Critical exponents describe leading singular powers near a continuous phase transition. For reduced temperature , common definitions are the spontaneous order parameter , magnetic susceptibility , singular heat capacity , correlation length , and the critical equation of state . At criticality the connected correlation function has decay power .
In Landau theory, solve the stationary equation, evaluate its stable minimum, and differentiate with respect to the conjugate variables. For a scalar quartic theory with , , the equation of state is . Its ordered solution and differentiated response give and , calculated explicitly in the unheaded solution below. At the same equation gives . The minimized zero-field singular potential is zero above the transition and below, giving a heat-capacity jump and . Expanding the gradient functional about a stable minimum gives inverse quadratic response , so and the Gaussian critical decay has .
These are mean-field critical exponents. Renormalization group methods obtain fluctuation-corrected powers from fixed-point scaling dimensions; they need not equal the Landau values below the upper critical dimension. At a marginal dimension logarithms can supplement the leading powers.
The Maxwell construction replaces a nonconvex homogeneous free-energy density by its coexistence envelope. If excludes the source, the coexisting values satisfy
Thus one straight line is tangent at both endpoints. Equivalently,
the equal-area form of the common-tangent construction. At fixed average between the endpoints, the lever rule gives . Such a mixture has the tangent-line bulk energy, below the energy of an unstable homogeneous state.
The two phases occupy spatial domains separated by domain walls. The gradient term gives a positive wall energy, but a macroscopic interface has area of order rather than bulk volume , so its cost per volume vanishes in the thermodynamic limit. For the symmetric quartic example a scalar quartic domain wall obeys and interpolates between as , with and . Multiplication of the differential equation by gives the first integral , explaining the finite positive interface cost. A uniform unconstrained system need not form both domains: phase separation is particularly relevant when the average order parameter is fixed or boundaries enforce competing phases.
Hysteresis occurs when the system follows a metastable local minimum after another minimum has become the equilibrium one. For , the equation of state is . The hysteresis spinodals of a scalar quartic potential follow from :
Following local minima until these limits gives different switching fields on increasing and decreasing . Equilibrium instead switches at . Actual switching can occur before a mean-field spinodal through nucleation, and depends on time scales; hysteresis is not a property of a globally minimized equilibrium potential. The second panel below distinguishes the local branches and their limiting switches.
A continuous phase transition has a continuously vanishing order parameter and no latent heat. A first-order phase transition exchanges globally stable phases with discontinuous first derivatives of a thermodynamic potential; the order parameter generally jumps, and an entropy jump gives latent heat. In Landau theory the distinction is determined by global minima, not just by the appearance of additional stationary points.
For the uniform scalar potential with and , its stationary equation is . For , is the minimum; for , the minima are . They approach zero continuously as , giving an ordinary continuous transition.
For and , use the sextic potential. Write . Stationarity of a nonzero phase requires , while equality of its potential with requires . Eliminating yields
At this value the potential factors as , proving that the three stationary zeros really are degenerate global minima. Crossing this line makes the order parameter jump: the transition is first order. The disordered and ordered spinodal points instead occur at and respectively, delimiting metastability; these are not the equilibrium transition. If the symmetry allows a cubic term, that term can also produce a first-order transition.
An order parameter distinguishes phases macroscopically. In an Ising model it is the local coarse-grained magnetization; the uniform mean is . In a selected ordered phase , whereas the disordered phase has at zero conjugate field. The ordered signs are related by spin inversion symmetry. A finite zero-field system may average both signs to zero, so spontaneous symmetry breaking is defined using the thermodynamic limit before removing a symmetry-selecting field. For other systems the order parameter can be a vector, a density difference, or a tensor rather than a single magnetization.
Landau-Ginzburg theory makes the order parameter a spatial field and expands a local free energy consistent with the symmetries. For a short-range scalar system with spin inversion symmetry at , a useful expression is
The linear source is the field conjugate to an order parameter. Odd terms at zero field are forbidden by spin inversion symmetry. The coefficients depend smoothly on control parameters away from singular fluctuation effects. With the quartic term suffices locally; if , take for stability. The Landau approximation minimizes this functional; statistical field theory also integrates fluctuations about those minima.
A Positron is the electron's antiparticle: it has the same mass and spin one-half, opposite electric charge , and fermionic statistics. A state has positive energy and physical momentum , as follows from the Hamiltonian and momentum operators
The charge commutator gives .
The term in the field has negative frequency, but it creates this positive-energy Positron. Treating its coefficient as an annihilator of a physical negative-energy electron would give an unbounded energy interpretation. Fermionic quantization instead reorders that sector, giving positive antiparticle occupation energies and the opposite signed charge. The historical Dirac sea describes a Positron as a hole in filled negative-energy electron levels; the field-theory interpretation uses an antiparticle creation operator and normal ordering, without requiring a literal infinite material sea.
Under charge conjugation of a Dirac field, , where , particle and antiparticle modes interchange. Electron–Positron pair creation and annihilation change total particle number while conserving electric charge. Positrons are positive-energy antiparticles, not physical negative-energy electrons.
With a vacuum annihilated by all , define the particle and antiparticle occupation operators of a Dirac field
Each counts its corresponding occupied modes. The total positive particle count is
Commuting it through either creation operator gives and , so both kinds of one-particle states have eigenvalue one. In a normalized discrete mode, by the anticommutation relations, so its occupation eigenvalues are .
For a Dirac field there is also a distinct conserved signed fermion number. Inserting the mode expansion into the normal-ordered local density gives
Spatial integration makes the same-sector momenta equal. The cross terms vanish by ; normal ordering of supplies the minus sign for the antiparticles. Thus the density integral is not the total positive particle count. It generates the global phase symmetry, and for electron charge the electric charge is .
The free normal-ordered Hamiltonian is
so and are conserved in the free theory. In an interacting theory, pair creation can change the total count while preserving the signed charge. This distinguishes a positive occupation number operator from the charge behind Dirac fermion number conservation.
For the specified mode normalization, the canonical anticommutation relations are
and every other anticommutator of the operators vanishes. For the Dirac field, they imply the Equal-time canonical anticommutator of a Dirac field
To check the normalization, the two sectors contribute and with opposite spatial Fourier signs. Reverse in the second contribution. The completeness relations then give
where . This cancels the mode denominator and leaves the Fourier representation of the delta function. Same-momentum orthogonality must not be used in place of this reversed-momentum identity.
For arbitrary spacetime points,
For spacelike , choose a Lorentz frame with ; reversing spatial momentum shows . Its derivatives vanish there as well, so the field anticommutators vanish at spacelike separation. This is fermionic microcausality. The anticommutation rules make the occupation of each normalized mode zero or one, implementing the Pauli exclusion principle, and make the normal-ordered free Hamiltonian positive for both particle sectors.
Use signature and the free Dirac equation . For positive , substitution of the two frequency signs gives
Multiplying by the opposite factor shows . Each equation has two independent solutions for a massive field.
In the Dirac basis, and . For orthonormal two-component spin labels , convenient solutions are
The two-block equations verify these expressions using . At rest the positive-frequency solutions have only upper components and the negative-frequency solutions only lower components. Their normalizations are and , . Their completeness relations are
The field expansion combines fermionic annihilation operators multiplying the positive-frequency waves with antiparticle fermionic creation operators multiplying the negative-frequency waves. The latter frequency sign is not the energy of a physical negative-energy antiparticle state.
For equal prior probabilities, . The projected score is normal under either class because it is a linear image of a multivariate normal vector, and its variance is
Moreover , so the midpoint threshold lies from either projected mean. Hence
Standardizing the two normal distributions, with the standard normal distribution function, gives
Thus the equal-covariance Gaussian classification error is . The Mahalanobis distance between the class means is their separation in units of within-class noise. It yields as and as . The stipulated means the two class means are distinct.
Under equal costs for either wrong assignment, the Bayes classifier chooses the class of greatest posterior probability. With positive prior probabilities, it assigns class one when ; ties can be resolved arbitrarily. For the two multivariate normal densities with common positive-definite covariance matrix , their normalizing factors cancel. The log posterior odds are
Therefore the Gaussian Bayes classifier has the required linear form, with
A larger prior probability for class one decreases , expanding its decision region. The prior-dependent Gaussian discriminant boundary is the hyperplane .
The within-group and between-group scatter decomposition separates noise from group separation: measures variation about each group sample mean, and measures separation of those means about the overall sample mean. In linear discriminant analysis, seek a projection maximizing between-group relative to within-group variation,
Normalize by ; a Lagrange multiplier calculation gives the generalized eigenvalue problem . The leading canonical discriminant directions have the largest eigenvalues, and subsequent directions are orthogonal in the inner product. There are at most nonzero separating directions. Projecting observations onto these directions gives a low-dimensional display of group separation.
For classification, estimate the common covariance matrix by and use the linear discriminant analysis scores
where are class prior probabilities. Assign to the largest score. The same within-group scatter sets the noise scale, while between-group scatter selects the directions in which the group means separate most strongly.
Let , and . Define the within-group and between-group scatter matrices by
The within-group and between-group scatter decomposition gives total scatter : in each group the residuals about sum to zero, so all cross terms vanish. Under unrestricted group means, the maximum-likelihood estimators are and . Under the common-mean null hypothesis, they are and .
For either fitted scatter matrix , substitution into the log-likelihood gives
since . Thus the maximized likelihood ratio is
A generalized likelihood-ratio test rejects for small , equivalently
The cutoff is chosen for the desired significance level. The determinant ratio itself is the Wilks lambda statistic. The derivation assumes is positive-definite, which for continuous normal samples requires enough within-group degrees of freedom, ; singular does not support this ordinary determinant-based test without a modified model.
Use the usual sample mean and the divisor- sample covariance matrix,
The residual sum is zero, so expanding gives
Multiplying the independent multivariate normal densities and taking logarithms yields the log-likelihood
Here is the determinant and denotes the trace. The maximum-likelihood estimators are
If instead denotes the divisor- sample covariance matrix, replace by and then . These formulas describe a nonsingular maximum-likelihood estimator when the empirical scatter is positive-definite; if it is singular, there is no maximizer over positive-definite covariance matrices.
Write and . Apply the Schur complement covariance to the first two coordinates, conditioning on the third. The conditional covariance and conditional variances are
Divide the conditional covariance by the product of the conditional standard deviations. The conditional correlation is therefore
This is the partial correlation after removing the linear effect of . For a nonsingular multivariate normal distribution, both residual variances are positive and the answer is independent of . If a residual variance vanishes, the requested correlation coefficient is undefined, even though the residual covariance formula remains valid.
Put , the precision matrix, with scalar and containing all remaining coordinates. Holding in the multivariate normal density, the terms involving in its exponent are
Completing the square gives a normal distribution with mean and
Equivalently, the block matrix inverse identity gives . The Gaussian conditional variance from precision is a reciprocal diagonal entry of the precision matrix, rather than the marginal variance .
Write the expected value as and assume the relevant covariance matrix is invertible. Remove the linear predictor from the first block by defining
A linear image of a multivariate normal vector is again multivariate normal. Direct covariance calculation gives
Thus and are independent: uncorrelated jointly Gaussian variables are independent. Expanding the covariance matrix of , using , gives
Conditioning on translates by the fixed vector and leaves its covariance matrix unchanged. Consequently the conditional multivariate normal distribution is
The conditional covariance is the Schur complement and does not depend on .
Define . Grouping equal claim sizes gives the pathwise identity
Conditional on , the independent claims receive category labels with probabilities . There are assignments producing prescribed category counts, and every assignment has probability . Thus the conditional probability distribution is a multinomial distribution:
when the nonnegative counts sum to , and is zero otherwise.
For , condition first on and use the probability generating function of a Poisson distribution:
This factorization is the joint probability generating function of independent Poisson distribution variables. Hence
A category with has zero count almost surely. The categorical thinning of a Poisson claim count produces independence after averaging over the random total, even though the counts are constrained to sum to conditional on that total.
Use the preceding construction with type- claims of deterministic size . For , the answer is
The characteristic function verifies this directly:
Hence has the stated compound Poisson distribution. Distinctness of the makes the atomic probabilities unambiguous.
A zero coefficient is allowed. It corresponds to zero-sized claims in this representation and contributes nothing to . If a representation using only positive claims is preferred, remove that category: its count parameter becomes , with probabilities on the positive values. The zero claim sizes in a compound Poisson representation explain why the count parameter need not be unique. If , or if all positive-size categories have zero rate, the aggregate is identically zero.
Let . For , the resulting compound Poisson distribution has
Thus the merged claim size probability distribution is a rate-weighted mixture distribution, not an equally weighted mixture distribution unless the rates agree.
To prove this without any moment-generating function assumption, let be the characteristic function of a claim of type . Conditioning on its Poisson distribution count gives
Independence of the aggregate risks therefore gives
The weighted characteristic function inside is precisely that of the displayed mixture distribution. This is the transform of a sum of independent claims with independent count , proving the claim by uniqueness of characteristic functions. This is Poisson superposition of insurance portfolios. If , all risks make no claims almost surely and ; the severity law is then irrelevant.
For these counts, start with any . The conditional missing count from part (ii) has mean
The E-step objective is
Its strictly concave M-step is explicit:
Equivalently, the update map is . Its fixed-point equation is
The unique root in is the maximum-likelihood estimate
For example, starting at gives , then approximately , and subsequent iterates approach the displayed root. Convergence is also transparent without relying only on the general EM theorem: is increasing and
This is positive below the feasible fixed point and negative above it. Monotonicity of prevents crossing that fixed point, so the iterates converge monotonically toward it from either side. The original log-likelihood is strictly concave and diverges to minus infinity at both endpoints, establishing that the fixed point is the unique global maximum.

Pinned article: Introduction to the OurBigBook Project

Welcome to the OurBigBook Project! Our goal is to create the perfect publishing platform for STEM subjects, and get university-level students to write the best free STEM tutorials ever.
Everyone is welcome to create an account and play with the site: ourbigbook.com/go/register. We belive that students themselves can write amazing tutorials, but teachers are welcome too. You can write about anything you want, it doesn't have to be STEM or even educational. Silly test content is very welcome and you won't be penalized in any way. Just keep it legal!
We have two killer features:
  1. topics: topics group articles by different users with the same title, e.g. here is the topic for the "Fundamental Theorem of Calculus" ourbigbook.com/go/topic/fundamental-theorem-of-calculus
    Articles of different users are sorted by upvote within each article page. This feature is a bit like:
    • a Wikipedia where each user can have their own version of each article
    • a Q&A website like Stack Overflow, where multiple people can give their views on a given topic, and the best ones are sorted by upvote. Except you don't need to wait for someone to ask first, and any topic goes, no matter how narrow or broad
    This feature makes it possible for readers to find better explanations of any topic created by other writers. And it allows writers to create an explanation in a place that readers might actually find it.
    Figure 1.
    Screenshot of the "Derivative" topic page
    . View it live at: ourbigbook.com/go/topic/derivative
  2. local editing: you can store all your personal knowledge base content locally in a plaintext markup format that can be edited locally and published either:
    This way you can be sure that even if OurBigBook.com were to go down one day (which we have no plans to do as it is quite cheap to host!), your content will still be perfectly readable as a static site.
    Figure 5. . You can also edit articles on the Web editor without installing anything locally.
    Video 3.
    Edit locally and publish demo
    . Source. This shows editing OurBigBook Markup and publishing it using the Visual Studio Code extension.
  3. https://raw.githubusercontent.com/ourbigbook/ourbigbook-media/master/feature/x/hilbert-space-arrow.png
  4. Infinitely deep tables of contents:
    Figure 6.
    Dynamic article tree with infinitely deep table of contents
    .
    Descendant pages can also show up as toplevel e.g.: ourbigbook.com/cirosantilli/chordate-subclade
All our software is open source and hosted at: github.com/ourbigbook/ourbigbook
Further documentation can be found at: docs.ourbigbook.com
Feel free to reach our to us for any help or suggestions: docs.ourbigbook.com/#contact