QPass: Multifaceted Applications in Quantum & Analytics

Updated 5 July 2026

QPass is a multi-context research term denoting four distinct constructs across quantum cryptography, quantum generative modeling, computer graphics, and soccer analytics.
Each variant employs a tailored methodology—from SWAP test-based authentication and QUBO energy models to branch-free tile traversal and field-value evaluations.
In sports analytics, QPass quantifies the strategic impact of soccer passes by measuring changes in field possession value.

QPass is a reused research label rather than a single established concept. In the arXiv literature, it denotes at least four technically unrelated constructs: a quantum password-checking protocol based on small quantum states and SWAP tests; a quantum adiabatic password-generation framework based on QUBO and UD-MIS formulations; a branch-free single-pass tile traversal algorithm inside the QuadBox rasterization pipeline for 3D Gaussian Splatting; and a merit-based metric for valuing soccer passes by change in field value (Garcia-Escartin et al., 2015, Mücke et al., 10 Jun 2025, Li et al., 6 May 2026, Gyarmati et al., 2016).

1. Disciplinary scope and terminological distribution

The term has appeared in quantum cryptography, quantum generative modeling, computer graphics, and sports analytics. These uses are not extensions of one another; they share a name but not a common formalism.

Usage of QPass	Domain	Core object
Simple quantum password checking	Quantum cryptography	Authentication protocol
Quantum adiabatic generation of human-like passwords	Quantum generative modeling	Password generator
QuadBox QPass	3D Gaussian Splatting	Tile traversal algorithm
Merit-based evaluation of soccer passes	Sports analytics	Pass valuation metric

This distribution matters because the surrounding technical vocabulary changes completely across papers. In one setting, QPass is defined by a Hilbert-space encoding with dimension $D=2^d$ and repeated SWAP tests; in another, it is defined by a learned QUBO energy $E_Q(z)=z^\top Q z$ ; in another, it is a discrete tile-space traversal scheme; and in another, it is the pass value

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$

(Garcia-Escartin et al., 2015, Mücke et al., 10 Jun 2025, Li et al., 6 May 2026, Gyarmati et al., 2016)

2. Quantum password checking

In "Simple quantum password checking" (Garcia-Escartin et al., 2015), QPass is a quantum authentication protocol in which Alice and Bob share an $m$ -bit classical password $p$ , but Alice never sends $p$ directly. Instead, each authentication round uses a fresh random string $r_i$ , computes

$j = H(p \| r_i),$

and encodes that value as a symmetric quantum state in a Hilbert space of deliberately restricted dimension

$D = 2^d,\qquad d<n.$

The essential design choice is that the state space is smaller than the number of hash bits, so the encoded object can be compared but not fully read out (Garcia-Escartin et al., 2015).

The privacy mechanism is information-theoretic rather than computational. Averaging uniformly over the family of encoded states yields the maximally mixed state,

$\frac{1}{N}\sum_{j=0}^{N-1} |\psi_j\rangle\langle\psi_j| = \frac{I}{D},$

so an adversary who does not know the password sees only the completely mixed state. The paper explicitly motivates this with quantum information limits, citing the intuition that the protocol encodes “too many bits into too small a space,” and relates this to the Holevo bound and stronger results such as Nayak’s and Ben-Aroya–Regev–de Wolf’s bounds (Garcia-Escartin et al., 2015).

Verification is performed by state comparison rather than decoding. Bob locally prepares the expected state $E_Q(z)=z^\top Q z$ 0 and applies the SWAP test to compare it with the received state. If the received state is $E_Q(z)=z^\top Q z$ 1 and Bob’s local state is $E_Q(z)=z^\top Q z$ 2, the probability of detecting a difference is

$E_Q(z)=z^\top Q z$ 3

and the pass probability is

$E_Q(z)=z^\top Q z$ 4

The protocol repeats this comparison $E_Q(z)=z^\top Q z$ 5 times with fresh $E_Q(z)=z^\top Q z$ 6; Bob accepts only if all $E_Q(z)=z^\top Q z$ 7 tests pass (Garcia-Escartin et al., 2015).

The security analysis emphasizes replay resistance. Under a random-state attack, the average fidelity of any fixed trial state $E_Q(z)=z^\top Q z$ 8 against legitimate states is

$E_Q(z)=z^\top Q z$ 9

which yields approximate adversarial success probability

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 0

over $\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 1 SWAP tests. The replay analysis further considers physically allowed transformations of captured states and extends to the case where up to $\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 2 valid states are captured, with the stated design condition

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 3

The paper also argues that the scheme is robust to dictionary attacks when the effective password space satisfies

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 4

while noting that very small structured dictionaries fall outside the guarantee; in that case, it suggests preprocessing the password to a shorter randomized string $\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 5 (Garcia-Escartin et al., 2015).

Practically, the paper stresses that the protocol does not require a full quantum computer. The listed primitives are quantum state preparation, random number generation, and state comparison via the SWAP test, with an optical implementation via the Hong-Ou-Mandel effect also mentioned. The protocol is therefore presented as an authentication scheme whose security derives from quantum information constraints rather than hardness assumptions (Garcia-Escartin et al., 2015).

3. Quantum adiabatic generation of human-like passwords

In "Quantum Adiabatic Generation of Human-Like Passwords" (Mücke et al., 10 Jun 2025), QPass denotes a quantum generative framework rather than an authentication protocol. The objective is to learn a probabilistic model from real password data and sample new passwords from a quantum system whose low-energy or Gibbs-state structure approximates the learned distribution. The paper instantiates this in two stages: a QUBO-based model and a UD-MIS-based hardware mapping for neutral-atom adiabatic hardware such as the QuEra Aquila 256-qubit device (Mücke et al., 10 Jun 2025).

The pipeline begins with passwords as bounded-length character strings

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 6

drawn from a multiset $\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 7. A tokenization function

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 8

maps each password to a unique token sequence, and the tokenized distribution is defined by

$\textrm{QPass}(p) = \begin{cases} f_e - f_s & \text{if successful} \ l_e - f_s & \text{otherwise.} \end{cases}$ 9

Tokens are then encoded into binary vectors through

$m$ 0

with padding to a fixed maximum of $m$ 1 tokens so that each password becomes a binary vector $m$ 2 (Mücke et al., 10 Jun 2025).

The learned model is a QUBO with energy

$m$ 3

which induces the Boltzmann distribution

$m$ 4

Training minimizes the KL divergence between the empirical dataset distribution and the model distribution. The gradient is expressed as a difference of second-order moments,

$m$ 5

and the paper updates $m$ 6 with ADAM. In the reported setup, the authors use a modified Byte Pair Encoding vocabulary of 256 tokens, restrict to passwords of at most 6 tokens, employ 5-fold cross-validation, sample $m$ 7 times from the QUBO’s Boltzmann distribution by Gibbs sampling with burn-in 100 and keeping every 10th sample, and repeat training for 1000 iterations (Mücke et al., 10 Jun 2025).

The second stage maps the learned QUBO to a Unit-Disk Maximum Independent Set instance so that it can run on neutral-atom hardware governed by Rydberg blockade constraints. The paper proposes a force-directed atom placement method inspired by Fruchterman–Reingold, interprets QUBO weights as edge likelihoods, and imposes hardware-specific geometric constraints including x-separation bounded by $m$ 8, y-separation bounded by $m$ 9, an Euclidean distance threshold around $p$ 0, and a y-coordinate difference condition of either $p$ 1 or $p$ 2 (Mücke et al., 10 Jun 2025).

Evaluation uses the RockYou dataset, simplified to printable ASCII characters. The principal realism proxy is minimum edit distance to the evaluation set, computed with a BK-tree; the paper also measures exact overlap with held-out passwords. The reported result is that the learned models achieve mean minimum edit distances more than one standard deviation closer to human-generated passwords than uniformly random token sequences, and that relatively small samples of 128 passwords generated on the QuEra Aquila 256-qubit neutral atom quantum computer contain examples such as "Tunas200992" and "teedem28iglove" (Mücke et al., 10 Jun 2025).

This usage of QPass is therefore an energy-based generative model implemented through adiabatic quantum optimization. It is explicitly framed as feasible because passwords are short, bounded-length, discrete sequences, unlike full-scale natural-language generation. The paper also states that the work is a proof of concept, with limitations arising from qubit count, noise, tokenization dependence, approximate QUBO training, and the nontrivial QUBO-to-UD-MIS embedding (Mücke et al., 10 Jun 2025).

4. QPass in 3D Gaussian Splatting rasterization

In "QuadBox: Accelerating 3D Gaussian Splatting with Geometry-Aware Boxes" (Li et al., 6 May 2026), QPass is not a security or password method but the traversal stage of a rasterization pipeline. It is introduced as the single-pass tile traversal algorithm that makes QuadBox practical for Gaussian-to-tile intersection testing in 3D Gaussian Splatting. The problem addressed is the preprocess-stage overhead caused by conservative relaxed bounding boxes in the standard rasterizer, which generate too many false-positive tiles and therefore increase memory traffic, sorting work, and render-kernel work (Li et al., 6 May 2026).

QuadBox constructs a tighter discrete approximation to a projected Gaussian by using four axis-aligned boxes around the ellipse-like support region. The construction proceeds through initialization and opacity filtering, DualBox extraction, and adaptive stretching to form QuadBox. The paper derives a geometry-aware stretching factor $p$ 3 from the covariance-inverse parameters $p$ 4, emphasizes that $p$ 5 depends only on $p$ 6, is independent of $p$ 7 and translation, satisfies $p$ 8, and equals 1 for circles or axis-aligned principal axes. The final representation is tile-aligned on an image partition of $p$ 9 tiles, with each sub-box mapped to integer tile-space bounds $p$ 0 (Li et al., 6 May 2026).

QPass then traverses this discrete representation in a single pass. Rather than scanning each of the four sub-boxes independently, it first aggregates global tile bounds, then chooses the shorter axis of the global range as the scan direction, and for each scanline performs fast interval overlap tests to determine which sub-boxes are active. The paper’s example uses column-wise traversal. The computational reduction comes from replacing analytic ellipse–tile tests with simple discrete comparisons such as

$p$ 1

and interval aggregation by min/max over the orthogonal coordinate. The paper explicitly characterizes QPass as a branch-free single-pass tile traversal algorithm, and states that each tile is visited exactly once (Li et al., 6 May 2026).

The performance claims are specific. The paper reports an average rendering speedup of about $p$ 2; on Mip-NeRF 360, 3DGS FPS improves from 180 to 322; and in the ablation the full QuadBox with QPass reaches 460 FPS on RTX 4090, compared with 425 FPS for AdR-AABB. The ablation also isolates QPass by comparing "3DGS + QuadBox $p$ 3" at 413 FPS with "3DGS + QuadBox" at 460 FPS, where QuadBox $p$ 4 denotes QuadBox without QPass (Li et al., 6 May 2026).

Within this literature, QPass is thus an implementation-level traversal engine. Its significance lies not in a new geometric primitive by itself, but in converting a tighter four-box discrete envelope into deterministic, cache-friendly, interval-based tile enumeration without duplicate visits (Li et al., 6 May 2026).

5. Merit-based evaluation of soccer passes

In "QPass: a Merit-based Evaluation of Soccer Passes" (Gyarmati et al., 2016), QPass is a football analytics framework for quantifying the intrinsic value of a pass by the change it induces in field possession value. The stated motivation is that standard descriptive statistics—pass counts, completion rate, assists, key passes, forward passes—are context-free and do not measure how a pass contributes to a team’s attacking or possession strategy (Gyarmati et al., 2016).

The method has three stated steps: create a partitioning of the field that reflects the playing style of a team, derive the value of having or not having the ball at a field region, and evaluate each pass based on the change in field values caused by the pass. The field partition is team-specific rather than uniform and is obtained with mini-batch K-means in an iterative procedure in which the number of clusters decreases from $p$ 5 to $p$ 6 in steps of $p$ 7. The paper explains that this yields finer resolution in tactically dense regions, such as advanced areas for FC Barcelona and the team’s own half for Levante (Gyarmati et al., 2016).

Field values are obtained from a system of linear equations over a transition model that includes team possession states, opponent possession states, and two shot states. The empirical setting uses goal value $p$ 8, conceded goal value $p$ 9, shot value $r_i$ 0, and opponent shot value $r_i$ 1, with

$r_i$ 2

Transitions are counted from pass starts to pass ends, from unsuccessful or terminal possessions to the opponent’s possession state at the end location, and from shots to goal or non-goal shot states. After normalization, solving the linear system yields field values that represent how beneficial it is for the team or the opponent to control the ball in each region (Gyarmati et al., 2016).

The pass metric itself is

$r_i$ 3

where $r_i$ 4 is the field value of the starting location, $r_i$ 5 is the value of the ending location under retained possession, and $r_i$ 6 is the value of the ending location when possession is lost. A positive value therefore means that the pass improves the team’s strategic state, while a negative value means the opposite (Gyarmati et al., 2016).

The paper applies the method to an event-based dataset from the 2015/16 Spanish La Liga season comprising more than 330,000 passes and nearly 8,600 shots. For player ranking, only players with at least 100 passes are considered, and median QPass rather than mean QPass is used because cluster-based values can create outliers. The reported qualitative findings include role-specific differences, with goalkeepers often exhibiting high QPass because they initiate possession and build-up safely, and attackers or strikers often exhibiting lower QPass because their passes occur in already valuable zones. The paper highlights Lucas Vazquez, Antoine Griezmann, and Lionel Messi among attackers with high QPass, contrasts Messi’s passes from outside the box into the box with Luis Suarez’s short lateral passes inside the box, and argues that nearly 50% of failed passes by defenders can still help their team. Its broader conclusion is the counterintuitive one stated in the abstract: losing the ball possession could lead to better chances to win a game (Gyarmati et al., 2016).

This usage of QPass is therefore a state-based marginal contribution metric. It values passes by tactical state transition rather than by surface success, which is why unsuccessful passes can receive positive scores when the turnover occurs in strategically favorable zones (Gyarmati et al., 2016).

The diversity of meanings attached to QPass is particularly visible in the password-related literature. In one paper, QPass is a password-checking protocol in which a shared secret is never transmitted and security comes from small-dimensional quantum encodings and state comparison (Garcia-Escartin et al., 2015). In another, QPass is a password generator that learns a password distribution through tokenization and a QUBO energy model, then samples from an adiabatically prepared quantum system (Mücke et al., 10 Jun 2025). These are adjacent only at the application level of “passwords”; their threat models, state spaces, and algorithmic primitives are different.

A neighboring acronymic overlap appears in "Pseudo Quantum Random Number Generator with Quantum Permutation Pad" (Kuang et al., 2023), where the central primitive is a Quantum Permutation Pad (QPP), a pad of quantum permutation matrices used as a high-entropy bijective transformation layer for a pseudo quantum random number generator, entropy booster, and whitening algorithm. That paper states that the quantum permutation space for an $r_i$ 7-qubit system has $r_i$ 8 elements and Shannon-equivalent entropy

$r_i$ 9

and reports a 64-matrix 8-bit pad with quoted entropy figures above 100 Kbits, while also noting numerical inconsistencies in its own text. Although QPP is not the same term as QPass, its inclusion in the surrounding literature illustrates how password- and quantum-themed acronyms can cluster without denoting the same construction (Kuang et al., 2023).

Across the four principal QPass usages, a common surface pattern is that each method compresses a complex object into a lower-dimensional operational representation: password knowledge into a small quantum state, a learned password distribution into a QUBO Hamiltonian, a projected Gaussian into four tile-aligned sub-boxes, and the tactical value of a pass into a field-value differential. This suggests a shared design instinct toward evaluable surrogates, but not a shared technical lineage. As a term of art, QPass is therefore best treated as context-dependent and domain-specific rather than as the name of a unified framework (Garcia-Escartin et al., 2015, Mücke et al., 10 Jun 2025, Li et al., 6 May 2026, Gyarmati et al., 2016).