{"id":"21ec1f2a-7dc2-4eb3-86f7-73662da8c84b","arxiv_id":"2502.03475","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The paper derives a constrained transformation between muscle moment arms and quaternion-coordinate muscle length Jacobians, allowing muscle torque calculation in quaternion-based musculoskeletal simulations.","lead":"Engineers derived a mathematical mapping that lets muscle force effects be computed in quaternion-based models of joints, rather than Euler-angle models. This matters for shoulder simulations, where Euler angles can hit a singularity called gimbal lock at large ranges of motion.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The constrained quaternion mapping in Eqs. 14-15 is singular at 180° rotations (Q0=0), and the proposed sign-flip to force Q0>0 introduces a discontinuous coordinate jump, so the torque calculation is not globally valid for the large range of motion claimed.","rationale":"The reader's weakest_assumption correctly identifies the Q0 = 0 singularity as the most load-bearing limitation of the central claim. I agree because the mapping in Eqs. 14-15 is the core of the proposed torque calculation, and its invertibility depends on Q0 != 0. The paper's own Discussion acknowledges the singularity, so the reader's condition (demonstrate in a full quaternion dynamic simulation and provide artifacts) is appropriate. I do not find an additional mathematically fatal flaw: the derivation of E and the validation against Euler-angle moment arms are internally consistent, and the coordinate discontinuity at Q0 = 0 is a consequence of the same singularity rather than a separate error. The lack of a dynamic demonstration is a verification gap, but the mathematical soft spot is the singular chart, which the reader already flagged. Therefore the verdict should remain CONDITIONAL, with the condition that the authors either restrict the claims to Q0 != 0 or extend the method to switch charts (e.g., select the largest-magnitude quaternion component as dependent) to cover all of SO(3).","tokens_in":7031,"tokens_out":12819,"duration_ms":133514,"concrete_test":"Compute the 3x3 matrix E = 2GT from Eq. (12) for the unit quaternion family Q(eps) = (eps, 0, 0, sqrt(1 - eps^2)), representing rotations of angle pi - 2*eps about the z-axis. For eps in {1e-1, 1e-2, 1e-3, 1e-4}, evaluate the condition number of E; if it grows as 1/eps, the mapping in Eqs. 14-15 is ill-conditioned near 180° rotations. Additionally, in the double-pendulum model of Section 2.4, drive the second joint through a continuous 180° rotation (e.g., about the z-axis) with constant muscle forces, and evaluate tau = R_S F_m from Eq. 16 near Q0 = 0; if the torque diverges or jumps discontinuously, the method fails at the claimed target range of motion.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central construction (Section 2.3) uses the local parametrization Q0 = sqrt(1 - Q1^2 - Q2^2 - Q3^2) with Q0 > 0. Matrix T in Eq. 11 divides by Q0, so E = 2GT is only invertible when Q0 != 0. At Q0 = 0, which corresponds to any 180° rotation, Eqs. 14-15 are undefined and R_S = (E^T)^-1 R*_Q blows up unless R*_Q lies in a special subspace. The paper acknowledges this singularity in the Discussion but does not provide a workaround. The suggested remedy of multiplying q by -1 when Q0 < 0 fails at Q0 = 0 (both q and -q have Q0 = 0) and, more importantly, causes Q* to jump from q_vec to -q_vec when a continuous motion crosses the equator Q0 = 0. Consequently, the muscle length expressed as a function of Q* is discontinuous at the equator, which would break polynomial approximations of muscle paths over a full range of motion. Since the paper explicitly targets shoulder models with large range of motion, and a composite rotation of exactly 180° is physically possible in the shoulder, the central claim of singularity-free muscle torque computation is not established for the intended application domain.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a mathematical framework for computing muscle torques in quaternion-based musculoskeletal models. It introduces a constrained mapping between the muscle length Jacobian in spatial coordinates and the Jacobian in a reduced quaternion parametrization, derived by imposing the unit-quaternion constraint and selecting Q0 = sqrt(1-Q1^2-Q2^2-Q3^2) with Q0>0. The mapping is validated on a three-dimensional double-pendulum model with six straight-line muscles, showing exact agreement with analytic YZY Euler-angle moment arms (Fig. 3), and is applied to polynomial approximation of the serratus anterior muscle length in a shoulder model (Section 3.2).","tokens_in":7299,"tokens_out":8524,"duration_ms":79937,"significance":"If valid, the method provides a practical route to muscle torque computation in quaternion-based dynamic simulations, potentially avoiding Euler-angle gimbal lock for joints with large range of motion such as the shoulder. The paper includes a strong validation step (machine-checkable exact agreement in Fig. 3) and demonstrates a relevant application to a realistic shoulder muscle path approximation. However, the singularity at 180-degree rotations is acknowledged but not resolved, and the absence of a full dynamic simulation limits the strength of the claims about robust torque estimation for the intended application domain.","major_comments":[{"comment":"The constrained mapping R*_Q = E^T R_S and R_S = (E^T)^{-1} R*_Q is singular when Q0=0, i.e., for any 180-degree rotation. The sign-flip remedy proposed in the Discussion (multiplying the quaternion by -1 when Q0<0) fails exactly at Q0=0 and introduces a discontinuous jump in the reduced coordinates Q* when a continuous motion crosses the Q0=0 hypersurface. Because the paper explicitly targets shoulder models with large range of motion, and a 180-degree rotation is physically possible, the method is not globally valid for the claimed application. The authors should either provide a global chart (e.g., switching the dependent quaternion component) or explicitly restrict the claims and characterize the region of validity.","section":"Section 2.3, Eq. (11) and Discussion"},{"comment":"The derivation is presented for a single quaternion, but the validation and application involve multiple joints (double pendulum with two 3D joints; shoulder with sternoclavicular and acromioclavicular joints). The paper does not define how the matrix E and the quaternion Jacobian R*_Q are assembled for multiple quaternions. While the empirical agreement in Fig. 3 suggests a correct block-diagonal extension, the formalism should be stated explicitly for the multi-quaternion case so that the reader can reproduce the mapping without inference.","section":"Section 2.3, Eqs. (13)-(15)"},{"comment":"The polynomial approximation of the muscle length is performed after mapping to a reduced quaternion representation requiring Q0>0. If the training data, which includes added random noise, contains configurations with Q0<0, the sign-flip makes Q* discontinuous at Q0=0, violating the differentiability assumptions of the polynomial fitting algorithm. The paper should report whether any training configuration had Q0 near zero, and specify how the sign-flip is handled in the polynomial approximation; otherwise, the comparison of RMSE for the quaternion-based approximation is not meaningful for motions that cross the singularity.","section":"Section 3.2 and Figure 4"},{"comment":"The validation compares muscle length Jacobians, not actual torques in a dynamic simulation. The abstract and introduction suggest that the method estimates muscle torques for dynamic models, and the Discussion claims 'robust estimation of muscle torques in dynamic motion'. A forward dynamics simulation using the quaternion-based torque (e.g., comparing the resulting motion with the Euler-angle model) would substantiate this claim more directly than a static Jacobian comparison.","section":"Section 3.1 and Discussion"}],"minor_comments":[{"comment":"The phrase 'transformation of muscle moment arms obtained from muscle paths based on Euler angles into quaternions description' is grammatically awkward; consider rephrasing to 'transformation of muscle moment arms from Euler-angle-based muscle paths into the quaternion representation'.","section":"Abstract"},{"comment":"The notation RQ, R*_Q, RS, and RE is easy to confuse. Please provide a table or explicit definitions of each Jacobian's dimensions and reference frame (e.g., RQ is a 4-vector, R*_Q is a 3-vector, RS is a 3-vector in body coordinates, RE is a 3-vector in Euler coordinates).","section":"Section 2.2 and 2.3"},{"comment":"The caption states that the mapped and analytic curves are 'identical', but the plot likely shows overlapping lines. Consider adding a second subplot with the pointwise difference to visually confirm the agreement, and state the numerical maximum error in the text.","section":"Figure 3"},{"comment":"The double-pendulum model is described as '3D' but the number of degrees of freedom (two joints with three rotational DOF each) is not explicitly stated. Clarify this to avoid ambiguity about the size of the Jacobians.","section":"Section 2.4"},{"comment":"The stopping criterion 'the change of the RMSE of an approximation and OpenSim generated values was less than 4%' is not fully specified. Please state the exact error measure, the direction of the change, and how the 4% threshold was chosen.","section":"Section 2.5"},{"comment":"The description of adding 'random noise to the angles' is not reproducible. Specify the noise distribution, standard deviation, and whether the noise was added to the measured angles or to the corresponding quaternions.","section":"Section 2.5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript presents a solid derivation for the local quaternion-to-spatial torque mapping, with a strong validation in Fig. 3. The main concern is the unresolved singularity at 180-degree rotations, which directly affects the paper's central motivation (large-range-of-motion shoulder simulations). The absence of a full dynamic simulation also makes the 'robust torque estimation' claim somewhat overstated. These issues are addressable, so I recommend major revision rather than rejection. The paper's scope is more appropriate for a biomechanics journal, but physics.med-ph is acceptable if the presentation is strengthened."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is worth a referee look. It derives a constrained mapping from spatial muscle length Jacobians to quaternion coordinates (Eqs. 14-15), resolving the 4-to-3 ambiguity that arises if you naively invert the quaternion-to-spatial velocity map. That specific mapping is not in Graf, van den Bogert, or Sobinov, and the authors validate it against an analytic YZY Euler-angle moment arm in a double pendulum; Figure 3 shows exact agreement. That is a genuine, reproducible-in-principle contribution.\n\nThe application to polynomial muscle-path approximation is sensible: quaternion polynomial with 13 terms vs. 15 for Euler, with comparable RMSE. The authors are transparent that approximation differences depend on algorithm configuration. Citation pattern is clean; self-citation is confined to the application stage and is appropriate.\n\nThe soft spots are real but in proportion. The 180-degree singularity (Q0=0) is acknowledged in the Discussion, but the proposed remedy—multiply q by -1 if Q0<0—fails at Q0=0 and, more importantly, makes the vector part jump discontinuously when a continuous motion crosses the equator. So the method is locally correct but not globally valid for the full range of motion they advertise for the shoulder. They also don't ship code or data, and the validation is a Jacobian comparison after a forward simulation with constant random forces, not a full muscle-driven dynamic simulation. Adding error bars or repeated trials to the RMSE comparison would be nice but is minor.\n\nNone of this sinks the central derivation. For any trajectory that avoids the equator, the math is sound. The gap is in the scope claim, not the core mapping.\n\nWho is this for: biomechanists doing musculoskeletal simulation or muscle-path polynomial approximation, especially anyone wanting to avoid Euler-angle gimbal lock in shoulder models. It deserves peer review: the method is novel, the algebra is defensible, and a good referee can push for a resolution of the singularity or at least a bounded-pose applicability statement.\n\nRecommendation: send it out, with a request for code/data and a clear treatment of the Q0=0 case.","headline":"A useful, honest methods paper with a locally correct mapping; the 180-degree singularity is acknowledged but the proposed workaround is incomplete, so the large-range-of-motion claim outruns the math.","tokens_in":7882,"tokens_out":1856,"would_cite":true,"duration_ms":19900,"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":"Muscle torque can be computed in quaternion coordinates through a constrained Jacobian map, avoiding the Euler-angle gimbal lock.","keywords":["quaternions","muscle torque","muscle moment arm","Jacobian","gimbal lock","shoulder kinematics","musculoskeletal dynamics","polynomial approximation"],"falsifier":"Let the double pendulum pass through an orientation where a joint is rotated exactly 180 degrees, and evaluate Eq. 15 for a muscle spanning that joint: the torque should diverge or become undefined at that instant, while the unconstrained quaternion Jacobian $R_Q = 2G^T R_S$ remains finite. A finite torque at $Q_0=0$ would contradict the paper's singularity analysis.","tokens_in":6816,"feed_emoji":"🦴","tokens_out":9707,"duration_ms":87174,"temperature":0.7,"pith_summary":"This paper establishes a way to compute the torque a muscle produces at a joint when the joint's rotation is described by quaternions (a four-number rotation representation) rather than by Euler angles. The key step is to express muscle length as a function of the quaternion components, differentiate it, and map the resulting Jacobian back to spatial coordinates using the unit-quaternion constraint. The authors show the mapping is unique only when the constraint is imposed, and they verify on a three-dimensional double pendulum that it reproduces the standard moment-arm result. They also fit a quaternion-based polynomial to a shoulder muscle path and show it matches moment arms exported from standard musculoskeletal software to about 10 millimeters RMSE. If the method holds, dynamic simulations of large-range joints such as the shoulder no longer need to switch rotation sequences to avoid gimbal lock.","feed_headline":"Quaternion mapping computes muscle torque without gimbal lock","feed_subtitle":"A constrained Jacobian turns muscle length changes into joint torques in any orientation—except an exact 180-degree flip.","key_machinery":"The load-bearing object is the constrained quaternion muscle-length Jacobian $R^*_Q = \\partial l_m/\\partial Q^*_i$, where $Q^* = (Q_1,Q_2,Q_3)$ and $Q_0$ is eliminated through the unit-norm constraint. The matrix $E = 2 G T$, formed from the quaternion angular-velocity matrix $G$ and the $4\\times3$ parametrization matrix $T$ from the constraint, converts between this Jacobian and the spatial Jacobian $R_S$ through $R^*_Q = E^T R_S$ and $R_S = (E^T)^{-1} R^*_Q$. Muscle torque is then $\\tau = R_S F_m$. The machinery works because constraining the quaternion removes the redundant degree of freedom, making the mapping from spatial to quaternion coordinates unique and invertible, at the price of a singularity when $Q_0=0$.","core_discovery":"The paper's central claim is that in a quaternion-based musculoskeletal model, muscle torque can be obtained from the muscle-length Jacobian through the constrained mapping $R^*_Q = E^T R_S$ and $R_S = (E^T)^{-1} R^*_Q$, with torque $\\tau = R_S F_m$. Here $R^*_Q$ is the derivative of muscle length with respect to the three independent quaternion components after enforcing $Q_0^2+Q_1^2+Q_2^2+Q_3^2=1$ and taking $Q_0 = \\sqrt{1-Q_1^2-Q_2^2-Q_3^2}$. The matrix $E$ encodes the relation between quaternion rates and spatial angular velocity; the constraint makes $E$ square and invertible, whereas the naive unconstrained map from spatial to quaternion Jacobian is ambiguous. The authors demonstrate in a double-pendulum simulation that this constrained route produces moment arms identical to the analytical Euler-angle calculation, and they apply it to approximate a shoulder muscle path. Their stated benefit is a dynamical muscle representation that is free of Euler-angle gimbal lock and suitable for joints with large ranges of motion.","pith_inferences":["The paper leaves implicit that its singularity at $Q_0=0$ can be handled by switching to the other quaternion chart or to the unconstrained four-component Jacobian near 180-degree rotations; the method as stated is local, not global.","Because the constrained mapping makes quaternion Jacobians and Euler moment arms algebraically interchangeable (Eqs. 5, 6, 14, 15), muscle paths defined in existing Euler-angle models could be re-expressed in quaternion form and run in quaternion-based solvers without re-deriving muscle geometry.","The shoulder example's comparable accuracy with fewer polynomial terms hints that quaternion coordinates may be a smoother domain for muscle-path regression, but the paper notes the comparison depends on the fitting algorithm; testing multiple muscles and motions would tell whether the hint is general.","An optimal-control extension follows naturally: a quaternion state avoids the discontinuous rotation-sequence switches that Euler coordinates force, which could make all feasible joint trajectories reachable in trajectory optimization."],"forward_implications":["Quaternion-based forward dynamics can include muscle actuation directly, without converting muscle paths to Euler angles at each step.","Muscle moment arms exported from Euler-angle-based models can be transformed into quaternion Jacobians (Eqs. 6 and 14) and used for approximation or control.","Polynomial muscle-path approximations in quaternion coordinates are feasible: the shoulder example required 13 polynomial terms versus 15 in Euler coordinates, with 10.01 mm versus 10.18 mm RMSE in moment arms.","Simulations of joints with large ranges of motion, such as the shoulder, can avoid changing rotation sequences to dodge gimbal lock.","The constrained form reduces the number of derivatives from four to three, lowering computational cost per muscle evaluation."],"supporting_citations":[{"why":"Supplies the matrix G that connects quaternion time derivatives to spatial angular velocity, the foundation for the E map and the Jacobian relations.","marker":"[Graf, 2008]"},{"why":"Defines the analytical tendon-excursion and moment-arm relation used as the Euler-angle baseline that the quaternion result must match.","marker":"[An et al., 1983]"},{"why":"Provides the polynomial approximation algorithm used to fit muscle length and moment arms in both Euler and quaternion forms.","marker":"[van den Bogert et al., 2011]"},{"why":"Supplies the shoulder model whose serratus anterior muscle path is used as the application example.","marker":"[Chadwick et al., 2014]"},{"why":"Documents the musculoskeletal modeling software used to export muscle lengths and moment arms for the approximation test.","marker":"[Delp et al., 2007]"}],"fun_headline_variants":["Quaternion muscle action without gimbal lock","Constrained quaternion Jacobian yields muscle torques","Muscle torques from quaternion model, no gimbal lock","Quaternion method prevents gimbal lock in muscle torque","New quaternion muscle torque model for complex joints"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All conversions rely on the scalar quaternion component $Q_0$ being positive and nonzero; at any 180-degree joint rotation $Q_0=0$ for both equivalent quaternions, the matrix $E$ cannot be inverted, and the muscle-torque calculation becomes singular.","fun_headline_variants_meta":{"raw":{"variants":["Quaternion muscle action without gimbal lock","Constrained quaternion Jacobian yields muscle torques","Muscle torques from quaternion model, no gimbal lock","Quaternion method prevents gimbal lock in muscle torque","New quaternion muscle torque model for complex joints"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000198,"raw_usage":{"total_tokens":1364,"prompt_tokens":936,"completion_tokens":428,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":348}},"tokens_in":552,"tokens_out":428,"duration_ms":4518,"temperature":1.0,"reasoning_tokens":348,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:10:58.402552+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Let the double pendulum pass through an orientation where a joint is rotated exactly 180 degrees, and evaluate Eq. 15 for a muscle spanning that joint: the torque should diverge or become undefined at that instant, while the unconstrained quaternion Jacobian $R_Q = 2G^T R_S$ remains finite. A finite torque at $Q_0=0$ would contradict the paper's singularity analysis.","supporting_citations":[],"review_version":1}