Priyam Srivastava

Research

Quantum sensing, estimation, and network control

My research investigates how quantum-information protocols can be designed and optimized under constraints imposed by physical interactions, available control operations, measurement design, and finite quantum-memory coherence.

I work primarily on two interconnected areas: quantum sensing and metrology, where I study variational probe preparation and measurement strategies, and quantum networking, where I investigate learning-based link control and entanglement distribution.

My approach combines analytical modeling, quantum-information methods, numerical simulation, and optimization. I am particularly interested in identifying which physical or operational constraints determine achievable performance, and how optimization can help reveal those limits.

Variational quantum sensing

Structured phase estimation: variational probe and readout design

Quantum-enhanced sensing depends not only on how much information a probe state can encode, but also on whether the available measurement operations can extract that information. I investigate this distinction in dipolar-interacting spin systems with restricted, finite-depth control.

Research question

How does including the measurement circuit during variational probe optimization affect the achievable precision of structured single-parameter quantum sensing?

Approach

I model systems of 2–8 dipolar-interacting spin-1/2 particles using PennyLane and JAX. Parameterized circuits prepare the sensing probe through alternating interaction dynamics and collective control rotations. CMA-ES is used for numerical optimization.

For structured phase estimation, the parameter is encoded through prescribed spatial profiles across the spins. The sensing performance is evaluated using classical Fisher information (CFI), which depends on the measurement probabilities, and quantum Fisher information (QFI), which characterizes the local information available in the quantum state.

I compare two optimization strategies using matched encoder and readout circuit resources.

Sequential versus joint optimization

Sequential optimization

Optimize the probe using a fixed Ramsey-readout baseline, then hold the probe fixed and optimize the readout parameters.

Joint optimization

Optimize probe-preparation and readout parameters together, using the measured CFI as the objective.

Both protocols use the same deployed circuit resources. The comparison isolates the effect of including the readout during probe selection rather than attributing an advantage to additional circuit depth.

Main findings

  • Joint probe–readout optimization improved CFI by 14.9% relative to sequential optimization for uniform encoding with eight spins and three circuit layers.
  • Fixed-probe readout re-optimization and direct-QFI optimization help distinguish limitations of probe preparation from limitations of measurement extraction.
  • Comparisons with Ramsey-based measurements and simpler measurement choices clarify when optimized probes require additional readout control.

Scientific perspective

A probe with high QFI does not necessarily yield equally high CFI under a restricted measurement. Probe optimization and measurement design must therefore be assessed together when evaluating implementable sensing protocols.

These results come from numerical simulations of the specified spin models. The optimized circuits have not been experimentally implemented in this work. A journal manuscript extending the QCE 2025 study is in preparation.

Earlier preprint IEEE QCE 2025 Publications

Multiparameter quantum metrology

Joint magnetometry and gradiometry on dipolar spin chains

Many sensing applications require estimating multiple physical quantities from the same quantum system. This introduces questions beyond maximizing sensitivity to a single parameter, including identifiability and trade-offs between different estimation directions.

Research question

Can variationally prepared quantum probes improve the simultaneous estimation of a magnetic-field offset and spatial gradient, and how should their joint sensitivity be quantified?

Approach

I study joint magnetometry and gradiometry using dipolar-interacting spin chains with variational probe-preparation circuits. The sensing parameters are a uniform magnetic-field component and a spatial field gradient.

Performance is characterized through the Fisher-information matrix, rather than a scalar sensitivity for either parameter alone. The determinant provides a measure of joint local information volume when the information matrix is nonsingular.

A central theoretical observation is that a conventional GHZ probe produces a rank-one QFI matrix for the two commuting sensing generators in this model. It is therefore insufficient for independently identifying both parameters locally.

Main result

For a five-spin system with three variational circuit layers, the optimized joint-estimation performance reached approximately 92% of the studied diagonal-encoder QFI determinant benchmark, corresponding to about 4.2 times the standard-quantum-limit determinant.

Significance

This problem demonstrates why sensitivity to an individual parameter is not sufficient for evaluating multiparameter sensing. Highly entangled states can be useful for one estimation direction while failing to distinguish multiple parameter combinations.

Variational optimization provides a numerical approach to exploring probe structures that distribute sensitivity across the parameters of interest, subject to the specified interaction and circuit constraints.

The result is specific to the studied finite spin system and optimization setting. The quoted benchmark does not represent a demonstrated universal optimum across all possible probe preparations.

arXiv:2605.03906 IEEE QCE 2026 · QTEM

Reinforcement learning and quantum networking

Quantum-memory control and entanglement-swapping strategies

Long-distance quantum communication relies on generating, storing, purifying, and swapping entanglement across intermediate nodes. The performance of these operations depends on stochastic link generation and finite memory coherence.

Research question

How do link-level quantum-memory control and network-level entanglement-swapping strategies interact, and when does one swapping protocol outperform another under finite-coherence constraints?

Approach

I developed a simulation framework using Python, PyTorch, and Gymnasium to investigate quantum repeater chains. At the elementary-link level, classical REINFORCE agents learn policies for managing entangled-pair resources.

The link controller selects among waiting, discarding, purification, and delivery decisions. The resulting entangled pairs are supplied to a network-level simulator, where different prescribed swapping strategies can be evaluated.

In the ACM PQQS 2026 study, the same link-control policy is used for both swapping strategies at each operating point. Thus, the comparison concerns the network protocol rather than an RL agent learning to choose between protocols.

Two network-level strategies

Sequential swapping

Entanglement is extended progressively along the chain. Previously established segments may need to remain in quantum memory while subsequent links become ready.

Synchronized swapping

The protocol waits until the required elementary links are simultaneously ready before performing the end-to-end swapping operation.

The strategies create different requirements for storing entanglement across link-generation cycles. This difference becomes particularly important when memory coherence times are short.

Main findings

  • In the studied four-hop chain, synchronized swapping remains viable in short-memory regimes where sequential swapping is strongly suppressed.
  • Sequential swapping can provide an advantage at sufficiently long memory coherence times, reaching 19.1% higher secret-key rate in a tested homogeneous 5 km-per-link configuration.
  • A fidelity-based analytical condition characterizes the onset of viable sequential swapping. The ratio of memory coherence time to link latency helps organize the observed regimes.
  • Heterogeneous link lengths introduce additional performance effects associated with bottleneck placement and memory occupancy.
  • The link-level RL study also identified an over-purification failure mode that could prevent successful training.

Scientific perspective

A swapping strategy cannot be evaluated independently of link-level resource generation and quantum-memory dynamics. Protocol performance depends on how these processes interact across different operating regimes.

This study uses simulated four-hop repeater chains, Werner-state noise modeling, and six-state QKD secret-key-rate evaluation. Its conclusions apply to the modeled architectures and operating assumptions, rather than all possible quantum networks.

ACM PQQS 2026 paper arXiv:2605.04047

Diagram of a quantum-network simulation with reinforcement-learning-based link management and network-level entanglement swapping.
Link-layer controllers manage elementary entanglement resources. Network-level simulations then compare sequential and synchronized swapping under identical link-control policies, isolating the effects of protocol choice and memory coherence.

Collaborative research and reviews

Alongside my primary research projects, I contribute to collaborative work connecting quantum information with sensing applications and machine-learning methods.

Quantum biomedical sensing

I am an equal-contribution coauthor of Four Generations of Quantum Biomedical Sensors, which develops a framework for understanding the evolution of quantum technologies in biomedical sensing.

Artificial intelligence for quantum information

I authored the section on AI for quantum networking in When AI Meets Quantum Information: A Comprehensive Review, examining the use of learning-based methods in quantum-network management and control.

Illustration: GHZ sensitivity under dephasing

Entanglement can improve quantum sensing precision, but it can also increase sensitivity to noise. The following idealized model compares product-state and GHZ-state probes at fixed interrogation time under independent dephasing.

This interactive illustration is educational and is separate from the variational sensing results presented above.

8
0.00

product uncertainty

0.354

GHZ uncertainty

0.125

product / GHZ uncertainty

2.83x

product GHZ number of qubits

Model assumptions: Each qubit retains coherence e−d, where d = Γt. At fixed interrogation time, the local phase-uncertainty bounds are ed/√N for a product probe and eNd/N for a GHZ probe, with one normalized repetition. Preparation and measurement are ideal, and repetitions are assumed independent. A product/GHZ uncertainty ratio above one favors GHZ, whereas a ratio below one favors the product probe. The comparison does not optimize interrogation time, account for preparation overhead, or include quantum error correction. Lines connect integer qubit counts. These are analytical model predictions, not experimental measurements or numerical results from my publications.

Earlier research

Research directions

I am interested in extending my work on sensing and networked quantum systems toward distributed quantum sensing, including the interplay between multiparameter estimation, entanglement resources, local control, and communication constraints. My broader interests also include quantum algorithms, quantum simulation, quantum machine learning, and optimization methods for quantum-information processing.