Three-Step Protocol (3SP) Overview

• Three-Step Protocol (3SP) is a control architecture defined by an ordered three-stage transformation applied across diverse fields such as cryptography, quantum communication, random dynamics, navigation, and epitaxy.
• Its implementations range from three-pass secret exchanges and successive quantum rotations to controlled temporal ensembles and hierarchical navigation planning, each tailored to domain-specific challenges.
• The protocol’s modular design refines processes—enhancing security protocols, optimizing quantum state evolution, and enabling precise material growth—highlighting its broad interdisciplinary significance.

Searching arXiv for recent and canonical uses of “Three-Step Protocol” / “3SP” across domains to ground the article. “Three-Step Protocol” (3SP) is not a single standardized construction in the arXiv literature. The label is used for multiple unrelated procedures whose common feature is an ordered three-stage transformation. In cryptography it often denotes a three-pass exchange over an insecure channel; in quantum information it denotes either three-stage secure communication or a three-step measurement circuit; in random-matrix dynamics it denotes a quenched temporal ensemble with three Hamiltonian evolutions; in zero-shot navigation it denotes a global–local–global control loop; and in epitaxy it denotes a three-regime growth schedule for pseudo-substrate formation (Onur et al., 2017, 1803.02157, Zhou et al., 5 Apr 2026, Zheng et al., 29 Apr 2026, Zhang et al., 17 Dec 2025).

1. Nomenclature and scope

The term 3SP is therefore best treated as an overloaded research label rather than as a uniquely defined protocol family. Across domains, “three-step” may refer to three transmissions of the same secret-bearing object, three successive unitary segments, three perceptual or planning views, or three growth-condition regimes.

Domain	Meaning of 3SP	Representative paper
Classical cryptography	Three-pass/no-key transport or mission flow	(Onur et al., 2017, Fattahi et al., 2017)
Quantum communication	Three-stage qubit transmission or state discrimination	(1803.02157, Kanamori et al., 2010, Li et al., 2017)
Random quantum dynamics	Three quenched Hamiltonian evolutions	(Zhou et al., 5 Apr 2026)
Vision-and-language navigation	“Look forward, look now, look backward”	(Zheng et al., 29 Apr 2026)
Epitaxial growth	Roughening, nanostructuring, coalescence	(Zhang et al., 17 Dec 2025)

This multiplicity matters because the same acronym can otherwise suggest a false continuity between unrelated literatures. The cryptographic three-pass lineage is historically older, but recent uses in quantum dynamics, embodied AI, and materials science are methodologically independent.

2. Classical cryptographic lineages

In the classical three-pass tradition, the best-known formulation is Shamir’s no-key protocol, formalized in one paper through a public Abelian group action $(G,*)$ on a set $S$ with right-action notation $a\circ g$ (Onur et al., 2017). Alice chooses a secret $k\in S$ and private $g\in G$, sends $c_1=k\circ g$; Bob chooses private $h\in G$, sends $c_2=c_1\circ h$; Alice removes her layer and sends $c_3=c_2\circ g^{-1}$; Bob applies $h^{-1}$ and recovers $S$0. Algebraic correctness requires commutativity, since $S$1 only when $S$2 is Abelian.

That same paper proves an impossibility result for public Abelian-group instantiations of the protocol: if the action is public and transitive, an eavesdropper can recover a usable transformation from the observed relation between $S$3 and $S$4, and then compute the secret from $S$5 (Onur et al., 2017). The attack is explicit. If Eve finds $S$6 such that $S$7, then $S$8. The paper’s conclusion is that secure secret transport by the three-pass protocol over public Abelian groups is impossible in the intended information-theoretic sense.

A different cryptographic use of a three-step flow appears in a protocol for multipart military missions with two independent decision levels (Fattahi et al., 2017). There the mission-relevant sequence is operational decision-maker $S$9 $a\circ g$0 logistic decision-maker $a\circ g$1 $a\circ g$2 mission executor $a\circ g$3. The operational action $a\circ g$4 is encrypted under $a\circ g$5’s public key so that $a\circ g$6 forwards but does not read it; $a\circ g$7 appends a signed logistic action $a\circ g$8; $a\circ g$9 acknowledges receipt with a signed hash. The protocol is presented as a tagged asymmetric-key design intended to ensure secrecy, authentication, non-repudiation, and resistance to man-in-the-middle attacks (Fattahi et al., 2017).

Taken together, these papers show that the classical 3SP motif is structurally simple but security-sensitive. Commutativity can make decryption feasible for the intended parties, yet in public algebraic settings the same structure can collapse secrecy.

3. Quantum communication and photonic network protocols

In quantum cryptography, the canonical three-stage construction is Kak’s protocol, revisited as a single-qubit secure communication scheme usable for both QKD and secure direct quantum communication (1803.02157). Alice prepares one of two orthogonal states, applies a secret rotation

$k\in S$0

Bob applies $k\in S$1, Alice applies $k\in S$2, and Bob applies $k\in S$3. In the ideal case the commuting rotations cancel cleanly, the protocol does not require entanglement, and it does not require quantum memory (1803.02157).

The same paper identifies a severe implementability constraint under realistic noise (1803.02157). In its original form, Kak’s protocol works in the presence of collective rotation noise, but not in its original form under amplitude damping, phase damping, or collective dephasing noise. Collective dephasing can be handled only by moving to logical qubits in a decoherence-free subspace, which removes the protocol’s single-qubit simplicity. The analysis quantifies this by explicit fidelity expressions, including $k\in S$4 under collective rotation noise (1803.02157).

A related paper studies man-in-the-middle vulnerabilities in Kak’s three-stage design and argues that real-valued orthogonal transforms are too easy to imitate (0706.2888). Its proposed variation replaces the real-valued family with a complex unitary family,

$k\in S$5

for which the commuting relation becomes highly restrictive. The same paper also proposes a single-stage protocol in which Bob already knows Alice’s $k\in S$6, and $k\in S$7 is periodically refreshed after blocks of transmitted qubits (0706.2888).

The “Quantum Three-Pass Protocol” is a separate quantum three-pass construction based on single-photon polarization superposition states (Kanamori et al., 2010). Alice and Bob choose fresh session keys $k\in S$8 and $k\in S$9, encode each qubit with polarization rotations $g\in G$0, and transmit the same quantum information three times. For a single bit $g\in G$1, the state evolution is

$g\in G$2

Its security rationale is the no-cloning theorem rather than computational hardness, and the paper emphasizes deterministic use of transmitted bits rather than BB84-style sifting, while still noting that authentication is required to prevent man-in-the-middle attacks (Kanamori et al., 2010).

Another distinct quantum 3SP replaces Bell-state measurement in entanglement swapping by a three-step quantum-walk-like state-discrimination circuit (Li et al., 2017). Clare acts locally on photons $g\in G$3 and $g\in G$4 from two nonmaximally entangled pairs $g\in G$5: the first step uses line-dependent coin operations and path exchange, the second uses NOT gates plus phase compensation, and the third uses Hadamard coins followed by final position readout. Successful branches identify $g\in G$6 or $g\in G$7, projecting photons $g\in G$8 and $g\in G$9 onto $c_1=k\circ g$0. The success probability is $c_1=k\circ g$1, so the protocol performs entanglement swapping and entanglement concentration simultaneously (Li et al., 2017).

4. Three-step temporal ensembles and unitary design

In quantum many-body dynamics, 3SP has acquired a sharply different meaning: a quenched temporal ensemble generated by three fixed Hamiltonians and random evolution times (Zhou et al., 5 Apr 2026). The protocol is

$c_1=k\circ g$2

where $c_1=k\circ g$3 are sampled once and held fixed, while the only randomness comes from independently sampled times $c_1=k\circ g$4. The paper contrasts this with the two-step protocol $c_1=k\circ g$5 and studies both through the $c_1=k\circ g$6-th frame potential.

The key mechanism is that time averaging acts as an energy filter: $c_1=k\circ g$7 so $c_1=k\circ g$8 as $c_1=k\circ g$9 for uniform $h\in G$0 on $h\in G$1 (Zhou et al., 5 Apr 2026). In 2SP this filtering leaves two independent permutations in the frame-potential combinatorics, and in the flat-overlap idealization $h\in G$2, which is larger than the Haar value $h\in G$3 for all $h\in G$4. In 3SP the additional quench introduces extra overlap phases. The apparent $h\in G$5 freedom collapses because the surviving permutations are forced to coincide, $h\in G$6, leaving a single $h\in G$7 sector (Zhou et al., 5 Apr 2026).

For independent GUE Hamiltonians, the paper proves rigorously that

$h\in G$8

so 3SP realizes a unitary $h\in G$9-design asymptotically, whereas 2SP does not (Zhou et al., 5 Apr 2026). Finite-time robustness is also better: with $c_2=c_1\circ h$0, the 3SP correction scales as $c_2=c_1\circ h$1, while the 2SP correction scales as $c_2=c_1\circ h$2, i.e. $c_2=c_1\circ h$3 in the same normalization. The numerical threshold time $c_2=c_1\circ h$4 for a fixed tolerance at $c_2=c_1\circ h$5 is reported as about $c_2=c_1\circ h$6 for 2SP and about $c_2=c_1\circ h$7 for 3SP (Zhou et al., 5 Apr 2026).

This use of 3SP is not cryptographic and not communication-theoretic. It is a protocol for engineering Haar-like random dynamics from only a few fixed chaotic Hamiltonians.

5. Hierarchical planning in zero-shot vision-and-language navigation

In embodied AI, “Three-Step Nav” defines 3SP as a global–local–global planner for zero-shot vision-and-language navigation in continuous environments (Zheng et al., 29 Apr 2026). The three views are explicit. “Look forward” parses the instruction into ordered sub-instructions and landmarks; “look now” grounds the current sub-goal against the current observation and candidate viewpoints

$c_2=c_1\circ h$8

and “look backward” audits the trajectory $c_2=c_1\circ h$9 to detect drift before termination (Zheng et al., 29 Apr 2026).

The protocol operates as a prompt-based wrapper around a frozen MLLM and requires no gradient updates or task-specific fine-tuning (Zheng et al., 29 Apr 2026). In the local step, the model also estimates a landmark distance $c_3=c_2\circ g^{-1}$0, and when $c_3=c_2\circ g^{-1}$1 falls below a predefined threshold the current sub-goal is treated as ready for inspection. In the audit step the planner can invoke one of four meta-abilities: continue, stay, backtrack with trajectory truncation $c_3=c_2\circ g^{-1}$2, or look-around by visiting neighboring viewpoints and returning for reassessment (Zheng et al., 29 Apr 2026).

Reported validation-unseen results are state of the art in the paper’s zero-shot setting. On R2R-CE, Three-Step Nav reports $c_3=c_2\circ g^{-1}$3, $c_3=c_2\circ g^{-1}$4, $c_3=c_2\circ g^{-1}$5, and $c_3=c_2\circ g^{-1}$6; on RxR-CE, it reports $c_3=c_2\circ g^{-1}$7, $c_3=c_2\circ g^{-1}$8, $c_3=c_2\circ g^{-1}$9, and $h^{-1}$0 (Zheng et al., 29 Apr 2026). The ablation evidence is central to the protocol’s rationale: “only local view / no global forward-backward reasoning” gives $h^{-1}$1, $h^{-1}$2, $h^{-1}$3, while removing look backward gives $h^{-1}$4, $h^{-1}$5, $h^{-1}$6 (Zheng et al., 29 Apr 2026). In this literature, 3SP denotes an explicit reasoning schedule rather than a message-exchange or physics protocol.

6. Three-step growth in (In,Ga)N pseudo-substrate fabrication

In materials science, a three-step protocol has been proposed for fully in situ fabrication of relaxed, smooth $h^{-1}$7 pseudo-substrates on GaN templates by plasma-assisted molecular beam epitaxy (Zhang et al., 17 Dec 2025). The sequence intentionally changes growth conditions from N-rich to metal-rich and passes through three morphological stages: a roughened GaN layer, relaxed $h^{-1}$8 nanostructures, and a coalesced smooth $h^{-1}$9 layer (Zhang et al., 17 Dec 2025).

The first step grows rough GaN under N-rich conditions at $S$00 for $S$01 min with N flux $S$02 atom/(s·cm$S$03) and Ga flux $S$04 atom/(s·cm$S$05) (Zhang et al., 17 Dec 2025). The second step opens the In shutter, ramps the In cell from $S$06 to about $S$07 over $S$08 min to an expected In flux of $S$09 atom/(s·cm$S$10), keeps N and Ga fluxes constant, and continues growth for $S$11 more minutes, producing a columnar “brain-like” nanowall structure with mean width $S$12 nm and gaps $S$13 nm (Zhang et al., 17 Dec 2025). The third step switches to metal-rich conditions with Ga flux $S$14 atom/(s·cm$S$15), In flux $S$16 atom/(s·cm$S$17), and $S$18 min growth time, enabling coalescence into a smooth, fully closed pseudo-substrate (Zhang et al., 17 Dec 2025).

The reported structural and optical gains are quantitative. Sample #B, produced by the pseudo-substrate route, has In content $S$19, relaxation degree $S$20, and in-plane lattice constant $S$21, whereas the direct-growth reference #R has In content $S$22, relaxation degree $S$23, and in-plane lattice constant $S$24 (Zhang et al., 17 Dec 2025). Sample #C, grown by uninterrupted 3SP, has In content $S$25, relaxation degree $S$26, photoluminescence peak wavelength $S$27 nm, and linewidth $S$28 meV / $S$29 nm (Zhang et al., 17 Dec 2025). Panchromatic cathodoluminescence maps show a bright-area fraction of $S$30 for sample #B versus $S$31 for #R, and the final surface is reported as smooth with RMS roughness about $S$32 nm (Zhang et al., 17 Dec 2025). Here 3SP denotes a process-integration strategy rather than an information-processing algorithm.

7. Structural commonalities, boundary cases, and recurring confusions

Across these literatures, the common invariant is not the object manipulated but the logic of staged constraint refinement. In public-key-style three-pass cryptography, the same secret-bearing object is transformed, retransformed, and partially unwrapped across three transmissions (Onur et al., 2017). In the temporal-ensemble unitary-design setting, a single Hilbert-space evolution acquires three independent random phases through successive quenches (Zhou et al., 5 Apr 2026). In zero-shot navigation, one instruction is processed by global planning, local grounding, and global verification (Zheng et al., 29 Apr 2026). In epitaxy, one substrate passes through roughening, nanostructuring, and coalescence (Zhang et al., 17 Dec 2025). This suggests that “three-step” is best read as a control architecture rather than as a domain-independent algorithm.

The term also invites confusion with other “three-” constructions that are not 3SPs. A three-state BB84 variant is a protocol with three signal states $S$33, not a three-step protocol (Krawec, 2016). “Three-step nilpotent” in Lie algebra theory denotes a lower-central-series property, not a protocol at all (Ray, 2015). Co-TAP is a three-layer multi-agent framework organized as HAI, UAP, and MEK, but the paper explicitly states that it is not called “3SP” (An et al., 9 Oct 2025). Conversely, some papers use “single-step” precisely to reject a multi-step framing: the three-qubit-gate protocol based on anisotropic chiral interactions presents its main construction as genuinely single-step and its alternative as four-step, not three-step (Nguyen et al., 15 Mar 2025).

For that reason, any reference to “3SP” requires disciplinary disambiguation. In cryptography it usually implies a three-pass exchange and immediate questions about commutativity, authentication, and active attacks. In quantum dynamics it implies frame potentials and design order. In navigation it implies a global–local–global loop around a frozen MLLM. In epitaxy it implies a staged change of flux regime and morphology. The abbreviation is stable; the underlying object is not.