Symmetry-Grouped Hadamard Test

• The paper’s main contribution is redefining the generalized Hadamard test by strategically swapping measured operators to exploit more groupable commuting-term structures.
• It introduces a method that leverages commuting-term grouping and coherent circuit designs to reduce measurement overhead in parameterized quantum-circuit derivative estimation.
• The approach integrates low-depth circuit strategies and ancilla-control elimination to enhance quantum algorithm efficiency on hardware with connectivity constraints.

Searching arXiv for the cited works and closely related Hadamard-test literature. arXiv search query: (Li et al., 2024) arXiv search query: (Mastorakis et al., 25 Jul 2025) arXiv search query: (Patti et al., 2022) arXiv search query: (Faehrmann et al., 21 May 2025) “Symmetry-Grouped Hadamard Test” is not the name of a formally introduced method in the cited arXiv literature. The closest direct precursor is the generalized Hadamard-test framework developed for parameterized quantum-circuit derivative estimation, where the decisive idea is to restructure which Hermitian factor is implemented through controlled evolution and which factor is measured, so that commutativity-based measurement grouping can be exploited on the more favorable side of the expression (Li et al., 2024). In that restricted and explicit sense, the term can be used as an interpretive umbrella for Hadamard-test constructions that combine operator-role flexibility, commuting-term grouping, coherent aggregation of many contributions into a single interferometric estimator, and low-depth circuit simplifications. The symmetry content remains indirect: the literature explicitly discusses commuting Pauli grouping and, in one case, commuting conserved observables, but it does not formally develop a symmetry-sector, irrep, or stabilizer-based Hadamard-test theory (Li et al., 2024).

1. Concept and nomenclature

Within current usage, the most precise technical meaning of a “symmetry-grouped Hadamard test” is an informed extrapolation from methods that exploit structure in the operator being measured. The clearest such structure is commuting-term grouping. For first-order derivatives of parameterized quantum circuits, the relevant paper explicitly states that reversing the roles of ansatz generators and observables can exploit whichever side is “more groupable,” quantified through the number of commuting Pauli groups, and that this is the closest direct analogue to a symmetry-grouped Hadamard-test idea (Li et al., 2024).

That formulation is narrower than a literal symmetry-based construction. The same paper explicitly states what is absent: no method called “symmetry-grouped Hadamard test,” no formal use of symmetry sectors, conserved charges, irreducible representations, stabilizer symmetry classes, or symmetry-adapted measurement partitioning (Li et al., 2024). Accordingly, the term denotes not a canonical algorithm but a family resemblance among Hadamard-test variants that make structural grouping operationally useful.

A plausible implication is that symmetry enters only insofar as it induces exploitable operator structure. If a symmetry makes a generator or observable more amenable to commuting decomposition, then the generalized Hadamard-test formalism can benefit from that structure without ever naming symmetry as the organizing principle. This suggests that, in present literature, “symmetry-grouped” is best interpreted as a conceptual extension of measurement grouping rather than an established formal category.

2. Gradient-estimation setting and the generalized Hadamard identity

The primary mathematical setting is a parameterized quantum circuit

$$U_{1:n}(\boldsymbol{\theta}) = U_n(\theta_n)\cdots U_1(\theta_1) = e^{-iH_n\theta_n/2}\cdots e^{-iH_1\theta_1/2},$$

with objective

$$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$

For a parameter $\theta_j$, the derivative is written as

$$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$

where

$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$

The form $-\Im\langle \tilde H_j O\rangle$ is the basis for the generalized Hadamard constructions (Li et al., 2024).

The key abstraction is the Flexible Hadamard Test. For a product of Hermitian operators,

$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$

the circuit estimates

$$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$

while allowing an arbitrary chosen factor $\hat H_i$ to be the measured operator (Li et al., 2024). Operationally, the ancilla is initialized in

$\frac{1}{\sqrt 2}(\ket 0 - i\ket 1),$

the left branch applies $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$0, the right branch applies $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$1, and the readout is $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$2. The resulting identity is

$$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$3

This freedom is the foundational mechanism behind any grouped or symmetry-aware interpretation. The crucial point is not merely that a different interference identity exists, but that the measured Hermitian factor can be selected strategically. That selection determines where measurement grouping becomes available.

3. Operator-role reversal and commuting-group optimization

Applying the flexible identity to the two-factor derivative expression produces the Reversed Hadamard Test. Standard Hadamard-test gradient estimation measures the observable $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$4 and implements generator-side terms as controlled operations. The reversed variant measures the generator side and implements observable-side terms as controlled operations; the authors describe this explicitly as exchanging the roles of ansatz generators and observables (Li et al., 2024).

If

$$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$5

then the reversed construction yields

$$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$6

and

$$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$7

The significance lies in the fact that the measured operator is now $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$8 rather than $$f(\boldsymbol{\theta}) = \bra{\Psi}U_{1:n}^\dagger O U_{1:n}\ket{\Psi}.$$9.

The paper makes the grouping logic explicit through commuting Pauli partitions:

$\theta_j$0

with

$\theta_j$1

The same logic can be applied to the generator side through $\theta_j$2. The optimization criterion is therefore whether $\theta_j$3 or $\theta_j$4 is more groupable (Li et al., 2024).

The resulting first-order tradeoff can be summarized as follows.

Method	Distinct circuits	Structural point
PSR	$\theta_j$5	Restricted generators
HT	$\theta_j$6	Measure $\theta_j$7
DHT	$\theta_j$8	Ancilla-free HT
RHT	$\theta_j$9	Measure $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$0
RDHT	$$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$1	Ancilla-free RHT

These formulas make the nearest “symmetry-grouped” reading precise. If structural information makes $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$2 easier to group than $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$3, then reversed methods reduce distinct measurement settings. The paper reports that first-order reversed methods can improve circuit execution count by up to $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$4 in practical settings, including QAQC-like regimes where hardware Hamiltonians have generator decompositions scaling as $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$5 (Li et al., 2024).

A common misconception is that reversal is merely a cosmetic rewriting of the same estimator. The explicit circuit-count formulas show otherwise: the measured object is the object to which commuting-group optimization applies, so exchanging measured and controlled roles changes the resource profile.

4. Higher-order derivatives as coherent grouping

The same work extends the grouping idea beyond first-order measurement reduction by introducing the $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$6-fold Hadamard Test for higher-order derivatives. The $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$7-th derivative is written as

$$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$8

Instead of expanding the nested commutator into $$\frac{\partial f(\boldsymbol{\theta})}{\partial \theta_j} = -\Im\!\left( \bra{\boldsymbol{\theta}}\tilde H_j O\ket{\boldsymbol{\theta}} \right) = \frac{i}{2}\bra{\boldsymbol{\theta}}[\tilde H_j,O]\ket{\boldsymbol{\theta}},$$9 terms and measuring them separately, the $\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$0-fold construction prepares

$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$1

applies controlled $\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$2 from ancilla $\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$3, and measures

$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$4

The estimator is

$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$5

This evaluates one $\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$6-th-order derivative entry with a single circuit (Li et al., 2024).

The relevant comparison is explicit. For one fixed derivative entry, the paper gives:

Method	Circuits per entry	Qubits
PSR	$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$7	$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$8
HT	$\tilde H_j = U_{(j+1):n} H_j U_{(j+1):n}^\dagger.$9	$-\Im\langle \tilde H_j O\rangle$0
DHT	$-\Im\langle \tilde H_j O\rangle$1	$-\Im\langle \tilde H_j O\rangle$2
$-\Im\langle \tilde H_j O\rangle$3-fold HT	$-\Im\langle \tilde H_j O\rangle$4	$-\Im\langle \tilde H_j O\rangle$5

This is not measurement grouping in the narrow commuting-Pauli sense. It is, however, a coherent grouping of exponentially many derivative contributions into one interferometric estimator. The paper explicitly notes that the full derivative tensor still carries an overall $-\Im\langle \tilde H_j O\rangle$6 scaling in the number of entries, but for each individual entry the $-\Im\langle \tilde H_j O\rangle$7-fold method avoids the $-\Im\langle \tilde H_j O\rangle$8 term explosion (Li et al., 2024).

An informed interpretation is that this constitutes a second axis of “grouped Hadamard testing.” First-order reversal groups by choosing the more groupable measured endpoint; higher-order interference groups by coherently packing an exponentially expanded commutator into a single circuit.

5. Related resource-compression variants of the Hadamard test

Other papers enlarge the notion of grouped Hadamard testing through structural compression rather than commuting-term grouping. In a variational SDP setting, the Hadamard test is used to encode an entire objective matrix into a single controlled unitary,

$-\Im\langle \tilde H_j O\rangle$9

so that a single ancilla expectation value estimates

$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$0

The paper states that this avoids “separately estimating exponentially many expectation values,” and therefore constitutes a coherent aggregation of many objective contributions into one overlap measurement (Patti et al., 2022). It is not symmetry grouping in the usual VQE sense, but it is a strong example of structure-compressed Hadamard testing.

That same work also combines the objective-side Hadamard compression with commuting Pauli-$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$1 constraint families. For low-order amplitude constraints,

$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$2

the terms commute and “can be estimated as $$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$3 different marginal distributions from a single set of $$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$4-qubit $$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$5-axis measurements,” with

$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$6

This is a genuine grouped-measurement mechanism, although it applies to constraints rather than to the core ancilla interference signal (Patti et al., 2022).

A distinct line of work studies low-depth Hadamard implementations. One paper states the following simplification rule: “Given the Hadamard test framework, together with the initialization of every qubit in the $$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$7 state, any quantum gate operation with at least one control on a non-ancilla qubit does not require a control from the ancilla qubit” (Mastorakis et al., 25 Jul 2025). In that construction, once an ancilla-triggered branch is established, subsequent conditionals that already have register controls can shed the ancilla control. The paper gives a concrete example: removing redundant ancilla control from a Toffoli-type gate can turn a three-qubit Toffoli into a single CNOT, saving five two-qubit gates and at least nine single-qubit gates in a standard decomposition (Mastorakis et al., 25 Jul 2025).

This low-depth work is highly relevant to grouped Hadamard-test design because grouped schemes often create shared ancilla skeletons. The paper does not explicitly introduce symmetry grouping, commuting-term grouping, or Pauli grouping, but it supplies a circuit-level backbone that could be combined with such strategies. In its Burgers-dynamics application, the low-depth design reduces two-qubit count $$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$8 by roughly a factor of 2–3x on both superconducting and trapped-ion hardware, and the trapped-ion results are especially favorable because all-to-all connectivity prevents routing overhead from reintroducing ancilla costs (Mastorakis et al., 25 Jul 2025).

6. Toward symmetry-resolved Hadamard testing

The paper closest to a genuine symmetry-oriented extension does not group terms at circuit-construction time; instead, it upgrades the Hadamard test into a joint ancilla–system measurement primitive. After the usual Hadamard circuit, the ancilla is measured together with the work register, and the work-register output is processed through classical shadows. The formal identities are

$$\left\langle \prod_{i=1}^m \hat H_i \right\rangle_{\ket{\boldsymbol{\theta}}} = \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}},$$9

together with

$$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$0

Thus the protocol estimates not only the usual Hadamard quantity $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$1, but also operator-dressed interference terms such as $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$2 through correlators like

$$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$3

and

$$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$4

(Faehrmann et al., 21 May 2025).

The paper explicitly states one symmetry-adjacent consequence: for $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$5, the expectation value of operators commuting with $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$6 can be estimated from $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$7. Since

$$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$8

any $$\Im\!\left( \bra{\boldsymbol{\theta}}\hat H_1\hat H_2\cdots \hat H_m\ket{\boldsymbol{\theta}} \right)$$9 with $\hat H_i$0 satisfies

$\hat H_i$1

This means one Hadamard-test data stream, supplemented by shadows on the work register, can estimate many conserved quantities simultaneously (Faehrmann et al., 21 May 2025).

This is the strongest formal route toward a true symmetry-grouped Hadamard test in the literature surveyed here. The same paper does not explicitly provide sector projectors, symmetry-adapted variance bounds, or a theorem specialized to grouped symmetry estimation. Nevertheless, the ingredients are present: ancilla–system correlators, linearity in the measured observable $\hat H_i$2, and a shadow-based mechanism for amortizing many target observables over the same circuit family. A plausible implication is that symmetry-sector projectors $\hat H_i$3 could be used to form symmetry-resolved interferometric quantities such as $\hat H_i$4 and $\hat H_i$5, but that step is not carried out explicitly in the paper (Faehrmann et al., 21 May 2025).

7. Limitations, misconceptions, and current status

The main limitation is terminological precision. None of the cited works formally defines a “symmetry-grouped Hadamard test.” The closest explicit constructions are: measurement-side flexibility and commuting-group optimization for derivatives (Li et al., 2024), coherent objective aggregation into a single controlled unitary (Patti et al., 2022), low-depth ancilla-control elimination for shared conditional structure (Mastorakis et al., 25 Jul 2025), and joint ancilla–system readout enabling simultaneous estimation of many observables, including conserved ones (Faehrmann et al., 21 May 2025).

A second limitation is that commuting-group optimization is not the same as symmetry decomposition. The derivative-estimation paper discusses full commutativity and notes that other grouping notions, such as qubit-wise commutativity, could also be integrated, but it does not formulate the method in terms of symmetry sectors (Li et al., 2024). Likewise, the SDP paper exploits operator synthesis and commuting $\hat H_i$6-basis constraints rather than graph symmetries (Patti et al., 2022).

A third limitation is hardware realism. Hadamard tests that rely on ancillas and controlled unitaries can incur substantial routing and SWAP costs on connectivity-limited devices. The low-depth trapped-ion study emphasizes exactly this point: ancilla-control elimination is most effective when all-to-all connectivity prevents routing from dominating the savings (Mastorakis et al., 25 Jul 2025). This matters for any future symmetry-grouped construction, since grouping often increases reliance on shared ancilla control structure.

The most accurate present-day summary is therefore narrow. A symmetry-grouped Hadamard test is not yet an established named technique. The extant literature instead provides four concrete building blocks: choose the measured Hermitian factor so that the more groupable operator is measured; compress many derivative or objective contributions into one interference circuit; strip redundant ancilla controls from already register-controlled branches; and reuse Hadamard-test shots to estimate many conserved or symmetry-related observables through joint ancilla–system measurement. Together these results define the current technical foundation from which an explicitly symmetry-adapted Hadamard-test formalism could plausibly emerge (Li et al., 2024).