus routing overhead on restricted-connectivity architectures. The dynamic-circuit QFT+M replaces these by \(O(n)\) mid-circuit measurements and \(O(n)\) classically controlled single-qubit rotations and Hadamards, requires no two-qubit gates, and has no connectivity constraints. The remaining depth is dominated not by entangling-gate layers but by sequential readout and feed-forward latency [2403.09514]. A large-scale experimental realization on IBM superconducting hardware demonstrates the practical impact of this reduction. On `ibm_kyiv`, dynamic-circuit QFT+M with mid-circuit measurement, feed-forward conditional operations, and qubit resets achieved certified process fidelities \(>50\%\) up to \(16\) qubits and \(>1\%\) up to \(37\) qubits, exceeding previous reports across platforms. For a periodic \(10\)-qubit input state, the dynamic implementation produced sharply peaked output distributions close to ideal, with measured \(\mathcal{F}_{\mathrm{proc}}\) of about \(88\%\), whereas the unitary implementation on the same hardware yielded about \(6\%\) [2403.09514]. The dominant hardware issue in this setting is idle-qubit decoherence during readout and feed-forward. The same work introduces “feed-forward-compensated dynamical decoupling” (FC-DD), tailored to dynamic circuits with hardware timing constraints. The reported timing parameters are \(\tau_{\text{readout}}=1244\,\mathrm{ns}\), comprising a \(781\,\mathrm{ns}\) measurement pulse and a \(463\,\mathrm{ns}\) delay, and \(\tau_{\mathrm{FF}}=653\,\mathrm{ns}\) of classical feed-forward latency. The protocol uses an \(X2\) sequence straddling \(\tau_{\mathrm{FF}}\) and an \(XY4\) sequence in the remaining \(\tau=\tau_{\text{readout}}-\tau_{\mathrm{FF}}=591\,\mathrm{ns}\) idle window, thereby mitigating measurement-induced dephasing without violating scheduling constraints [2403.09514]. The same paper also gives a process-fidelity certification tailored to “unitary followed by measurement,” with efficient state preparation via virtual \(R_Z\) frame updates and sampling over \(m=20\) inputs with \(2000\) shots each. This is significant because ordinary classical result-fidelity metrics do not certify the underlying quantum process. A common misconception is that QFT+M is merely an implementation shortcut; the dynamic-circuit results show instead that, when the algorithmic contract is “QFT followed immediately by measurement,” the semiclassical form is the exact primitive of interest rather than an approximation [2403.09514]. ## 3. Measurement and feed-forward inside coherent approximate QFT A distinct line of work uses measurement and feed-forward not to measure the QFT output, but to reduce the fault-tolerant synthesis cost of a fully coherent approximate QFT. In this construction, small controlled \(Z\)-rotations are converted into uncontrolled \(Z^a\) rotations by ancilla-assisted measurement gadgets, and these rotations are then batched into phase-gradient operations driven by a reusable resource state. The resulting fully coherent AQFT achieves \(O(n\log n)\) T-count rather than the \(O(n\log^2 n)\) scaling of gate-by-gate synthesis [1803.04933]. The starting point is the standard exact QFT decomposition into Hadamards and controlled rotations \[ R_k=\operatorname{diag}\!\left(1,e^{2\pi i/2^k}\right), \] followed by qubit reversal. Approximation is introduced by discarding rotations below a threshold corresponding to a cutoff \(b\), leaving \(O(nb)\) retained rotations. Choosing \[ b=\Theta\!\big(\log(n/\varepsilon)\big) \] ensures \[ \|U_{\mathrm{QFT}}-U_{\mathrm{AQFT}(b)}\|=O(n/2^b)\le \varepsilon. \] The measurement-assisted construction then keeps the coherent output while using measured ancillae and classical corrections internally [1803.04933]. Its central resource is the \((b+1)\)-qubit phase-gradient state \[ |v_{b+1}\rangle = \frac{1}{\sqrt{2^{b+1}}} \sum_{j=0}^{2^{b+1}-1}e^{-2\pi i j/2^{b+1}}|j\rangle, \] which is an eigenstate of an in-place adder \(A_{b+1}\) in the sense that \[ A_{b+1}\big(|k\rangle\otimes |v_{b+1}\rangle\big) = e^{2\pi i k/2^{b+1}}|k\rangle\otimes |v_{b+1}\rangle. \] Because the resource is returned unchanged, it can be reused across layers. The one-time T-cost to prepare it is \[ T_{\mathrm{prep}}=O\!\big(b\log(b/\alpha)\big) \] using RUS synthesis of the requisite \(Z\)-axis rotations [1803.04933]. At circuit level, each controlled rotation is replaced by a measurement-based gadget with constant T cost, after which the corresponding uncontrolled rotations are induced collectively by a phase-gradient slice. The paper reports a total T-count \[ T_{\mathrm{total}}=8n(b-1)+O\!\big(b\log(b/\varepsilon)\big), \] with logical-qubit count \[ n_g=n+3b-4. \] For \(b=13\), explicit T-counts include \(18{,}876\) for AQFT128, \(38{,}844\) for AQFT256, \(78{,}780\) for AQFT512, and \(392{,}842\) for AQFT4096, corresponding to approximately \(10\)–\(12\times\) T-count reductions relative to prior gate-synthesized AQFT implementations [1803.04933]. This usage broadens the meaning of “QFT with measurement” in the fault-tolerant literature. The measurements are not the terminal readout of the Fourier basis; rather, they are internal control primitives that enable a cheaper coherent transformation. A plausible implication is that QFT+M should be understood not as a single circuit family, but as a design principle: measurement may either terminate the Fourier computation or be inserted strategically to reduce synthesis overhead without sacrificing coherence [1803.04933]. ## 4. Photonic interferometric QFT+M In linear optics, QFT+M takes a physically different form. The QFT is realized as a multiport interferometer on path, polarization, or orbital-angular-momentum modes, and measurement is photon counting in the resulting output basis. A prominent implementation uses a polarization–path factorization \[ F_{2d} = P\left[\bigoplus_{k=0}^{d-1}H\!\left(\frac{k\pi}{d}\right)\right](F_d\oplus F_d)P, \] where \(H(\theta)\) mixes horizontal and vertical polarizations within each path and \(P\) permutes the mode ordering. This construction uses polarization as an extra rail, removes entire layers of \(50{:}50\) path couplers, and reduces beam splitters by as much as \(75\%\) relative to path-only implementations in the same Reck/Clements framework; for FFT-like \(O(m\log m)\) QFT decompositions it still yields \(>50\%\) reduction [1708.00296]. Small-\(N\) realizations were demonstrated explicitly. For \(N=2\), a single non-polarizing \(50{:}50\) beam splitter implements \(F_2\). For \(N=3\), three modes were encoded as \((|H\rangle_1,|V\rangle_1,|H\rangle_2)\) and implemented using a polarization-dependent beam splitter with reflectivities \(R_H=1/3\) and \(R_V=1\), plus wave-plate phases. For \(N=4\), the encoding \((|H\rangle_1,|V\rangle_1,|H\rangle_2,|V\rangle_2)\) was realized with a single \(50{:}50\) beam splitter and polarization transformations \(H\) and \(HS=H\cdot S\), where \(S=\mathrm{diag}(1,i)\) [1708.00296]. These interferometers were used to observe generalized Hong–Ou–Mandel suppression in \(F_n\) multiports for \(n=2,3,4\) photons, with transition amplitudes governed by permanents. The measured output-distribution fidelities were \(0.973\pm0.001\) for \(n=2\), \(0.871\pm0.004\) for \(n=3\), and \(0.765\pm0.008\) for \(n=4\). Suppression-law violation ratios for indistinguishable photons were \(\nu_2=0.052\pm0.001\), \(\nu_3=0.24\pm0.01\), and \(\nu_4=0.41\pm0.03\), all well below the distinguishable-photon baselines \(0.47\pm0.01\), \(0.68\pm0.08\), and \(0.75\pm0.14\). The average second-order-correlation witness \(\overline{G}_n\) yielded \(0.052\pm0.005\), \(0.396\pm0.007\), and \(0.556\pm0.011\), below the classical bounds \(1-1/n\), confirming genuine multiphoton interference [1708.00296]. The same architecture operationalized QFT+M for metrology through a multimode Mach–Zehnder interferometer \(F_n\rightarrow \Phi(\phi)\rightarrow F_n^\dagger\). With one indistinguishable single photon injected into each input mode, the first QFT generates number–path entanglement deterministically; the phase operation \(\Phi(\phi)\) imprints the unknown parameter; the inverse QFT converts the phase information back into structured counting probabilities. For the delta-phase scheme \(f_j^\delta=\delta_{j,1}\), measured phase sensitivities were \(0.515\pm0.013\) for \(n=2\), \(0.491\pm0.015\) for \(n=3\), and \(0.458\pm0.027\) for \(n=4\), beating the corresponding shot-noise limits \(0.707\), \(0.577\), and \(0.500\) deterministically. The \(n=2\) case reaches the Heisenberg limit ideally, whereas \(n=3,4\) lie between shot-noise and Heisenberg scaling, consistent with theory. The same paper emphasizes that linear-phase superresolution does not by itself constitute a genuine quantum advantage, because the apparent \(O(1/n^{3/2})\) scaling is attributable to a multiple-phase resource counting issue [1708.00296]. A second photonic direction realizes single-photon high-dimensional QFT+M using orbital angular momentum and path as dual degrees of freedom. There the transform is decomposed as \[ F_{\mathrm{OAM}}^{(d)} = \mathrm{SWAP}\cdot F_{\mathrm{Path}}^{(d_P)}\cdot \mathrm{CZ}\cdot F_{\mathrm{OAM}}^{(d_O)}, \qquad d=d_Od_P, \] with deterministic measurement either in the path outputs or, after an OAM sorter, in the OAM basis. For \(d=2^M\), the resource counts scale linearly with dimension: approximately \(N_{BS}\approx 6.037\,d\) beam splitters, \(N_{\mathrm{Dove}}\approx 4.860\,d\) Dove prisms, \(N_{\mathrm{holo}}\approx 3.037\,d\) holograms, and \(N_{\mathrm{phase}}\approx 2.976\,d\) phase shifters. This improves over \(O(d\log d)\) path-only recursive designs and contrasts with earlier OAM Fourier schemes based on specially designed phase plates [2004.12788]. Together these results show that, in photonics, QFT+M is not merely an algorithmic readout device. It is also a method for certifying bosonic interference, deterministically generating number–path entanglement, and implementing metrological or high-dimensional mode-analysis tasks directly in hardware [1708.00296], [2004.12788]. ## 5. Period extraction and the local period problem In period-finding contexts, QFT+M is the stage that converts hidden periodic structure into a classical integer suitable for continued-fraction recovery. The “Amplified Quantum Fourier Transform” for the local period problem makes this role explicit. One is given an oracle \(f:L\to\{0,1\}\) over \(L=\{0,1,\dots,N-1\}\), where the marked subset \[ A=\{\,j:s+rP,\ r=0,1,\dots,M-1\,\} \] has size \(M\), offset \(s\), and local period \(P<\sqrt N\), with \(M\ll N\) known. After preparing the uniform superposition and applying \(k\approx \lfloor \pi/(4\theta)\rfloor\) Grover iterations, where \(\sin\theta=\sqrt{M/N}\), the state becomes \[ |\psi_k\rangle = \sin((2k+1)\theta)|\psi_A\rangle + \cos((2k+1)\theta)|\psi_{\overline A}\rangle. \] The subsequent QFT+M step produces an outcome \(y\) whose distribution is concentrated near multiples of \(N/P\) [1010.0033]. The mechanism is the geometric-series factor \[ \sum_{r=0}^{M-1}e^{2\pi i rPy/N} = \frac{1-e^{2\pi i MPy/N}}{1-e^{2\pi i Py/N}}, \qquad \left|\sum_{r=0}^{M-1}e^{2\pi i rPy/N}\right| = \left|\frac{\sin(\pi MPy/N)}{\sin(\pi Py/N)}\right|, \] which is large when \(y\approx dN/P\). Exact probability laws are derived for the cases \(y=0\), \(Py\equiv0\pmod N\), \(Py\neq0\pmod N\) with \(MPy\neq0\pmod N\), and \(MPy\equiv0\pmod N\), the last yielding \(\Pr(y)=0\). These formulas explain both the enhanced peaks and the structural notches in the spectrum [1010.0033]. The recovery of \(P\) proceeds by continued fractions. If \[ \left|\frac{y}{N}-\frac{d}{P}\right|\le \frac{1}{2P^2} \] and \(\gcd(d,P)=1\), then \(d/P\) is a convergent of \(y/N\), so \(P\) can be recovered uniquely under the promise \(P^2