{"id":"d1f84133-2585-40d1-a257-d2d5d0573e56","arxiv_id":"2412.18771","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A closed-form secret key rate expression is derived for RIS-assisted MIMO continuous-variable QKD at THz with least-squares channel estimation and collective Gaussian eavesdropping.","lead":"This paper derives a formula for the secret key rate of a quantum key distribution system that uses many antennas and a reconfigurable intelligent surface at terahertz frequencies, when the wireless channel is imperfectly estimated. It shows how pilot power, distance, and surface phase affect the key rate for two detector types.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (44) assumes the LS channel-estimation error is diagonalized by the SVD of the estimated channel; the paper gives no proof, and C_RIS in Eq. (19) is generically full, so the summed per-mode SKR is not an exact total SKR.","rationale":"The reader's weakest assumption is the right one. The per-mode decomposition is load-bearing because every subsequent quantity in Eq. (44) -- the Shannon term (24) and the Holevo terms (32) and (43) -- is evaluated per mode with a scalar sigma_RISi^2. If C_RIS is not diagonal in the SVD basis, these terms cannot be summed to give the total SKR: mutual information and Holevo information are not additive across modes unless the modes are independent, and the received vector in (16) has cross-couplings from the off-diagonal part of U^dag_{RISLS} Delta H_RIS V_{RISLS}. I also note the separate ML-covariance bias in (21)-(22) and the signal-dependent nature of n_RIS; these reinforce caution about the numerical SKR values, but the diagonal assumption is the primary mathematical gap. The proposed small-scale check would determine whether the gap is quantitative or merely formal. If the off-diagonal correlations are tiny in the regime of interest, Eq. (44) could still be a good approximation, but the paper does not currently make or support that claim. Thus the verdict remains CONDITIONAL: the closed-form result should be re-derived with the full covariance matrix, or the diagonalization condition should be explicitly proved and verified.","tokens_in":20204,"tokens_out":16145,"duration_ms":153762,"concrete_test":"Run a minimal 2x2 case (N_T=N_R=2, L_p>2, generic H_d, H_f, H_g, V_0>0, heterodyne detection). Compute C_RIS from Eq. (19) and rho = |C_RIS(1,2)|/sqrt(C_RIS(1,1) C_RIS(2,2)). If rho is non-negligible (e.g., >0.1), compute the exact RR SKR from the full r-mode Gaussian covariance matrix of the received modes (standard multimode Gaussian SKR formula) and compare it with Eq. (44) using only sigma_RISi^2. A difference above numerical tolerance (e.g., >5%) would settle that the per-mode sum is not the exact total SKR and that Eq. (44) needs either a diagonalization proof or replacement by a full-covariance formula.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (44) is exact only if the r parallel modes are independent Gaussian channels whose estimation-error noises have variances sigma_RISi^2 = 0.5 C_RIS(i,i). The paper does not establish this. From (19), C_RIS = (2 V_a N_T)/(V_p L_p) U^dag_{RISLS} C_tilde_n U_{RISLS}; because tilde_N = H_RIS Psi_0 + N_het, C_tilde_n contains 2 V_0 H_RIS H_RIS^dag, and U_{RISLS} comes from the SVD of H_RISLS = H_RIS + Delta, not of H_RIS. Hence U^dag_{RISLS} C_tilde_n U_{RISLS} is generically a full matrix: the off-diagonal entries of U^dag_{RISLS} Delta H_RIS V_{RISLS} in (16) are cross-mode couplings, not independent per-mode noises. Replacing the vector noise by its diagonal variances and summing per-mode rates in (23)-(44) discards the joint covariance of the multimode Gaussian state. The exact SKR under a collective Gaussian attack is a function of the full covariance matrix of all r modes; no proof is given that it reduces to (44) unless C_RIS is diagonal in the SVD basis. As written, the closed-form claim is therefore an unproven approximation, not a derived exact expression.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper considers an RIS-assisted MIMO CV-QKD link at THz frequencies. Bob estimates the effective channel by least squares from pilot symbols, and the estimated channel state information is fed back to Alice over a public authenticated channel that is also accessible to Eve. Alice uses Gaussian modulation, Bob uses homodyne or heterodyne detection, and Eve is modeled as employing a collective Gaussian entanglement attack. The main claim is Eq. (44), a closed-form secret key rate (SKR) expression for reverse reconciliation that includes the channel estimation overhead and the noise covariance induced by LS estimation. Numerical evaluations study the SKR versus distance, pilot power, pilot length, number of RIS elements, and RIS phase configuration.","tokens_in":20503,"tokens_out":16772,"duration_ms":157388,"significance":"If correct, the closed-form SKR would be a useful design tool for RIS-assisted THz CV-QKD under imperfect channel knowledge, extending the perfect-CSI analyses in [48] and [50]. The paper is clearly structured and the algebraic chain leading to Eq. (44) follows the standard CV-QKD security template, which makes the claimed extension attractive. However, the exactness of Eq. (44) rests on at least two security-sensitive assumptions that are not proved—the diagonalization of the estimation-error covariance and the inclusion of the estimation noise in Eve's output state—and the numerical section uses a biased covariance estimator. Until these points are resolved, the quantitative SKR values and design conclusions should be treated as tentative. The paper does not provide independent Monte Carlo validation or reproducible code.","major_comments":[{"comment":"The per-mode summation in Eq. (44) is exact only if the channel-estimation noise n_RIS has independent components in the SVD basis. The covariance C_RIS computed in Eq. (19) is generically a full matrix: C_RIS = (2Va NT/(Vp Lp)) U†_RISLS C_tilde_n U_RISLS, and C_tilde_n includes 2V0 H_RIS H_RIS† plus detector noise. Because U_RISLS is the left singular-vector matrix of H_RISLS = H_RIS + Δ, not of H_RIS, there is no reason for U†_RISLS C_tilde_n U_RISLS to be diagonal; the off-diagonal entries of the cross term U†_RISLS Δ H_RIS V_RISLS in Eq. (16) are exactly cross-mode couplings. The paper defines σ²_RISi = 0.5 C_RIS(i,i) but never proves that the off-diagonal entries vanish or that the multimode Gaussian state separates into r independent single-mode channels. Thus Eq. (44) is an unproven approximation rather than a derived exact closed-form SKR. The authors should either prove the diagonalization under the specific channel/pilot model, or compute an exact multimode SKR from the full covariance matrix, or explicitly present Eq. (44) as an approximation with a justification and a bound on the error.","section":"§IV, Eq. (19); §V, Eq. (44)"},{"comment":"The same realization n_RIS appears in Eve's output mode in Eq. (18), and consequently σ²_RISi enters the Eve-side variance V_eoi = (1−β_i)V_a + β_i V_e + σ²_RISi in Eq. (28). In a collective Gaussian entangling-cloner attack, Eve's stored half of the two-mode squeezed vacuum is not affected by Bob's channel-estimation noise realization. An additive noise at Bob's measurement stage should appear in Bob's measured variance V_bi, but it should not appear in Eve's unconditional covariance matrix unless the attack model explicitly copies that noise to Eve. No physical justification is given for this common-noise assumption. This modeling choice reduces the Holevo information and therefore inflates the SKR. The authors should derive the joint state of Alice, Bob, and Eve from the actual attack and measurement model, recompute the conditional entropies, and, if a common-noise assumption is intended, state and defend it explicitly.","section":"§IV, Eqs. (17)–(18); §V, Eq. (28)"},{"comment":"The ML estimator in Eq. (21) is a biased residual sample covariance. Because the residual vectors are (I − P)Ñ, where P is the projector onto the row space of Ψ_p, the expectation satisfies E[Ĉ_ML] = ((Lp − NT)/Lp) C_tilde_n. Substituting Ĉ_ML into Eq. (22) therefore underestimates C_RIS by the factor (Lp − NT)/Lp. In the numerical section, where NT is as large as 256 and Lp can be comparable (Fig. 4), this bias materially changes the σ²_RISi values and hence the reported SKR curves. The paper should use an unbiased estimator, or explicitly correct for the bias and show that the numerical conclusions are unchanged.","section":"§IV, Eqs. (21)–(22)"},{"comment":"The text reports that the SKR degrades with increasing pilot power Vp and explains this by saying that higher pilot power implies a longer estimation duration. Vp is a power, not a duration, so the explanation is dimensionally inconsistent with the authors' own formula: in Eq. (22), Vp appears only in the denominator of the estimated noise covariance, so an increase in Vp should reduce σ²_RISi and, all else being equal, increase the SKR. The figure or the explanation should be reconciled with Eq. (22).","section":"§VI, Fig. 3"}],"minor_comments":[{"comment":"The caption reads 'NT = NR = {32, 12}'; this appears to be a typo, since the surrounding text specifies NT = NR = {32, 128}.","section":"§VI, Fig. 6 caption"},{"comment":"The LS solution and the DFT pilot design require Lp ≥ NT so that Ψ_p Ψ_p† is invertible; this constraint should be stated explicitly, and the plotted Lp ranges for large NT in Fig. 4 should be checked against it.","section":"§III, Eqs. (10)–(11)"},{"comment":"The section says the numerical results validate the analytical framework, but all curves are evaluations of Eq. (44) and no independent simulation is reported; the wording should be changed to 'illustrate the derived formula' or an actual Monte Carlo validation should be added.","section":"§VI"},{"comment":"The claim that this is 'the first work proposing the use of RISs for such quantum communication systems employing a practical channel estimation scheme' should be benchmarked against the existing literature, including the authors' own RIS-assisted CV-QKD study [50], to avoid an unsupported priority statement.","section":"§VI"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope and the engineering question is timely, but I would strongly recommend a separate security-model review before acceptance. The most consequential issue is the treatment of the channel-estimation noise in Eve's state (Eqs. (17)–(18) and (28)), which is not derived from the stated collective attack and currently acts in Eve's favor. The diagonalization gap in Eq. (44) is also substantial, and the biased ML covariance estimator affects the numerical claims. These are fixable, but they require a genuine revision rather than a light edit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper combines two strands that hadn't been put together before: RIS-assisted MIMO THz CV-QKD with LS channel estimation. Prior work had MIMO THz QKD with estimation error but no RIS ([16]), and RIS-assisted MIMO THz QKD with perfect CSI ([50]). This paper inserts the RIS channel and the estimation-error covariance into the standard CV-QKD SKR machinery, and the algebraic chain from (16) to (44) is internally consistent under its stated Gaussian assumptions. That is a real contribution: it gives a closed-form design tool, and the authors are upfront about the pilot overhead and the noise terms. The numerical study is thorough in terms of parameter sweeps.\n\nThe soft spots are in proportion. The biggest one is the step from (19) to (44). Expression (44) sums per-mode rates and uses only the diagonal entries σ²_RISi = 0.5 C_RIS(i,i). But C_RIS in (19) is generically a full matrix: U†_RISLS C_tilde_n U_RISLS is not diagonal, because C_tilde_n includes the term H_RIS H_RIS† and U_RISLS comes from the SVD of the estimated channel, not the true one. The off-diagonal entries couple the r modes, so the r parallel channels are not independent. The authors never prove that the covariance is diagonalized in this basis, and I don't see how it could be. As written, (44) is an unproven approximation, not a derived exact expression. This is load-bearing, because the whole claim of a closed-form SKR depends on it.\n\nSecond, the ML estimate in (21) is the biased residual sample covariance: its mean is ((Lp - NT)/Lp) C_tilde_n. They plug it into (22) without a degrees-of-freedom correction. For small Lp relative to NT, the numerical values will shift. This is fixable but genuine, and it means the computed SKR curves should be read with caution.\n\nThird, the per-mode independence assumption also affects the Holevo information: Eve's attack is modeled as per-mode with a common Ve, but if the modes are coupled, the optimal collective attack should account for the joint state. The authors don't justify that choice physically.\n\nThe central argument—that RIS helps at larger distances and that heterodyne beats homodyne at low pilot power—probably survives a proper fix, but the numbers as printed are not reliable. My recommendation: send it to peer review, but reviewers should demand either a proof of the diagonalization or an explicit acknowledgment that (44) is an approximation, plus the bias correction. The paper is for researchers working on THz CV-QKD and RIS-assisted quantum links; it's a useful incremental extension, not a breakthrough.","headline":"New combination of RIS, MIMO THz CV-QKD, and LS channel estimation, but Eq. (44) relies on an unproven diagonalization of the estimation-error covariance, so the closed-form claim is premature.","tokens_in":21057,"tokens_out":3031,"would_cite":false,"duration_ms":28048,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the secret key rate of an RIS-assisted MIMO CV-QKD link at THz frequencies, under least-squares channel estimation and a collective Gaussian attack, is given by a closed-form per-mode sum in Eq.","keywords":["continuous-variable quantum key distribution","reconfigurable intelligent surface","terahertz MIMO","least-squares channel estimation","secret key rate","collective Gaussian attack","homodyne detection","heterodyne detection"],"falsifier":"Compute the full multi-mode secret key rate from the joint Gaussian state using the entire estimated covariance $C_{RIS}$, including its off-diagonal entries, for a scenario where $C_{RIS}$ is strongly non-diagonal, and compare it with Eq. (44). If the difference is non-negligible, the closed-form sum is not the exact secret key rate. A Monte Carlo simulation of the LS pilot phase followed by key exchange would provide the same check.","tokens_in":19986,"feed_emoji":"🔐","tokens_out":8846,"duration_ms":78936,"temperature":0.7,"pith_summary":"The paper sets out to answer a practical question: how much secret key can an RIS-assisted MIMO terahertz link still generate when the channel is learned from pilots rather than assumed known. Its main claim is a closed-form secret key rate, Eq. (44), that takes the least-squares channel estimate, the extra noise it injects, detector noise, and a collective Gaussian eavesdropping attack into account. The formula splits the estimated MIMO channel into $r$ parallel modes, adds each mode's estimation-noise variance to the signal-to-noise ratio, and subtracts the corresponding Holevo information. A sympathetic reader would care because this gives a parameter-level handle on RIS element count, phase configuration, pilot power and length, and homodyne versus heterodyne detection, all in one expression. The paper's numerics use that expression to argue that the RIS keeps the key rate from falling at longer distances and that heterodyne detection is better at low pilot power.","feed_headline":"New formula sets secret-key rates for RIS-aided THz quantum links","feed_subtitle":"A single equation folds channel-estimation noise, RIS phase, and detector choice into the key rate, guiding system design.","key_machinery":"The argument is carried by the SVD of the least-squares channel estimate, $H_{RIS}^{LS} = U_{RISLS} \\Sigma_{RISLS} V_{RISLS}^{\\dagger}$, which splits the MIMO link into $r$ parallel Gaussian channels with gains $\\beta_i$. The estimation-error covariance $C_{RIS} = \\mathbb{E}[n_{RIS} n_{RIS}^{\\dagger}] = \\frac{2 V_a N_T}{V_p L_p} U_{RISLS}^{\\dagger} \\tilde{C}_n U_{RISLS}$ supplies the extra noise variance $\\sigma_{RISi}^2 = 0.5 C_{RIS}(i,i)$ used in each parallel channel; the paper estimates $\\tilde{C}_n$ by maximum likelihood from pilot residuals. The Holevo information terms are evaluated through symplectic eigenvalues $\\lambda_{i1},\\dots,\\lambda_{i4}$ of Eve's covariance matrices, giving closed-form entropy expressions. This machinery turns a full MIMO quantum protocol into a sum over independent scalar Gaussian channels.","core_discovery":"The central claim is that Eq. (44) is a closed-form expression for the secret key rate of the RIS-assisted MIMO CV-QKD system under reverse reconciliation: $SKR_{RR}^{MIMO} = (1 - L_p/T_c) \\sum_{i=1}^{r} \\left( \\eta \\frac{d_m}{2} \\log_2\\left(1 + \\frac{\\beta_i V_s}{\\beta_i V_o + (1-\\beta_i)V_e + \\sigma_d^2 + \\sigma_{RISi}^2}\\right) - h_o(\\lambda_{i1}) - h_o(\\lambda_{i2}) + h_o(\\lambda_{i3}) + h_o(\\lambda_{i4}) \\right)$, with $\\sigma_{RISi}^2 = 0.5 C_{RIS}(i,i)$ and $C_{RIS} = \\frac{2 V_a N_T}{V_p L_p} U_{RISLS}^{\\dagger} \\tilde{C}_n U_{RISLS}$, where $\\tilde{C}_n$ is the maximum-likelihood estimate of the pilot-noise covariance. The $\\beta_i$ are the squared singular values of the LS-estimated channel, the $\\lambda$'s are symplectic eigenvalues of Eve's covariance matrices, and $h_o(\\cdot)$ is the Holevo entropy function. This extends the earlier MIMO THz CV-QKD analysis to the practical case of imperfect channel state information, using the estimated channel for both beamforming and key extraction.","pith_inferences":["The per-mode sum in Eq. (44) uses only the diagonal of $C_{RIS}$; if the off-diagonal entries are large, the true multi-mode secret key rate would require the full covariance, so Eq. (44) may be an approximation. A numerical comparison against the full-covariance rate would settle this.","Eve in this model learns the estimated channel from the feedback link but does not appear to influence the estimation itself. An active pilot-corruption attack, where Eve injects noise during training, is a natural stress test of the formula.","Because pilot power and length enter both the estimation-noise covariance and the overhead factor $(1 - L_p/T_c)$, the paper's degradation curves imply an optimal pilot budget per coherence block, though the paper does not solve for it.","The same derivation structure should carry to MMSE or other estimators: replace the LS error covariance with the corresponding estimator's covariance and re-derive $C_{RIS}$; the secret key rate formula would keep its shape."],"forward_implications":["With Eq. (44), designers can compute the secret key rate of an RIS-assisted THz CV-QKD link directly from system parameters without running full protocol simulations, provided the per-mode independence assumption holds.","The number of RIS elements $K$ and their common phase $\\phi$ enter the key rate through the effective channel and through $C_{RIS}$; the numerics show the rate rises with $K$ and varies sinusoidally with $\\phi$, giving a tuning knob for the RIS.","Heterodyne detection is preferable at low pilot powers, while homodyne can win at longer distances for smaller MIMO arrays, yielding a detector-selection rule.","Longer pilot lengths and higher pilot powers do not simply help: both reduce the fraction of coherence time available for key exchange and change the noise covariance, so the secret key rate can decrease with $L_p$ and $V_p$.","The RIS prevents the secret key rate from decaying with distance in the simulated range, which is exactly the regime where THz links are otherwise loss-limited."],"supporting_citations":[{"why":"Supplies the MIMO CV-QKD decomposition into parallel Gaussian channels and the reverse-reconciliation per-mode secret key rate structure that Eq. (44) extends.","marker":"[48]"},{"why":"Provides the LS channel estimation with DFT pilots and the estimation-error noise covariance approach that the paper adapts to the RIS-assisted setup.","marker":"[16]"},{"why":"Gives the reverse-reconciliation secret key rate formula and the Holevo information expression used for each parallel channel.","marker":"[59]"},{"why":"Defines the Von Neumann entropy via symplectic eigenvalues and the $h_o$ function used in Eqs. (26)-(27).","marker":"[60]"},{"why":"Provides Lemma 4 used to evaluate the expectation $\\mathbb{E}[\\tilde{N}\\Psi_p^{\\dagger}\\Psi_p \\tilde{N}^{\\dagger}]$ in the derivation of $C_{RIS}$.","marker":"[58]"},{"why":"Supplies the collective entanglement attack model with TMSV/EPR modes and the beamforming structure that the paper adopts.","marker":"[57]"},{"why":"Presents the RIS-assisted MIMO THz CV-QKD system with perfect channel information that this paper extends to imperfect LS-based estimation.","marker":"[50]"}],"fun_headline_variants":["RIS-assisted THz QKD gets a closed-form key-rate equation","Single equation sets secret-key rates for THz MIMO QKD","Channel estimation folded into key-rate formula for THz QKD","Exact key-rate formula for RIS-aided MIMO CV-QKD","RIS boosts THz QKD secret-key rates with new formula"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The formula assumes the $r$ parallel SVD modes are independent and that each mode's channel-estimation noise is fully described by the single diagonal variance $\\sigma_{RISi}^2 = 0.5 C_{RIS}(i,i)$; if the covariance $C_{RIS}$ has significant off-diagonal correlations, summing per-mode rates does not give the exact total secret key rate.","fun_headline_variants_meta":{"raw":{"variants":["RIS-assisted THz QKD gets a closed-form key-rate equation","Single equation sets secret-key rates for THz MIMO QKD","Channel estimation folded into key-rate formula for THz QKD","Exact key-rate formula for RIS-aided MIMO CV-QKD","RIS boosts THz QKD secret-key rates with new formula"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000721,"raw_usage":{"total_tokens":3308,"prompt_tokens":1093,"completion_tokens":2215,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":709,"completion_tokens_details":{"reasoning_tokens":2125}},"tokens_in":709,"tokens_out":2215,"duration_ms":15614,"temperature":1.0,"reasoning_tokens":2125,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:30:42.953233+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the full multi-mode secret key rate from the joint Gaussian state using the entire estimated covariance $C_{RIS}$, including its off-diagonal entries, for a scenario where $C_{RIS}$ is strongly non-diagonal, and compare it with Eq. (44). If the difference is non-negligible, the closed-form sum is not the exact secret key rate. A Monte Carlo simulation of the LS pilot phase followed by key exchange would provide the same check.","supporting_citations":[{"cited_title":"MIM O terahertz quantum key distribution,","cited_arxiv_id":null,"evidence_quote":"Supplies the MIMO CV-QKD decomposition into parallel Gaussian channels and the reverse-reconciliation per-mode secret key rate structure that Eq. (44) extends."},{"cited_title":"Cha nnel estimation and secret key rate analysis of MIMO terahertz qu antum key distribution,","cited_arxiv_id":null,"evidence_quote":"Provides the LS channel estimation with DFT pilots and the estimation-error noise covariance approach that the paper adapts to the RIS-assisted setup."},{"cited_title":"Q uantum cryptography approaching the classical limit,","cited_arxiv_id":null,"evidence_quote":"Gives the reverse-reconciliation secret key rate formula and the Holevo information expression used for each parallel channel."},{"cited_title":"Continuou s-variable quantum key distribution using thermal states,","cited_arxiv_id":null,"evidence_quote":"Defines the Von Neumann entropy via symplectic eigenvalues and the $h_o$ function used in Eqs. (26)-(27)."},{"cited_title":"General capacity bounds for s patially correlated Rician MIMO channels,","cited_arxiv_id":null,"evidence_quote":"Provides Lemma 4 used to evaluate the expectation $\\mathbb{E}[\\tilde{N}\\Psi_p^{\\dagger}\\Psi_p \\tilde{N}^{\\dagger}]$ in the derivation of $C_{RIS}$."},{"cited_title":"MIMO terahertz quantum key distribution under restricted eavesdrop- ping,","cited_arxiv_id":null,"evidence_quote":"Supplies the collective entanglement attack model with TMSV/EPR modes and the beamforming structure that the paper adopts."},{"cited_title":"RIS-Assisted THz MIMO Wireless System in the Presence of Direct Link for CV-QKD with Limited Quantum Memory","cited_arxiv_id":"2410.16731","evidence_quote":"Presents the RIS-assisted MIMO THz CV-QKD system with perfect channel information that this paper extends to imperfect LS-based estimation."}],"review_version":1}