Fisher information as a generalized measure of coherence in classical and quantum optics

We show that metrological resolution in the detection of small phase shifts provides a suitable generalization of the degrees of coherence and polarization. Resolution is estimated via Fisher information. Besides the standard two-beam Gaussian case, this approach provides also good results for multiple field components and nonGaussian statistics. This works equally well in quantum and classical optics. © 2012 Optical Society of America OCIS codes: (030.1640) Coherence; (260.5430) Polarization; (270.5290) Photon statistics; (260.2110) Electromagnetic optics. References and links 1. L. Mandel and E. Wolf, Optical Coherence and Quantum Optics (Cambridge U. Press, 1995). 2. J. W. Goodman, Statistical Optics (John Wiley and Sons Inc., 1985). 3. B. Karczewski, “Degree of coherence of the electromagnetic field,” Phys. Lett. 5, 191–192 (1963). 4. H. M. Ozaktas, S. Yüksel, and M. A. Kutay, “Linear algebraic theory of partial coherence: discrete fields and measures of partial coherence,” J. Opt. Soc. Am. A 19, 1563–1571 (2002). 5. E. Wolf, “Unified theory of coherence and polarization of random electromagnetic beams,” Phys. Lett. A 312, 263–267 (2003). 6. J. Tervo, T. Setälä, and A. T. Friberg, “Degree of coherence for electromagnetic fields” Opt. Express 11, 1137– 1143 (2003). 7. T. Setälä, J. Tervo, and A. T. Friberg, “Complete electromagnetic coherence in the space-frequency domain,” Opt. Lett. 29, 328–330 (2004). 8. E. Wolf, “Comment on Complete electromagnetic coherence in the space-frequency domain,” Opt. Lett. 29, 1712–1712 (2004). 9. T. Setälä, J. Tervo, and A. T. Friberg, “Reply to comment on Complete electromagnetic coherence in the spacefrequency domain,” Opt. Lett. 29, 1713–1714 (2004). 10. P. Réfrégier and F. Goudail, “Invariant degrees of coherence of partially polarized light,” Opt. Express 13, 6051– 6060 (2005). 11. F. Gori, M. Santarsiero, and R. Borghi, “Maximizing Young’s fringe visibility through reversible optical transformations,” Opt. Lett. 32, 588–590 (2007). 12. R. Martı́nez-Herrero and P.M. Mejı́as, “Maximum visibility under unitary transformations in two-pinhole interference for electromagnetic fields,” Opt. Lett. 32, 1471–1473 (2007). 13. A. Luis, “Degree of coherence for vectorial electromagnetic fields as the distance between correlation matrices,” J. Opt. Soc. Am. A 24, 1063–1068 (2007). 14. A. Luis, “Maximum visibility in interferometers illuminated by vectorial waves,” Opt. Lett. 32, 2191–2193 (2007). 15. J. J. Gil, “Polarimetric characterization of light and media,” Eur. Phys. J. Appl. Phys. 40, 1–47 (2007). 16. I. San José and J. J. Gil, “Invariant indices of polarimetric purity. Generalized indices of purity for nxn covariance matrices,” arXiv:0807.2171v1 [physics.optics]. #172647 $15.00 USD Received 16 Jul 2012; revised 7 Oct 2012; accepted 7 Oct 2012; published 12 Oct 2012 (C) 2012 OSA 22 October 2012 / Vol. 20, No. 22/ OPTICS EXPRESS 24686 17. P. Réfrégier, “Mean-square coherent light,” Opt. Lett. 33, 1551–1553 (2008). 18. A. Luis, ”Quantum-classical correspondence for visibility, coherence, and relative phase for multidimensional systems,” Phys. Rev. A 78, 025802 (2008). 19. P. Réfrégier and A. Luis, “Irreversible effects of random unitary transformations on coherence properties of partially polarized electromagnetic fields,” J. Opt. Soc. Am. A 25, 2749–2757 (2008). 20. R. Martı́nez-Herrero and P.M. Mejı́as, “Maximizing Youngs fringe visibility under unitary transformations for mean-square coherent light,” Opt. Express 17, 603–610 (2009). 21. A. Luis, “Coherence and visibility for vectorial light,” J. Opt. Soc. A 27, 1764–1769 (2010). 22. A. Luis, “Coherence versus interferometric resolution,” Phys. Rev. A 81, 065802 (2010). 23. A. Luis, “An overview of coherence and polarization properties for multicomponent electromagnetic waves,” in Advances in Information Optics and Photonics, International Commission for Optics, vol. VI, A. T. Friberg and R. Dändliker, eds. (SPIE, 2009) pp. 171–188. 24. J. Tervo, T. Setälä, and A. T. Friberg, “Phase correlations and optical coherence,” Opt. Lett. 37, 151–153 (2012). 25. A. Luis, “Degree of polarization in quantum optics,” Phys. Rev. A 66, 013806 (2002). 26. A. Luis, “Degree of polarization of type-II unpolarized light,” Phys. Rev. A 75, 053806 (2007) 27. A. Luis, “Polarization distributions and degree of polarization for quantum Gaussian light fields,” Opt. Commun. 273, 173–181 (2007). 28. A. Luis, “Ensemble approach to coherence between two scalar harmonic light vibrations and the phase difference,” Phys. Rev. A 79, 053855 (2009). 29. A. Picozzi, “Entropy and degree of polarization for nonlinear optical waves,” Opt. Lett. 29, 1653–1655 (2004). 30. P. Réfrégier, “Polarization degree of optical waves with non-Gaussian probability density functions: Kullback relative entropy-based approach,” Opt. Lett. 30, 1090–1092 (2005). 31. P. Réfrégier and F. Goudail, “Kullback relative entropy and characterization of partially polarized optical waves,” J. Opt. Soc. A 23, 671–678 (2006). 32. A. Rivas and A. Luis, “Characterization of quantum angular-momentum fluctuations via principal components,” Phys. Rev. A 77, 022105 (2008). 33. A. Luis, “Quantum-limited metrology with nonlinear detection schemes,” SPIE Reviews 1, 018006 (2010). 34. H. Cramér, Mathematical Methods of Statistics (Asia Publishing House, 1962). 35. S. L. Braunstein and C. M. Caves, “Statistical distance and the geometry of quantum states,” Phys. Rev. Lett. 72, 3439–3443 (1994). 36. Exploratory Data Analysis Using Fisher Information, B. R. Frieden, ed. (Springer-Verlag, 2007). 37. B. R. Frieden, Physics from Fisher information: A Unification, (Cambridge U. Press, 1999). 38. A. Luis and L. L. Sánchez-Soto, “A quantum description of the beam splitter,” Quantum Semiclass. Opt. 7, 153–160 (1995). 39. T. M. Cover and J. A. Thomas, Elements of Information Theory, (Wiley Interscience, 1991). 40. A. D. C. Nascimento, R. J. Cintra, and A. C. Frery, “Hypothesis testing in speckled data with stochastic distances,” IEEE Trans. Geos. Remot. Sens. 48, 373–385 (2010). 41. F. Goudail, P. Réfrégier, and G. Delyon, “Bhattacharyya distance as a contrast parameter for statistical processing of noisy optical images,” J. Opt. Soc. Am. A 21, 1231–1240 (2004). 42. G. Björk, J. Soderholm, L. L. Sánchez-Soto, A. B. Klimov, I. Ghiu, P. Marian, and T. A. Marian, “Quantum degrees of polarization,” Opt. Commun. 283, 4440–4447 (2010). 43. P. Réfrégier, “Mutual information-based degrees of coherence of partially polarized light with Gaussian fluctuations,” Opt. Lett. 30, 3117–3119 (2005). 44. P. Réfrégier and A. Roueff, “Visibility interference fringes optimization on a single beam in the case of partially polarized and partially coherent light,” Opt. Lett. 32, 1366–1368 (2007). 45. I. Afek, O. Ambar, and Y. Silberberg, “High-NOON states by mixing quantum and classical light,” Science 328, 879–881 (2010). 46. H.-W. Lee, “Theory and application of the quantum phase-space distribution functions,” Phys. Rep. 259, 147–211 (1995). 47. C. M. Caves, “Quantum-mechanical noise in an interferometer,” Phys. Rev. D 23, 1693–1708 (1981). 48. A. Rivas and A. Luis, “Precision quantum metrology and nonclassicality in linear and nonlinear detection schemes,” Phys. Rev. Lett. 105, 010403 (2010). 49. Z. Y. Ou, “Fundamental quantum limit in precision phase measurement,” Phys. Rev. A 55, 2598–2609 (1997). 50. A. Luis and L. L. Sánchez-Soto, “Quantum phase difference, phase measurements and Stokes operators,” in Progress in Optics, vol. 41, E. Wolf, ed. (Elsevier, Amsterdam, 2000), pp. 421–482. 51. A. Luis, “Visibility for anharmonic fringes,” J. Phys. A: Math. Gen. 35, 8805–8815 (2002). 52. A. Sehat, J. Söderholm, G. Björk, P. Espinoza, A. B. Klimov, and L. L. Sánchez-Soto,“Quantum polarization properties of two-mode energy eigenstates,” Phys. Rev. A 71, 033818 (2005). 53. R. Simon, E. C. G. Sudarshan, and N. Mukunda, “Gaussian-Wigner distributions in quantum mechanics and optics,” Phys. Rev. A 36, 3868–3880 (1987). 54. A. Luis, “Quantum mechanics as a geometric phase: phase-space interferometers,” J. Phys. A 34, 7677–7684 (2001). #172647 $15.00 USD Received 16 Jul 2012; revised 7 Oct 2012; accepted 7 Oct 2012; published 12 Oct 2012 (C) 2012 OSA 22 October 2012 / Vol. 20, No. 22/ OPTICS EXPRESS 24687 55. A. Luis, “Classical mechanics and the propagation of the discontinuities of the quantum wave function,” Phys. Rev. A 67, 024102 (2003). 56. B. N. Simon, S. Simon, F. Gori, M. Santarsiero, R. Borghi, N. Mukunda, and R. Simon,“Nonquantum entanglement resolves a basic issue in polarization optics,” Phys. Rev. Lett. 104, 023901 (2010).


Introduction
Interference and polarization are leading manifestations of coherence [1,2].In spite of being a very basic topic the complete relationship between coherence, interference and polarization for arbitrary fields is still an open question and is currently the subject of very active investigation .
For two electromagnetic fields E 1,2 we may say that interference reflects coherence between fields in the same vibration mode, while polarization displays coherence between fields in orthogonal vibration modes.Interference and polarization are easily interchangeable by means of simple energy-preserving transformations, such as phase plates, so that they are not independent field properties.
More specifically, coherence manifests in interference through the modulation of the intensity of the superposition E 1 e iφ 1 + E 2 e iφ 2 2 , where φ j are the phases acquired by the waves within the interferometer.This modulation can be conveniently assessed by the fringe visibility where μ is the complex second-order degree of coherence, and the angle brackets represent ensemble averages.In the polarization context, coherence manifests in the purity of the polarization state assessed through the degree of polarization where I M,m are the eigenvalues of the correlation matrix Γ.A key point is that P is invariant under deterministic unitary transformations E → E = UE, Γ(E) → Γ(E ) = UΓ(E)U † , being U 2 × 2 unitary matrices, so we have P(E ) = P(E) while V (E ) = V (E) and |μ(E )| = |μ(E)|.More specifically, it holds that V and P satisfy the equality [21] This, along with Eq. ( 1), implies that P is the maximum degree of coherence that can be obtained for a given field state after any deterministic unitary transformation The standard definitions of the degrees of coherence and polarization in Eqs. ( 1) and (2) hold provided that three conditions are satisfied: i) we are restricted to two scalar electromagnetic fields, ii) we perform statistical evaluations of second order in the amplitudes, and iii) the fields obey Gaussian statistics.Beyond these three restrictions the standard approach may be undetermined or lack usefulness when applied to more general situations involving either more than two electric fields or nonGaussian statistics, as it is the case of the problems addressed in Refs. .A particularly clear example of these limitations is provided by quantum field states reaching maximum resolution in phase-shift detection, since they present vanishing degrees of polarization and coherence μ = P = 0 in spite of their interferometric usefulness [22,32].
In this work we look for a more general approach to coherence that can be applied to any situation with multi-component fields and nonGaussian statistics, including naturally Eqs. ( 1) and (2) in the particular case of just two electric fields in a thermal-chaotic light state.To this end we focus on the practical application of interference and polarization to the detection of small phase shifts, as the cornerstone of the metrological applications of optics.
The key point is the idea of coherence as equivalent to good interference.This idea emerges after Eq. ( 1), where coherence is the maximum fringe visibility, as well as after Eq. ( 4), where visibility reaches its maximum provided that the unitary transformation U is properly chosen.This suggests that the degrees of coherence and polarization indicate the best interference that can be achieved with a suitable preparation of the interference for a given input field state.
Therefore, we will identify the generalized amount of coherence χ of a given field state with the maximum phase-shift resolution after a suitable optimization of the detection arrangement, so that the phase-shift measurement exploits all the interfering capabilities of the field state.Note that this approach can be equally well applied to quantum and classical fields.
The resolution can be properly assessed by means of the Fisher information entering in the Cramér-Rao lower bound for the uncertainty of phase-shift estimators [34,35].It is worth pointing out that the Fisher information has been extensively used to provide original analyses and interpretations of many different problems in science, from the formation of basic physical laws, to biology and financial economy, for example [36,37].

Phase shifts and phase-shift detection
By phase shifts we will understand linear, energy-preserving, unitary transformations of the complex amplitudes E of an arbitrary number k of complex electric fields, E → U φ E, where U φ is a k × k unitary matrix.In general terms we may express U φ as U φ = exp(iφ G), where φ is the phase shift, and G an Hermitian k × k matrix G † = G.
In quantum optics the amplitudes E become complex-amplitude operators a, with [a i , a † j ] = δ i, j and [a i , a j ] = 0.The phase-shift transformation reads a → U φ a = T † aT , with T = exp(iφ Ĝ) and Ĝ = a † Ga [38].
Both in classical and quantum optics the objective is to infer the value of the unknown phase shift φ from measurements performed on the transformed electric fields U φ E. Let us call P(θ |φ ) the statistics of the measurement outcomes θ when the true value of the phase shift is φ .From a Bayesian perspective P(θ |φ ) defines a posteriori probability distribution P( φ |θ ) for the inferred value of the signal φ depending on the N outcomes θ obtained after N repetitions of the measurement.This is of the form where Φ( φ ) is the prior distribution for the signal.

Fisher information
A convenient assessment of the uncertainty Δφ on the value of φ after the measurement is the Fisher information F entering in the Cramér-Rao lower bound It is worth noting that the Cramér-Rao lower bound is not necessarily reachable in every situation, i. e., there is no guarantee that an efficient estimator can be found that reaches the minimum variance 1/(NF), specially for small N.However, for increasing N, the maximum likelihood estimator [the φ value that maximizes P( φ |θ )] is unbiased and asymptotically efficient, so that Δ φ → 1/ √ NF when N → ∞ [34].After a large enough number of repetitions, 1/(NF) becomes the width of P( φ |φ ) as a function of φ since P( φ |φ ) ∝ exp[−NF( φ − φ ) 2 /2], which is consistent with the above mentioned asymptotic efficiency.
Besides, the Fisher information is also the approximate form of several distances, or divergences [39,40], between the statistics associated with two close enough phase shifts φ and φ + δ φ, such as the Hellinger distance, the Kullback-Liebler divergence or the Rényi-Chernoff distance that becomes the Bhattacharyya distance for s = 1/2.All them have been already applied for the assessment of coherence and polarization in different contexts [29-31, 41, 42].

Generalized measure of coherence
In general, F depends on the light state, on the generator G, on the measurement performed, and also on φ .Following the idea of coherence as optimum visibility, we will consider that the measurement performed extracts from the transformed fields U φ E the maximum information possible about φ .Throughout we assume φ → 0 since precision metrology is always interested in the detection of small phase shifts, which can be, without loss of generality, brought back around a zero phase-shift value.Nevertheless, there would be no fundamental difficulty to apply this approach beyond precision metrology to large phase shifts, maybe with the drawback of lacking simple expressions and obtaining different conclusions depending on the particular values of φ considered.The larger F the lesser the uncertainty and the larger the resolution so we may say that larger F means larger coherence.Since F is not bounded from above, ∞ ≥ F ≥ 0, it might be convenient to normalize it, so we may introduced a generalized assessment of coherence χ comprised between 0 and 1 in the form The factor 4tr G 2 is introduced so that χ becomes exactly the degree of coherence for two electric fields in a classical thermal-chaotic light state [see Eq. ( 21) below].The dependence on G means essentially that the usefulness of interference fringes depends on whether the phaseshift transformation exploits all the interfering capabilities of the light state.A coherence assessment depending just on the field state can be obtained as the maximum of χ when G is varied.This might be regarded as the generalization of the degree of polarization as the ultimate measure of coherence in accordance with the standard result for two electric fields and Gaussian statistics recalled in the introduction.
Quantum and classical statistics differ in very fundamental issues, such as the uncertainty relation for example.Thus, at this point the analysis must be split into classical and quantum channels, although we will see that they largely run parallel.

Classical sector
In classical optics every measurement statistics P(θ |φ ) can be derived as a marginal distribution of the probability distribution for the transformed amplitudes W (E|φ ) = W (U † φ E), where W (E) is the probability distribution for the complex amplitudes in the field state before the signal-dependent transformation U φ .More specifically, taking into account that any measured observable is a function of the complex amplitudes θ (E) we have Formally we can consider a suitable change of variables including θ within the new set of arguments of the transformed distribution W , so that where ϑ represents the rest of variables required for a complete specification of the system.After Eqs. ( 11) and ( 12) we have that any measurement statistics P(θ |φ ) is extracted as a marginal distribution from the complete distribution W (U † φ E).Therefore, W (U † φ E) contains no less information than any P(θ |φ ), so that (see Appendix A) This is to say that F C is the maximum Fisher information reachable by any measurement.Therefore, provides a suitable estimation of the degree of coherence, understood as metrological usefulness under phase shifts generated by G. Let us consider some meaningful examples.

Thermal-chaotic light
A fundamental example is provided by the coherence between two electric fields in a thermalchaotic light state.This allows us to check the compatibility of this approach with the standard definition.Thermal-chaotic light is fully characterized by second-order moments of the field amplitudes as where M = Γ −1 and Γ i, j = E i E * j , with Γ j, j = |E j | 2 = 0 since otherwise we shall accordingly reduce the number k of electric fields.A simple calculus leads to (see Appendix B) The dependence on G reflects the fact that for a given field state the usefulness of the interference depends on the particular interferometric arrangement considered and whether it exploits the capabilities of the incoming light state.To exploit all the coherence conveyed by the field state we have to look for the maximum of F C by varying G.To this end we particularize the above expressions to the field basis where M is diagonal, leading to where only the terms i = j contribute, P i, j is the two-field degree of polarization of modes i, j, and P max is the maximum of P i, j for all pairs i, j: where I M,m are the maximum and minimum intensities.Therefore, after Eq. ( 14) The equality in Eqs. ( 17) and ( 19) is reached when G just mixes the two electric fields with the maximum degree of polarization.
The upperbound in Eq. ( 19) agrees with previous results showing that the maximum visibility in the interference of two partially polarized transversal waves is reached by combining just the two components with the largest degree of polarization [14].The bound in Eq. ( 17) reproduces also basic results of classical metrology: i) the resolution increases without limit as P max → 1, and ii) resolution is independent of the total field intensity.We will see below that these two properties no longer hold in the quantum sector.

Interference between two scalar fields
Let us particularize this result to the standard case with two electric fields k = 2 where G generates phase-difference shifts.In this case, after Eqs. ( 10) and ( 16) we obtain where μ is the standard two-mode degree of coherence in Eq. (1).

Interference between two partially polarized fields
Let us consider the interference between two partially polarized fields in schemes of the Young type.This is k = 4 and where I is the 2 × 2 identity matrix.We are considering the field basis in which E 1 = E 1,x , E 2 = E 1,y denote the two components, say x and y, of one of the fields, while represent the components of the other one.Note that as reflected by the form of G in this scheme the two components x, y of each wave experience the same phase shift.The corresponding 4 × 4 correlation matrix can be expressed as and where U 1,2 are the unitary matrices performing the singular-value decomposition of U 2 , and μ S,I are the corresponding real and positive singular values, with μ S ≥ μ I for example.These singular values were introduced in Ref. [10] to represent the coherence between two partially polarized waves independently of their polarization states.
Applying Eq. ( 16) to this case and taking into account that For example, if μ S = μ I we get χ = μ S which is exactly the case of two scalar fields in Eq. ( 21).
On the other hand F C diverges when μ S → 1.All this is fully consistent with the properties of μ S,I as measures of coherence between two partially polarized waves [43,44].

Fields with vanishing second-order degrees of coherence and polarization
The above equivalences in Eqs. ( 21) and ( 26) between resolution and the standard degree of coherence apply just to two fields in a thermal-chaotic light state, but fails to be true in general for other more sophisticated field statistics, as already shown in Ref. [22].This is the case of field states whose probability distribution for the field amplitudes is an even function of any of the amplitudes, say 1) leads automatically to μ = 0 irrespectively of any other statistical properties of the field state.Some examples are provided by most of the quantum field states reaching maximum phase-shift resolution that have E i E * j = Cδ i, j , for all i, j, where C is a constant [32] leading to μ = P = 0.This includes the so-called N00N states [45] that are examined in more detail in the quantum sector.

Quantum sector
In the quantum sector the upper bound F C in Eq. ( 13) cannot be defined because there is no proper probability distribution W (E) providing observable statistics via marginals.Actually, this is the distinctive feature of nonclassical light [1,46].
Nevertheless, it is possible to find an upper bound to F in terms of the so-called quantum Fisher information F Q [35], where p i and |ψ i are the eigenvalues and eigenstates of the density matrix ρ representing the field state In Eq. ( 27) the sum is extended to all pairs i, j with p i + p j = 0.When the field state is pure, i. e., ρ = |ψ ψ| we have that Eq. ( 27) greatly simplifies becoming the variance of the generator Ĝ in the field state |ψ Thus, in any case F Q is the quantum analog of F C in Eq. ( 13).Accordingly, a suitable estimation of the degree of coherence conveyed by the field state under phase shifts generated by Ĝ in the quantum sector is

Thermal-chaotic light
In order to compare the classical and quantum sectors with a meaningful example let us consider quantum thermal-chaotic light.In the mode basis where the correlation matrix is diagonal a † i a j = ni δ i, j the field state factorizes ρ = ⊗ k i=1 ρ i , with [1] where |n i are the corresponding photon-number states.
In order to check compatibility with the classical sector as simply and clearly as possible let us consider the two-field case k = 2.By symmetry, any generator of the form Ĝ ∝ exp(iϕ)a † 1 a 2 + exp(−iϕ)a 1 a † 2 is optimum for any ϕ, leading to a maximum Fisher information (see Appendix C) where P is the standard degree of polarization in Eq. ( 2), and n is the total mean number of photons: It can be appreciated that for P = 1 the classical result in Eq. ( 17) is obtained in the limit n → ∞, as it should be expected.On the other hand, for P → 1 we have that: i) the Fisher information does not diverge, contrary to the equivalent situation in the classical sector, but remains always finite, and ii) the Fisher information depends on the field intensity, so that for P = 1 it holds that F Q = 2tr(G 2 ) n. Actually, the scaling F Q ∝ n is known as the standard quantum limit, which is the largest Fisher information that can be reached with classical light in quantum optics [47,48].This may be compared with the scaling F Q ∝ n2 obtained for suitable nonclassical light, such as N00N states, as shown next in more detail.

N00N state
In the photon-number basis, the N00N states are This is an example of field states with vanishing degrees of polarization and coherence μ = P = 0, since for all n > 1 it holds that E 1 E * 2 = 0, and Despite this, these states are extremely useful in interference since they provide much larger resolution than other states with larger values of μ and P.
To show this we note that the maximum quantum Fisher information holds for Ĝ ∝ a † 1 a 1 − a † 2 a 2 leading to F Q = 2tr(G 2 ) n2 , where n = n is the total mean number of photons [32].The scaling F Q ∝ n2 is known as Heisenberg limit, that clearly surpasses the scaling F Q ∝ n that can be obtained with classical light [49].
The optimum Fisher information can be approached via ideal phase measurement whose statistics is given by projection on the relative-phase states [50] leading to the phase-difference distribution with Fisher information F = n 2 .The relative phase states emerge from a polar decomposition of the product of complex amplitudes a 1 a † 2 into a product of amplitude modulus and exponential of the phase difference.This problem has unitary solution for the exponential of the relative phase at difference with the equivalent single-mode polar decompositions [50].
It is worth noting that the total number n and the relative phase φ are commuting observables as the phase-number counterpart of commutation between total momentum and relative position for two particles [p 1 + p 2 , x 1 − x 2 ] = 0 since, roughly speaking, we may say that n = n 1 + n 2 and φ = φ 1 − φ 2 , where n 1,2 and φ 1,2 are the corresponding variables in the respective field mode.This does not contradict that single-mode number and phase do not commute, [n j , φ j ] = 0.

Conclusions
The classic degrees of coherence and polarization may admit different legitimate generalizations focusing on different field properties.For example, the metrological-based approach presented here is naturally different from global assessments of coherence and polarization in terms of distances of the light state to fully incoherent or unpolarized light [25][26][27][28]51].This is also different from the degree of polarization in Ref. [52] defined in terms of the minimum overlap between the original and transformed field state under any SU(2) transformation.In this work we are considering exclusively infinitesimal transformations in the spirit of precision metrology, while typically minimum overlap is obtained for large phase shifts.
We highlight the applicability of one and the same formalism both to the classical and quantum optical realms, obtaining parallel and consistent results in both of them [18,[53][54][55][56].
We have checked the consistency of this approach in the case of thermal light and N00N states.We think that there will be no consistency problems in any situation with multicomponent fields and nonGasussian statistics because of the practical nature of this formalism and its focus on the observables consequences of coherence.

A. Upper bound to Fisher information in classical optics
In this Appendix we demonstrate the upper bound for Fisher information in classical optics in Eq. (13).To simplify notation let us consider discrete indices instead of continuous variables so that the measured statistics P j of the j variable can be expressed in terms of the joint statistics of the j and k variables W j,k as where for simplicity we omit expressing explicitly the φ dependence.Let us demonstrate that F ≤ F C where where primes indicate derivation with respect to φ .Then where Λ j,k is the conditional probability of k given j Finally As a further alternative demonstration let us show that the Hellinger distance in Eq. ( 7) H P between the probability distributions P j = P j (φ ) and Pj = Pj (φ + δ φ) is smaller than the Hellinger distance H W between W j,k = W j,k (φ ) and Wj,k = Wj,k (φ + δ φ) This is because P j Pj ≥ ∑ k W j,k Wj,k .To show this let us consider the square of the left-hand side and the square of the right-hand side so that P j Pj ≥ ∑ k W j,k Wj,k holds after comparing the right-hand side of Eqs. ( 43) and (44) because a + b ≥ 2 √ ab for all a, b ≥ 0. Finally this implies H P ≤ H W and thus F ≤ F C after Eq. ( 7).

B. Fisher information for classical thermal-chaotic light
Let us particularize F C in Eq. ( 13) to the case of thermal-chaotic light.After Eq. ( 15) we have where Thus Eq. ( 13) leads to Using the Gaussian momentum theorem [1] we readily get leading to In the field basis where M is diagonal we get where Note the formal similarity with the quantum Fisher information in Eq. (27).We finally get G i, j 2 P 2 i, j where P i, j is the two-mode degree of polarization of modes i, j, and P max is the maximum of P i, j for all pairs i, j: where I M,m are the maximum and minimum intensities.The equality in Eq. ( 53) is reached when G mixes just the electric fields with the maximum and minimum intensities.

C. Quantum Fisher information for thermal-chaotic light
Since the thermal-chaotic states in Eq. ( 31) are mixed we have to resort to Eq. ( 27) to compute its quantum Fisher information.Since the product of number states |n 1 |n 2 are the eigenstates of ρ, for Ĝ = exp(iϕ)a † 1 a 2 + exp(−iϕ)a 1 a † 2 we have to compute In the first place we have The two Kronecker deltas are incompatible so that in the modulus square there are no crossed contributions and where After some simple algebra This expression can be fruitfully simplified using the total mean number of photons n = n1 + n2 and the degree of polarization in Eq. ( 2) P = | n1 − n2 |/( n1 + n2 ) where we have taken into account that for the thermal state in Eq. ( 31) E 1 E * 2 = 0, so that where we have used that tr(G 2 ) = 2.