May 15, 2026
A standalone extension note connecting distributed iQFT routing costs with the path-volume framework of Path-Length Distributions, Inverse Geometry, and Designed Neighborhood Spectra.
← back to research indexThe distributed inverse Quantum Fourier Transform is usually analyzed by counting which remote controlled-phase interactions remain after approximation and pruning. A routing extension then asks how far those retained interactions travel through a physical QPU interconnect. This note adds a more geometric layer, applying the path-length distribution program from Path-Length Distributions, Inverse Geometry, and Designed Neighborhood Spectra to distributed iQFT routing [2]. The exponential phase law of the iQFT induces a metric on logical QPU blocks; thresholding weak phase interactions is equivalent to taking a ball graph in that induced metric. Once a placement into hardware is chosen, the retained interactions define a phase-weighted routing spectrum: a probability measure on physical route lengths. The ordinary routing cost is the first moment of this spectrum. A Gray-code hypercube placement becomes a spectral compactification result, placing all retained phase-weighted routing mass inside radius $\min\{D,\log_2 P\}$. Replacing shortest-route distance by route-volume distributions then yields an inverse-design problem: choose a physical QPU graph whose path-volume law matches the communication spectrum demanded by the distributed iQFT.
This note joins two related viewpoints. The first is architectural: the inverse Quantum Fourier Transform is a circuit-level object on a single processor, but in a distributed architecture it acquires geometry. If a controlled phase rotation acts between qubits on the same QPU, it is local. If it acts between qubits on different QPUs, it becomes a communication event. Approximate distributed iQFT constructions exploit the fact that phase angles decay exponentially with separation, so weak long-range rotations can be pruned. After pruning, one obtains a retained logical communication graph and must route its edges through a physical interconnect [1, 3].
The second viewpoint comes directly from the companion path-volume paper Path-Length Distributions, Inverse Geometry, and Designed Neighborhood Spectra [2]. In Euclidean space, the collection of one-waypoint paths between two endpoints has a specific path-volume law: equal-length sets are ellipses, and the density of available detours follows from area growth. More generally, a metric-measure space induces path-length spectra. Conversely, one may try to infer or design a geometry whose path-volume or neighborhood spectra have a prescribed shape.
Central idea The distributed iQFT does not merely run on a physical graph. Its phase decay law induces a logical geometry. A hardware placement turns that logical geometry into a phase-weighted routing spectrum, and architecture design becomes an inverse problem over physical path-volume geometry.
In slogan form, \[\text{iQFT phase decay}\quad\Longrightarrow\quad \text{logical phase geometry}\quad\Longrightarrow\quad \text{physical routing spectrum}.\]
Let there be \(P\) QPUs, each holding \(Q\) logical qubits, so \(n=PQ\). Logical node \(p\in\{0,\ldots,P-1\}\) holds the block \[B_p=\{pQ,pQ+1,\ldots,(p+1)Q-1\}.\] The inverse QFT contains controlled phase rotations whose magnitudes decay exponentially with qubit separation. At the node level, after replacing many qubit-level interactions by a block-level abstraction, this motivates a horizon graph.
Model: horizon-\(D\) distributed iQFT graph Let \[G_D=(V,E_D),\qquad V=\{0,\ldots,P-1\},\] where \[(p,q)\in E_D\quad\Longleftrightarrow\quad 0<|p-q|\le D.\] Exact distributed iQFT corresponds to \(D=P-1\). Approximate distributed iQFT corresponds to a finite communication horizon \(D\ll P\).
This graph records which QPU blocks must communicate after pruning. It does not yet say how those communications move through hardware.
The horizon \(D\) is a geometric shadow of a quantum approximation. Weak long-range phase corrections are removed; the surviving phase corrections form a banded graph on the QPU blocks. The circuit has become a weighted interaction geometry.
The horizon graph itself comes from the phase law. Let \(a_{pq}\ge 0\) denote effective retained phase magnitude, communication demand, or phase-weighted interaction strength between logical QPU blocks \(p\) and \(q\). The idealized iQFT scaling is \[a_{pq}=a_0 b^{-|p-q|},\qquad b>1.\] For the usual binary phase hierarchy, one may take \(b=2\), up to convention-dependent constants.
Definition: phase-induced distance For positive interaction weights \(a_{pq}\), define \[d_\phi(p,q)=-\frac{1}{\log b}\log\left(\frac{a_{pq}}{a_0}\right).\] When \(a_{pq}=a_0b^{-|p-q|}\), this gives \(d_\phi(p,q)=|p-q|\).
Proposition 1: exponential phase decay induces a line metric Suppose \(a_{pq}=a_0b^{-|p-q|}\), with \(b>1\). Then \[d_\phi(p,q)=|p-q|.\] Consequently, thresholding weak iQFT interactions is equivalent to taking a ball graph in the induced metric: \[a_{pq}\ge \tau\quad\Longleftrightarrow\quad d_\phi(p,q)\le R_\tau, \qquad R_\tau=-\frac{1}{\log b}\log\left(\frac{\tau}{a_0}\right).\]
Proof. Substituting the exponential form gives \[d_\phi(p,q)=-\frac{1}{\log b}\log(b^{-|p-q|})=|p-q|.\] The threshold equivalence follows by applying the monotone logarithm to \(a_0b^{-|p-q|}\ge\tau\). \(\square\)
A robust version is useful when constants or block aggregation distort the ideal law.
Proposition 2: approximate exponential decay gives a quasi-line geometry Suppose there are constants \(0<c_1\le c_2\) and \(\alpha>0\) such that \[c_1e^{-\alpha |p-q|}\le \frac{a_{pq}}{a_0}\le c_2e^{-\alpha |p-q|}.\] Define \(d_\phi(p,q)=-(1/\alpha)\log(a_{pq}/a_0)\). Then \[|p-q|-\frac{\log c_2}{\alpha}\le d_\phi(p,q) \le |p-q|-\frac{\log c_1}{\alpha}.\] Thus the induced phase geometry differs from the block-line metric by only an additive constant determined by multiplicative distortion in the phase weights.
Proof. Take logarithms of the two-sided bound and multiply by \(-1/\alpha\), reversing the inequalities. \(\square\)
The iQFT already contains a geometry. Large phase corrections are close. Small phase corrections are far. Taking a negative logarithm turns multiplicative phase decay into additive distance. The familiar banded communication pattern is a metric ball graph, not merely an engineering cutoff.
Let the physical QPU interconnect be \[G_{\mathrm{phys}}=(V_{\mathrm{phys}},E_{\mathrm{phys}}),\qquad |V_{\mathrm{phys}}|=P.\] A placement is a bijection \(\sigma:V\to V_{\mathrm{phys}}\). Let \[d_{\mathrm{phys}}(u,v)=\operatorname{dist}_{G_{\mathrm{phys}}}(u,v)\] be physical graph distance, and attach nonnegative weights \(w_{pq}\) to retained logical edges. These weights may represent remote gates, EPR-pair demand, latency demand, or phase-weighted communication mass. The usual routing functional is \[C_D(\sigma)=\sum_{(p,q)\in E_D}w_{pq}\,d_{\mathrm{phys}}(\sigma(p),\sigma(q)).\] This is a scalar summary. The richer object is the distribution whose first moment it computes.
Definition: phase-weighted routing spectrum Let \(W=\sum_{(p,q)\in E_D}w_{pq}\). For a placement \(\sigma\), define \[\mu_\sigma=\frac{1}{W}\sum_{(p,q)\in E_D}w_{pq}\,\delta_{d_{\mathrm{phys}}(\sigma(p),\sigma(q))}.\] This is the phase-weighted distribution of physical route lengths required by the retained distributed iQFT interactions.
Proposition 3: routing cost is the first spectral moment For any placement \(\sigma\), \[\frac{C_D(\sigma)}{W}=\int r\,d\mu_\sigma(r), \qquad C_D(\sigma)=W\,\mathbb{E}_{\mu_\sigma}[r].\]
Proof. Substitute the definition of \(\mu_\sigma\) into the integral and multiply by \(W\). \(\square\)
The scalar routing cost is not discarded. It becomes the mean of a more informative distribution. The spectrum \(\mu_\sigma\) says how much phase-weighted communication occurs at radius one, radius two, radius three, and so on. A good architecture should not only lower the mean; it should control the tail.
The routing spectrum immediately converts geometric support information into cost information.
Proposition 4: support bound If \(\operatorname{supp}(\mu_\sigma)\subseteq[0,R]\), then \[C_D(\sigma)\le WR.\]
Proof. Since \(r\le R\) on the support of \(\mu_\sigma\), \[C_D(\sigma)=W\int r\,d\mu_\sigma(r) \le W\int R\,d\mu_\sigma(r)=WR.\] \(\square\)
Proposition 5: tail-mass routing bound Assume the physical graph has diameter \(\Delta\). If \(\mu_\sigma([R,\Delta])\le\beta\), then \[C_D(\sigma)\le W\bigl((1-\beta)R+\beta\Delta\bigr) = W\bigl(R+\beta(\Delta-R)\bigr).\]
Proof. At least \(1-\beta\) of the mass lies at radius at most \(R\), and at most \(\beta\) lies beyond \(R\). Since no radius exceeds \(\Delta\), the stated expectation bound follows. \(\square\)
The long-distance tail is not merely a visual feature of a distribution. It directly bounds how much weighted communication cost can leak into expensive routes.
Assume \(P=2^m\). Let the physical graph be the \(m\)-dimensional hypercube \[H_m=\{0,1\}^m,\] where two vertices are adjacent if their binary labels differ in exactly one coordinate. The graph distance is Hamming distance, so \[\operatorname{diam}(H_m)=m=\log_2 P.\] Place logical blocks by Gray code, \[\sigma(p)=\operatorname{Gray}(p)=p\oplus(p\gg 1),\] so consecutive Gray-code words differ in exactly one bit.
Proposition 6: Gray-code hypercube embedding of the phase geometry Let \(P=2^m\), let \(G_{\mathrm{phys}}=H_m\), and place logical nodes by \(\sigma(p)=\operatorname{Gray}(p)\). For every pair \(p,q\), \[d_{H_m}(\sigma(p),\sigma(q))\le \min\{|p-q|,\log_2 P\}.\] If \(d_\phi(p,q)=|p-q|\), then \[d_{H_m}(\sigma(p),\sigma(q))\le \min\{d_\phi(p,q),\log_2 P\}.\] Consequently, for every retained horizon-\(D\) edge, \[d_{H_m}(\sigma(p),\sigma(q))\le \min\{D,\log_2 P\}.\]
Proof. Every pair of vertices in \(H_m\) is separated by at most \(m=\log_2P\), giving the diameter bound. Gray-code ordering gives \(d_{H_m}(\sigma(r),\sigma(r+1))=1\). If \(q>p\), then \(\sigma(p),\sigma(p+1),\ldots,\sigma(q)\) is a path of length \(q-p\), so \(d_{H_m}(\sigma(p),\sigma(q))\le |p-q|\). Combining the two bounds proves the claim; retained edges satisfy \(|p-q|\le D\). \(\square\)
Corollary 7: hypercube support bound for the routing spectrum Under the assumptions of Proposition 6, \[\operatorname{supp}(\mu_\sigma)\subseteq[0,\min\{D,\log_2P\}],\] and therefore \[C_D(\sigma)\le \min\{D,\log_2P\} \sum_{(p,q)\in E_D}w_{pq}.\]
Proof. Proposition 6 places every retained edge at radius no larger than \(\min\{D,\log_2P\}\). Proposition 4 then gives the cost bound. \(\square\)
The hypercube is not merely a graph with small diameter. In this setting it compactifies the iQFT phase geometry. Consecutive logical blocks remain adjacent through Gray code, while the worst possible retained communication is forced into a logarithmic-radius ball.
The previous spectrum places one atom at the shortest physical distance for each retained logical interaction. The link to Path-Length Distributions, Inverse Geometry, and Designed Neighborhood Spectra is that shortest distance is only the first geometric observable: the richer object counts not only the shortest route, but the available routes around it.
For each retained logical edge \(e=(p,q)\in E_D\), let \(\Gamma_e=\Gamma_{\sigma(p),\sigma(q)}\) be a family of admissible physical routes from \(\sigma(p)\) to \(\sigma(q)\). Let \(c_a\ge0\) be an edge cost, and define route length \[L_c(\gamma)=\sum_{a\in\gamma}c_a.\] Let \(\nu_e\) be a counting measure or weighted route measure on \(\Gamma_e\). Define \[F_e(\ell)=\nu_e\{\gamma\in\Gamma_e:L_c(\gamma)\le\ell\}, \qquad \rho_e=dF_e.\]
Definition: phase-weighted iQFT path-volume spectrum The global phase-weighted route-volume spectrum is \[\rho_{\mathrm{iQFT}}=\frac{1}{W}\sum_{e=(p,q)\in E_D}w_e\rho_e,\] or equivalently \[F_{\mathrm{iQFT}}(\ell)=\frac{1}{W}\sum_{e=(p,q)\in E_D}w_eF_e(\ell).\]
The shortest-route spectrum \(\mu_\sigma\) is recovered by replacing each route family with a single atom at shortest distance: \[\rho_e=\delta_{d_{\mathrm{phys}}(\sigma(p),\sigma(q))}.\] Thus the earlier theory is the degenerate shortest-path case of the path-volume theory.
For a fixed retained iQFT interaction, hardware may offer many possible routes. Some are shortest, some are slightly longer, and some are long detours. The path-volume spectrum asks how much phase-weighted routing capacity exists at each route length.
The previous definitions suggest a genuine inverse problem, in the same spirit as reconstructing or designing geometry from prescribed path-length spectra. Instead of choosing a familiar physical graph and then analyzing its cost, choose a physical graph whose route-volume law matches the structure demanded by the distributed iQFT.
Let \(\rho^\star\) be a desired route-volume spectrum: compact support for hard latency control, an exponential law matching iQFT phase decay, a shifted lognormal law for multiplicative route distortions, or an anchor-wise target spectrum around each QPU block. Then an inverse architecture design problem is \[(G_{\mathrm{phys}}^\star,\sigma^\star,c^\star) = \arg\min_{G,\sigma,c} D(\rho_{\mathrm{iQFT}},\rho^\star) +\lambda\Omega(G,\sigma,c)+\eta\mathcal L_{\mathrm{hardware}}(G,c).\] Here \(D\) may be Wasserstein distance, binned squared error, KL divergence when densities are available, or another discrepancy. The term \(\Omega\) prevents degeneracy, while \(\mathcal L_{\mathrm{hardware}}\) may encode degree constraints, congestion, fabrication cost, latency, fidelity, or platform limits.
An anchor-wise version mirrors designed neighborhood spectra. For each logical block \(p\), define \[W_p=\sum_{q:(p,q)\in E_D}w_{pq},\qquad \mu_{\sigma,p}=\frac{1}{W_p}\sum_{q:(p,q)\in E_D}w_{pq}\, \delta_{d_{\mathrm{phys}}(\sigma(p),\sigma(q))}.\] Then optimize \[\sigma^\star=\arg\min_\sigma\sum_pD(\mu_{\sigma,p},\mu_p^\star) +\lambda C_D(\sigma)+\Omega(\sigma).\]
Proposition 8: spectrum matching controls mean routing cost Let \(\mu_\sigma\) be the shortest-route spectrum and let \(\mu^\star\) be a target probability measure on route lengths. If \[W_1(\mu_\sigma,\mu^\star)\le\varepsilon,\] where \(W_1\) is Wasserstein-1 distance, then \[\left|\frac{C_D(\sigma)}{W}-\int r\,d\mu^\star(r)\right|\le\varepsilon.\] In particular, \[C_D(\sigma)\le W\left(\int r\,d\mu^\star(r)+\varepsilon\right).\]
Proof. The function \(f(r)=r\) is 1-Lipschitz on the nonnegative real line. By the Kantorovich-Rubinstein dual representation of \(W_1\), \[\left|\int r\,d\mu_\sigma(r)-\int r\,d\mu^\star(r)\right| \le W_1(\mu_\sigma,\mu^\star)\le\varepsilon.\] Using Proposition 3 gives the stated cost bound. \(\square\)
This is the cleanest bridge from inverse geometry back to engineering cost. If the learned or designed architecture matches a target routing spectrum in \(W_1\), then its normalized communication cost is close to the target mean. Matching the spectrum is not just aesthetic; it controls the routing objective.
The inverse-geometry viewpoint also suggests a sparsification principle: remove edges according to a controlled spectrum, then require replacement paths to bound shortest-path distortion. The same logic applies here.
Let \(G_{\mathrm{phys}}\) be a full physical routing fabric and let \(H\subseteq G_{\mathrm{phys}}\) be a sparse physical routing fabric. Let \(d_H\) denote shortest-path distance inside \(H\).
Proposition 9: phase-weighted stretch bound Suppose that for every retained distributed-iQFT edge \((p,q)\in E_D\), \[d_H(\sigma(p),\sigma(q))\le \tau d_{G_{\mathrm{phys}}}(\sigma(p),\sigma(q))\] for some \(\tau\ge1\). Then \[C_D^H(\sigma)\le \tau C_D^{G_{\mathrm{phys}}}(\sigma).\]
Proof. Multiply the pointwise stretch inequality by \(w_{pq}\) and sum over \((p,q)\in E_D\). \(\square\)
A tail-controlled version is also useful. Suppose a sparse fabric preserves all retained interactions up to radius \(R\), and the remaining long-distance phase mass is at most \(\beta\). If the sparse fabric has worst-case replacement stretch \(\tau\) on the long tail, then the total penalty is concentrated on a controlled portion of the phase mass. Distributional tail control says how much important structure is exposed to pruning; stretch control says how much error that pruning can introduce.
The routing note already gives a practical theorem: hypercube geometry with Gray-code placement bounds retained iQFT communication by \(\min\{D,\log_2P\}\). The companion inverse-geometry note, Path-Length Distributions, Inverse Geometry, and Designed Neighborhood Spectra, gives a broader principle: geometries can be represented, inferred, or designed through path-length and neighborhood spectra. The present extension adds four things.
An intrinsic iQFT geometry. Exponential phase decay induces a phase metric \(d_\phi\), and the horizon graph is a ball graph in that metric.
A spectral view of routing. The scalar cost \(C_D\) is the first moment of a phase-weighted routing spectrum \(\mu_\sigma\).
A compactification interpretation. The Gray-code hypercube result says the retained phase-weighted spectrum has bounded support inside \([0,\min\{D,\log_2P\}]\).
An inverse-design problem. Physical QPU geometry can be optimized so that its route-volume spectrum matches the phase-weighted communication spectrum demanded by the iQFT.
The model abstracts away effects that a full architecture paper would need to include: edge congestion, routing conflicts, entanglement generation latency, purification overhead, correlated noise, and platform-specific degree constraints. A natural congestion functional is \[C_D^{\mathrm{cong}}(\sigma)= \sum_{a\in E_{\mathrm{phys}}}\Psi\left( \sum_{(p,q)\in E_D}w_{pq}\mathbf 1\{a\in\gamma_{pq}\} \right),\] where \(\gamma_{pq}\) is the chosen route and \(\Psi\) penalizes edge load. The route-volume spectrum could then be paired with a congestion spectrum.
The best next experiment is probably computational: fix \(P,D\), and \(w_{pq}\); compute \(\mu_\sigma\) for several candidate physical graphs, such as a line, ring, grid, hypercube, expander, or sparse learned fabric; then compare their spectra. The picture should make the architecture choice visible: where is the phase mass, how heavy is the tail, and how much of the iQFT is forced into expensive routes?
The distributed iQFT induces a phase-weighted logical geometry; a hardware placement converts that geometry into a routing spectrum; and architecture design becomes the inverse problem of choosing a physical QPU path-volume geometry whose short routes carry the important phase mass.