---
title: Barren Plateaus in Quantum Algorithms
url: https://www.emergentmind.com/topics/barren-plateaus-bps
type: topic
---

# Barren Plateaus in Quantum Algorithms

A barren plateau (BP) in variational quantum algorithms (VQAs) is characterized by the exponential suppression of the variance of cost-function gradients with respect to the number of qubits or circuit depth. This exponential vanishing of the gradient renders parameter optimization landscapes almost featureless as the system size increases, making the efficient training of parameterized quantum circuits (PQCs) fundamentally intractable in practice. The BP phenomenon poses a critical challenge to scalable quantum machine learning, quantum simulation, and hybrid quantum–classical optimization workflows, dominating the field’s theoretical discourse and influencing ansatz design, initialization protocols, and mitigation strategies.

## 1. Formal Definition and Mechanisms

Barren plateaus are defined mathematically by the property that, for a cost function $C(\boldsymbol{\theta}) = \langle 0|U^\dagger(\boldsymbol{\theta})\,\mathcal{O}\,U(\boldsymbol{\theta})|0\rangle$ associated with an $n$-qubit PQC $U(\boldsymbol{\theta})$ and observable $\mathcal{O}$, the variance over parameter randomizations satisfies
\[
\operatorname{Var}[\partial_k C] \le F(n), \quad F(n)\in o(b^{-n}),\, b > 1.
\]
Typically, the mean gradient vanishes: $\mathbb{E}[\partial_k C] = 0$, and the probability of observing a nonzero gradient is exponentially suppressed with increasing $n$ (or circuit depth $L$), as guaranteed by Chebyshev's inequality. This vanishing variance also extends to higher-order derivatives such as the Hessian and all mixed partials of the cost, precluding the efficacy of gradient- or Hessian-based optimization methods in the BP regime [2008.07454].

The mechanistic origin of BP in most expressive PQCs is closely tied to the emergence of approximate unitary $2$-designs in the circuit ensemble. In randomly initialized, sufficiently deep PQCs, the distribution of unitaries quickly approaches the Haar measure, so that expectation values and their derivatives “concentrate” (concentration of measure) around zero exponentially with $n$. This effect is also modulated by the entanglement properties of the state, the locality of the observable, the Lie algebra generated by the gates, and noise in the quantum hardware [2405.00781, 2309.09342].

## 2. Destructive Interference and Diagnostic Frameworks

Recent advances reframe the BP phenomenon through the lens of destructive interference among termwise gradient contributions. For Hamiltonians decomposed as $H = \sum_\alpha c_\alpha P_\alpha$, and for gradient components $a_{\alpha,k} := c_\alpha \partial_k \langle P_\alpha \rangle$, the net gradient $\partial_k \langle H \rangle = \sum_\alpha a_{\alpha,k}$ can be made exponentially small not because each $|a_{\alpha,k}|$ is small individually, but due to random sign cancellations across terms ("random-sign cancellation regime"). Quantitative diagnostics include:

- **Cancellation ratio** $R_k = |\sum_\alpha a_{\alpha,k}| / \sum_\alpha |a_{\alpha,k}|$
- **Effective term count** $N_{\mathrm{eff},k} = (\sum_\alpha |a_{\alpha,k}|)^2 / \sum_\alpha a_{\alpha,k}^2$
- **Interference-quality measure** $B_{\mathrm{eff},k} = R_k \sqrt{N_{\mathrm{eff},k}}$

An ensemble in the random-sign cancellation regime typically shows $B_{\mathrm{eff},k}\approx 0.8$ (for random $a_{\alpha,k}=\pm 1$ with equal probability), with the mean gradient norm vanishing exponentially with system size. Crucially, this framework distinguishes between suppression caused by cancellation and that arising from pre-cancellation activity scale $Q_k = \sum_\alpha a_{\alpha,k}^2$ [2605.01319].

## 3. Origins: Expressibility, Locality, Entanglement, and Noise

The emergence of BPs in VQAs can be attributed to several interrelated factors:

- **Circuit expressibility**: Hardware-efficient, unstructured ansatzes with depth $L \sim \mathrm{poly}(n)$ typically generate large dynamical Lie algebras ($\dim g \sim 4^n$) and quickly approach a 2-design, enforcing exponential suppression of gradient variance [2309.09342, 2405.00781].
- **Observable and cost locality**: Global observables ($\mathcal{O}$ acting nontrivially on $O(n)$ qubits) inevitably cause BPs at any circuit depth. In contrast, strictly local costs can avoid BPs if the circuit is sufficiently shallow ($L=O(\log n)$) [2406.14285, 2407.17706].
- **State entanglement**: Volume-law entangled input states can push even local observables into BP regimes via the pre-formation of maximal mixedness in local reductions [2309.09342, 2201.08194].
- **Noise**: Unital noise channels (e.g., depolarizing) induce deterministic flattening ("noise-induced BPs"), driving both expectation values and gradients to their maximally mixed values exponentially in depth [2310.15037, 2507.02043].

## 4. Consequences for Optimization and Trainability

In the BP regime, the required number of measurement shots or function evaluations to estimate gradients with any fixed precision scales exponentially with the number of qubits. This affects all derivative-based strategies, including higher-order methods, as not only are the gradients exponentially small, but so too are the Hessians and all higher derivatives—thereby creating a fundamental obstruction to efficient training [2008.07454]. 

Attempts to escape BPs by exploiting Hessian structure, high-order derivatives, or "jumping" to stochastic noise for escaping plateaus have been shown analytically to not circumvent the exponential cost [2008.07454, 2205.05786]. Even initialization or ansatz selection strategies that boost initial gradient norms can merely shift, but not fundamentally remove, the exponential scaling [2606.18515, 2412.06462, 2603.18479].

Moreover, it has been demonstrated that even in shallow circuits that avoid BPs, the loss landscape is “swamped with traps”—an overwhelming number of poor local minima, where most parameter choices fail to escape suboptimal configurations. In the BP regime, exponentially many trivial local minima persist, optimally solving only a few cost-function terms while leaving the rest flat [2405.05332, 2205.05786]. This establishes that non-vanishing gradients alone are not sufficient for trainability.

## 5. Mitigation Strategies and Their Limitations

A variety of methods have been developed to delay or sidestep barren plateaus. These can be organized into several categories [2407.17706, 2406.14285]:

| Strategy Category         | Representative Methods                                              | Principle                    |
|--------------------------|---------------------------------------------------------------------|------------------------------|
| Initialization           | Small-angle (narrow range) init, classical-network seeding, LLM-driven bootstrapping, first-moment engineering [2412.06462, 2205.13418, 2502.13166, 2606.18515] | Avoid early 2-design formation; steer toward high-gradient regions |
| Optimization Procedure   | Layerwise/blockwise growth, staged/rolling training, classical shadows for entropy monitoring [2402.05227, 2201.08194] | Keep circuit expressibility limited during optimization |
| Model/Architecture       | Problem-aligned ansatz (e.g., HVA/QAOA), symmetry-restricted, local-depth circuits [2605.01319, 2407.17706, 2309.09342] | Align generator structure with physics; reduce effective Lie algebra dimension |
| Regularization/Noise     | Engineered dissipation (Markovian loss), entanglement penalties, reset protocols [2310.15037, 2507.02043] | Locally inject purity, break up expressibility, penalize volume-law entanglement |
| Measurement/Hybrid       | Post-selection, classical-quantum hybrids, intermediate measurement [2407.17706] | Conditioning on structured subspaces |

While these approaches can delay the onset of BPs or partially restore trainability in regimes of moderate $n$, none are universal. In particular, mitigation strategies that lift the average variance may bias the optimizer toward different “trainable pockets” of the landscape, and exponentially many first-moment-distinct initializations exist, yielding non-equivalent minima [2606.18515]. For problems requiring global exploration (e.g., learning unknown scrambling unitaries), BPs remain unavoidable, representing a fundamental resource barrier [2009.14808].

Dimension is a further aggravating factor: for qudit PQCs, the gradient variance scales as $O((d')^{-n})$, amplifying the plateau problem with increased local dimension [2405.08190].

## 6. Unified Theoretical Characterization

A unified theoretical framework ties the sources of BPs to the dynamical Lie algebra $g$ generated by the circuit’s gate set [2309.09342]. The variance of the loss is given by
\[
\operatorname{Var}[l(\theta)] = \sum_{j=1}^{k-1} \frac{P_{g_j}(\rho)\,P_{g_j}(O)}{\dim(g_j)}
\]
where $P_{g_j}(X)$ is the $g_j$-purity of $X$. Gradient suppression arises due to (i) large $\dim(g_j)$ (expressivity), (ii) small $P_{g_j}(\rho)$ (input-state entanglement), (iii) small $P_{g_j}(O)$ (observable locality/non-alignment), or (iv) noise-induced purity loss.

This framework accommodates and extends all known origins of BPs, and provides an exact predictive theory for whether a given architecture, cost, and initial state will admit trainable gradients [2309.09342].

## 7. Open Problems and Future Directions

Key open problems include:

- Formalizing the interplay between destructive interference and expressibility/Lie-algebraic mechanisms in more general Hamiltonians, cost functions, and physical architectures [2605.01319, 2603.18479].
- Systematically designing ansatz architectures that balance expressibility against trainability, possibly guided by Lie algebra size, $g$-purity, and first-moment diagnostics.
- Rigorous development of initialization strategies that selectively break average-case concentration without introducing exponential ambiguity in the optimal pocket [2606.18515].
- Quantifying the relationship between the absence of BPs and classical simulability, the effect of hardware-specific non-unital noise, and the interplay of BPs with local minimum “traps” [2405.00781, 2405.05332].
- Connecting classical shadow-based entropy-monitoring and real-time regularization with hardware-realizable error-mitigation techniques [2201.08194, 2310.15037, 2507.02043].
- Generalizing strategies to the qudit setting, high-dimensional architectures, and hybrid quantum-classical models [2405.08190].

Resolving these challenges will be crucial for scaling quantum variational algorithms beyond the limits set by the barren plateau phenomenon.

Source: https://www.emergentmind.com/topics/barren-plateaus-bps