Qimax: Efficient quantum simulation via GPU-accelerated extended stabilizer formalism
Vu Tuan Hai 1,2 , Bui Cao Doanh 1,2 , Le Vu Trung Duong 3 , Pham Hoai Luan 3 and Yasuhiko Nakashima 3 · 2025
Read Qimax: Efficient quantum simulation via GPU-accelerated extended stabilizer formalism by Vu Tuan Hai 1,2 , Bui Cao Doanh 1,2 , Le Vu Trung Duong 3 , Pham Hoai Luan 3 and Yasuhiko Nakashima 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-03307.txt.
Introduction
Vu Tuan Hai and co-authors present Qimax in a 2025 paper on faster quantum circuit simulation. Simulating Clifford and near-Clifford circuits with the extended stabilizer formalism needs far fewer resources than tracking full state vectors, which is why the method is popular in quantum error correction work.
The catch is speed: with high-rank stabilizers, the usual formalism runs step by step and wastes modern hardware. Qimax parallelizes it so the work spreads across many cores, including graphics processors. In tests, the Python implementation beat well-known simulators such as Qiskit and Pennylane in certain scenarios. The paper targets researchers who need bigger near-Clifford simulations without new hardware.
Catalog record
| CREATOR | Vu Tuan Hai 1,2 , Bui Cao Doanh 1,2 , Le Vu Trung Duong 3 , Pham Hoai Luan 3 and Yasuhiko Nakashima 3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-03307
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Qimax: Efficient quantum simulation via GPU-accelerated extended stabilizer formalism Vu Tuan Hai 1,2 , Bui Cao Doanh 1,2 , Le Vu Trung Duong 3 , Pham Hoai Luan 3 and Yasuhiko Nakashima 3 Abstract Simulating near-Clifford quantum circuits using the extended stabilizer formalism remains computationally challenging due to rapid stabilizer rank growth and limited parallel scalability. In this work, we present Qimax, a reformulation of the extended stabilizer framework designed to shift the practical boundary of classical quantum circuit simulation. Rather than applying gates sequentially to stabilizer generators, Qimax introduces an operator-level decomposition that restructures circuit execution into grouped single- and two-qubit operators. We further propose a GPU-native tensor encoding of stabilizer generators that enables parallel weight propagation and index-based Pauli accumulation. To mitigate the combinatorial explosion inherent in non-Clifford operations, we develop a sparse representation that preserves computational parallelism while substantially reducing memory overhead. This design transforms extended stabilizer simulation from a sequential procedure into a parallel model suitable for a GPU. Experimental results demonstrate that Qimax significantly extends the feasible simulation regime for deep near-Clifford circuits, achieving substantial speedups over other simulators while maintaining accuracy. I Introduction Simulating a quantum system classically demands significant resources, scaling exponentially with the number of particles, posing a challenging computational problem. For instance, Google’s quantum computer, Sycamore, demonstrated quantum supremacy by producing results in days that would take classical computers thousands of years to simulate [ 1 , 2 , 3 ] . Various simulation methods have been proposed, including state-vector [ 4 , 5 ] , tensor-network [ 6 ] , decision diagram [ 7 ] , and stabilizer formalism [ 8 ] . Each method offers different levels of fidelity and resource requirements depending on the quantum circuit’s size and type. Quantum simulator development aims to address domain-specific and general-purpose applications, ultimately targeting the boundary of quantum advantage, where classical computers cannot simulate quantum systems within an acceptable execution time [ 9 , 10 , 11 ] . Simulator scalability is assessed by the number of qubits ( n , or #Qubits) and gates ( m , or #Gates). In the state-vector approach, the quantum state |\psi(\bm{\theta})\rangle is represented as a 2^{n} -dimensional complex tensor, leading to an exponential increase in execution time and memory requirements with #Qubits, while #Gates affects execution time linearly. This exponential scaling is inherent to quantum systems in the state-vector approach. Recent stabilizer formalism research focuses on reducing gate application times for Clifford circuits. Stabilizer formalism based on Heisenberg’s picture [ 12 , 13 ] provides a compact representation for quantum systems and gate applications where the quantum system can be described as stabilizers. Quantum gates act on the stabilizers by mapping between Pauli strings rather than matrix-vector multiplication. Compared to the state-vector approach, the stabilizer formalism can solve the same problems with fewer computational resources, as it operates on stabilizers rather than full state vectors. Existing stabilizer formalism packages have been developed; although they can run a hundred-qubit Clifford circuit with a million gates, the stabilizer formalism limits the application of the quantum simulator, especially for algorithms that require non-Clifford gates. On the other hand, using non-Clifford gates in (extended) stabilizer formalism may lead to a fast increase in complexity. For instance, [ 14 ] shows the exponential rise of stabilizer rank in cases of Quantum Neural Network (QNN), which require non-Clifford gates. Other studies address this by employing approximation methods, balancing complexity and accuracy [ 15 ] . There are two use cases of stabilizer formalism: (1) large-qubit Clifford circuits, and (2) low-rank Clifford and non-Clifford. Because Stim [ 8 ] effectively handles the first, our research focuses on the second. Clifford circuits are used in many problems, such as: quantum error correction [ 16 , 17 , 18 ] , randomized benchmarking [ 19 , 20 ] , shadow tomography [ 21 , 22 , 23 ] and magic state distillation [ 24 ] . Near-Clifford circuits have wider applications. Our proposed package, Qimax, addresses this limitation by offering three modes: v1 (standard extended stabilizer formalism, CPU-based), v2 (parallel stabilizer formalism, GPU-based [ 25 ] , using one-hot vector Pauli encoding), and v3 (parallel stabilizer formalism, GPU-based, using tensor index encoding). II Background II-A Extended stabilizer formalism An n -qubit quantum state |\psi\rangle is defined as a stabilizer state with eigenvalue +1 for every operator in its stabilizer group \{\mathcal{P}_{n,0},\mathcal{P}_{n,1},\dots\} , meaning U|\psi\rangle=+1|\psi\rangle . The stabilizer states form a strict subset of all quantum states that can be uniquely described by \{\mathcal{P}_{n}\} . The elements of the stabilizer group are called stabilizers. Any stabilizer group can be specified by a set of generators G_{|\psi\rangle}\equiv G=\{\mathbb{P}_{n}\} so that every element \mathcal{P}_{n,j} in the stabilizer group can be obtained through matrix multiplication between any pair \{\mathbb{P}_{n,i},\mathbb{P}_{n,j}\} . The number of generators (order) of G is always n , each generator \mathbb{P}_{n,j} is also an n -qubit Pauli string. These generators evolve under the application of m quantum gates, as shown in ( II-A ): \displaystyle G^{(0)}=\left\langle\begin{array}[]{c}\mathbb{P}_{n,0}^{(0)}\\ \vdots\\ \mathbb{P}_{n,n-1}^{(0)}\\ \end{array}\right\rangle\xrightarrow{g^{m}}G^{(m)}=\left\langle\begin{array}[]{c}\mathbb{P}_{n,0}^{(m)}\\ \vdots\\ \mathbb{P}_{n,n-1}^{(m)}\\ \end{array}\right\rangle, where \mathbb{P}_{n,j}^{(0)} can be set as \mathbb{Z}_{n,j}=I^{\otimes j}\otimes Z\otimes I^{\otimes(n-j-1)} and G^{(0)} is the set of generator for initial state |0\rangle^{\otimes n} . Gate g^{(t)} update \mathbb{P}_{n,j}^{(t)} to \mathbb{P}_{n,j}^{(t+1)} . To achieve universal quantum computation, non-Clifford gates, such as rotation gates, must be included. These gates transform stabilizer states into non-stabilizer states. A generator may transform from a single Pauli string into a linear combination of Pauli strings \mathbb{P}_{n}=\sum_{j}\lambda_{j}P_{n,j} with \lambda P_{n}=\lambda p_{0}\otimes\ldots\otimes p_{n-1},\;\equiv\lambda p_{0}\ldots p_{n-1} where \lambda\in\mathbb{R} is the weight. We can relate G^{(m)} directly to the target density operator either as a product of stabilizers or as a sum of Pauli operators, as shown in ( 7 ), respectively: \displaystyle\rho^{(m)}=\frac{1}{2^{n}}\prod_{j=1}^{n}\left(I^{\otimes n}+\mathbb{P}^{(m)}_{n,j}\right)=\frac{1}{2^{n}}\sum_{\mathcal{P}_{n}\in\mathcal{P}_{|\psi\rangle}}\lambda_{\mathcal{P}_{n}}\mathcal{P}_{n}. (7) In the other cases, we can simply track G to get measurement value as p_{j,k}=\text{tr}([\frac{1}{2}(I^{\otimes n}+Z_{k})]|\psi\rangle\langle\psi|) . II-B Single-qubit Pauli measurement Let \mathbb{P}_{n,k,b}=\underbrace{I\otimes\cdots\otimes|b\rangle\langle b|_{k}\otimes\cdots\otimes I}_{n\;\text{terms}}=\frac{1}{2}((-1)^{b}Z_{n,k}+I^{\otimes n}) for k\in[n] . When measuring the k^{\text{th}} qubit of an n -qubit state |\psi\rangle using projectors \{\mathbb{P}_{n,k,0},\mathbb{P}_{n,k,1}\} , we get two possible outcomes: 0 with probability p_{k,0} and 1 with probability p_{k,1} . It follows that p_{k,0}=\mathrm{Tr}(\mathbb{P}_{n,k,0}|\psi\rangle\langle\psi|) where \mathrm{Tr} is the trace mapping. As shown in ( 7 ), the density operator |\psi\rangle\langle\psi| can be written using generators i.e., |\psi\rangle\langle\psi|=\frac{1}{2^{n}}\sum_{\mathbb{P}\in G}\mathbb{P} . The probability p_{k,0} can be obtained as: \displaystyle p_{k,0} \displaystyle=\text{Tr}(\mathbb{P}_{n,k,0}|\psi\rangle\langle\psi|) \displaystyle=\text{Tr}\left(\frac{1}{2}(I^{\otimes n}+Z_{n,k})|\psi\rangle\langle\psi|\right) \displaystyle=\frac{1}{2}\left(\text{Tr}(|\psi\rangle\langle\psi|)+\text{Tr}(Z_{n,k}|\psi\rangle\langle\psi|)\right) \displaystyle=\frac{1}{2}+\frac{1}{2^{n+1}}\sum_{\lambda_{\mathcal{P}_{n}}\mathcal{P}_{n}}\lambda_{\mathcal{P}_{n}}\text{Tr}(Z_{n,k}\mathcal{P}_{n}) For any Pauli string P_{n} , we can express \mathrm{Tr}(P_{n}) as \mathrm{Tr}(P_{0})\cdots\mathrm{Tr}(P_{n-1}) . Since \mathrm{Tr}(X)=\mathrm{Tr}(Y)=\mathrm{Tr}(Z)=0 , \mathrm{Tr}(P_{n})\neq 0 if and only if P_{n}=I^{\otimes n} . Furthermore, \mathbf{P}=Z_{k} if and only if Z_{k}P_{n}=I^{\otimes n} . Thus p_{k} can be simplified as \frac{1}{2}+\frac{1}{2}\sum_{\mathcal{P}_{n}=Z_{k}}\lambda_{\mathcal{P}_{n}} . II-C Limitation of extended stabilizer formalism The current generation of quantum computers, such as IBM Quantum Heron, IBM Quantum Eagle, Rigetti Anka-3, use a native gate set that includes \{I,X,SX,R_{z}(\theta)/R_{x}(\theta),CZ/ISWAP\} . There is only one non-Clifford gate that can build any other gate using the circuit transpiler. Qimax is not limited to these basic gates; additional gates can be supported with minor modifications to the create_LUT() function. The Qimax’s gate set used in this paper including \{H,S,CX,R_{x}(\theta),R_{y}(\theta),R_{z}(\theta)\} . The 3- and 4-qubit gates will be considered in the future version. If a Clifford gate g is applied to the stabilizer state, then U\mathcal{P}_{n}U^{\dagger}g|\psi\rangle=g|\psi\rangle .