A Survey of Methods for Mitigating Barren Plateaus for Parameterized Quantum Circuits

TL;DR

This survey addresses barren plateaus in variational quantum algorithms (VQAs) by integrating formal measure-theoretic frameworks with circuit-design perspectives. It surveys how Haar measure, unitary ensembles, and unitary $t$-designs relate to expressibility and gradient concentration, and it connects these ideas to practical metrics like frame potential and Quantum Fisher Information. It reviews a taxonomy of circuit ansätze (hardware-efficient, physically motivated, QAOA/HVA, tensor-network-based) and analyzes how circuit depth, locality, initialization, and entanglement impact trainability, including noise-induced effects. It highlights mitigation strategies such as local cost functions, identity initialization, entanglement management, and geometry-aware updates (quantum natural gradient, Quantum Neural Tangent Kernel), arguing for problem-informed designs that balance expressibility and trainability to enable scalable VQAs on NISQ devices with applications in quantum chemistry, optimization, and quantum machine learning.

Abstract

Barren Plateaus are a formidable challenge for hybrid quantum-classical algorithms that lead to flat plateaus in the loss function landscape making it difficult to take advantage of the expressive power of parameterized quantum circuits with gradient-based methods. Like in classical neural network models, parameterized quantum circuits suffer the same vanishing gradient issue due to large parameter spaces with non-convex landscapes. In this review, we present an overview of the different genesis for barren plateaus, mathematical formalisms of common themes around barren plateaus, and dives into gradients. The central objective is to provide a conceptual perspective between classical and quantum interpretations of vanishing gradients as well as dive into techniques involving cost functions, entanglement, and initialization strategies to mitigate barren plateaus. Addressing barren plateaus paves the way towards feasibility of many classically intractable applications for quantum simulation, optimization, chemistry, and quantum machine learning.

Figures (10)

• Figure 1: A Narrow Gorge Parameter Landscape. Since the cost values are exponentially concentrated about the mean value on average, the parameter landscape derivative concentrates about this value leaving the majority of the landscape flat as defined by Eq. \ref{['eq: derivative-bp']} ̇arrasmith_equivalence_2022
• Figure 2: The current landscape of research surrounding PQC design in connection with algorithmic performance for variational tasks. Initial works have studied relationships between cost functions and barren plateaus as well as developed metrics to circuit properties including expressibility. Most recently, the geometry of a PQC has surfaced as a framework for understanding algorithmic performance.
• Figure 3: A layered brickwork ansatz scheme. Here, the 2-qubit gates have nearest neighbor connectivity and are used as entangling gates. Parameterized gates are single qubit rotation gates. Dimensionality is the topology of the physical qubits in a chain (1D). leone_practical_2022
• Figure 4: An HVA ansatz. In this example, each layer, d, contains parameterized 2-qubit local ZZ rotation gates followed by CZ entangling gates and a layer of single qubit parameterized X rotation gates wiersema_exploring_2020. In this example, there are 2d parameters for a d-depth ansatz.
• Figure 5: A typical mapping for a 1D tensor network to an MPS ansatz. The staircase pattern is a typical mapping of the local entangling properties of the system. Here, the MPS is used as an approximation solution to pre-train a HEA circuit dborin_matrix_2022.

Theorems & Definitions (1)

• Theorem 2.1: Solovay-Kitaev