Indefinite orthogonal group¶

The real linear group O(p,q) preserving a nondegenerate symmetric form with both positive and negative directions.

Version: v1 · 2026-10-07 · History
Domain-specific #: 13914
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Linear and Lie Groups → Mathematics
Aliases: Pseudo Orthogonal Group

Core Idea¶

An indefinite orthogonal group is the group of invertible real linear transformations that preserve a nondegenerate symmetric bilinear form with both positive and negative directions. If a basis represents the form by a matrix G of signature (p,q), with p,q > 0, its members satisfy AᵀGA = G. The notation O(p,q) names the resulting group up to change of basis. Interchanging p and q replaces G by −G and gives an isomorphic group.^[1]

The word indefinite qualifies the form, not the certainty of the definition. A transformation either preserves the specified form or it does not. The mixed signature distinguishes this family from the definite O(n), while the varying values of p and q keep the entry from collapsing into the Lorentz group alone.^[1]

Structural Signature¶

Real linear carrier: a finite-dimensional real vector space V, with dimension p+q. The group acts linearly on V rather than by arbitrary spacetime motions.^[1]
Declared invariant: a nondegenerate symmetric bilinear form B represented by G. Both p and q are positive, so there are directions of both signs.^[1]
Membership test: A is an invertible real linear map and AᵀGA = G. Composition and inverse preserve this condition, making the solution set a group.^[1]
Family parameter: (p,q) identifies the signature; the chosen coordinate basis and the particular diagonal representative of G do not define a different abstract group.^[1]

Removing the mixed signature changes the family to a definite orthogonal group. Removing nondegeneracy changes the form stabilizer. Keeping only invertibility without the preservation equation admits matrices outside O(p,q). These are identity failures, not small variations.^[1]

What It Is Not¶

O(p,q) is not every symmetry of a space carrying an indefinite form. Translations of Minkowski spacetime preserve intervals between points but are affine operations; they are not elements of the linear Lorentz group O(1,3). Nor does preservation of determinant alone establish membership: the declared form must be preserved.^[1]

It is not the definite O(n), obtained when p or q is zero. It is also not only the identity component. For p,q > 0 the full group has disconnected components; an application may select a component, but that selection removes some matrices that still satisfy AᵀGA = G.^[1]^[2]

Scope of Application¶

The entry applies to finite-dimensional real spaces with a specified nondegenerate symmetric form of mixed signature. It covers the Lorentzian case O(1,3) and split cases O(n,n), including matrices used in a fermionic quantum Monte Carlo construction. Those applications give the same membership equation different physical and computational interpretations.^[1]^[2]

One must state the signature and the form before asserting a group member. Complex Hermitian forms, alternating forms, degenerate metrics, nonlinear transformations, and translations require different definitions. Calling each one “orthogonal” by analogy would not pass this entry's membership test.

Clarity¶

The equation AᵀGA = G is a compact yes-or-no test. It tells a reader exactly what remains invariant even when individual coordinates change. For G = diag(I_p,−I_q), A can mix positive and negative coordinates; preserving the form does not mean preserving each coordinate block separately.^[1]^[2]

The signature separates questions often blurred by the generic word “orthogonal.” Positive definite O(n) preserves a Euclidean form. Lorentzian O(1,3) preserves one positive and three negative directions in the cited convention. Split O(n,n) preserves equal positive and negative counts. The same formula alone is insufficient until G and its signature are supplied.^[1]^[2]

Manages Complexity¶

A large matrix can have many entries, but the invariant equation organizes them as a group of allowed transformations. One can first check the form and the equation, then ask the application-specific question about components, time direction, determinant sign, or simulation weights. This prevents an inference about one component from being silently applied to the full group.^[1]^[2]

The distinction matters in the split example: Wang and colleagues classify four components of O(n,n) and give different signs for det(I+M) depending on the component. Membership in O(n,n) alone does not make every determinant nonnegative. In their positive-shift auxiliary-field spinless t–V construction, the relevant real generators exponentiate to products in the identity component O⁺⁺(n,n), so det(I+M) is nonnegative. Their alternative negative-shift construction can use O⁻⁻(n,n) and cancel its nonpositive determinant with a negative prefactor.^[2]

Abstract Reasoning¶

Given a proposed instance, write the bilinear form B and its matrix G. Check that G is symmetric and nondegenerate, find the numbers p and q of positive and negative directions, and require p,q > 0. Then test AᵀGA = G. A basis change changes the matrix presentation of both the form and transformation, but not the fact of preservation.^[1]

Only after membership is established should one infer consequences tied to a particular subgroup or application. If orientation or time direction matters, determine the relevant component. If a determinant sign matters, check the theorem's component hypotheses. A counterexample to those extra conditions does not refute membership in the full indefinite orthogonal group.^[1]^[2]

Knowledge Transfer¶

The Lorentz example teaches the invariant-form test in a familiar four-dimensional setting. The same test transfers to Wang and colleagues' split group by changing the signature to (n,n) and the carrier to a 2n-dimensional real single-particle coordinate space, on which real 2n×2n matrices act. The transfer is mathematical membership, not a claim that a fermionic simulation has a physical time direction or that every Lorentz transformation solves a Monte Carlo sign problem.^[1]^[2]

The reverse transfer also helps: the split example makes it clear that O(p,q) is a family, not a synonym for relativistic boosts. When the source changes, retain the four structural roles and check any added application conditions independently.

Examples¶

Lorentz symmetry. Let V be the real four-dimensional coordinate space of Minkowski spacetime and G = diag(1,−1,−1,−1). The form has signature (1,3), and a Lorentz matrix A lies in O(1,3) exactly when AᵀGA = G. Feng and Stange distinguish the full group from its identity component, which also preserves the selected time direction and orientation. The spacetime interpretation and those component choices belong to this case, not to the whole O(p,q) family.^[1]

Split orthogonal matrices in fermionic QMC. In Wang and colleagues' construction, V is a real 2n-dimensional single-particle coordinate space organized by two n-site sublattices, with η = diag(I_n,−I_n). Real 2n×2n transformation matrices M satisfying MᵀηM = η lie in O(n,n). In the paper's positive-shift auxiliary-field spinless t–V construction, the relevant matrix products lie in O⁺⁺(n,n), making det(I+M) nonnegative. Other component choices can give a nonpositive determinant that a method-specific prefactor cancels. Neither sign conclusion follows from bare O(n,n) membership.^[2]

Structural Tensions¶

Full preservation group versus component restriction. Every element of O(p,q) preserves the mixed form, but the full group includes disconnected transformations. The Lorentz application may require preservation of time direction and orientation; the positive-shift auxiliary-field split QMC argument requires the O⁺⁺(n,n) component for its nonnegative determinant, while a negative-shift variant uses a prefactor to compensate the sign of an O⁻⁻(n,n) determinant. Choosing a proper component gains those added guarantees while excluding some otherwise valid form-preserving transformations. The question is which invariant the application actually needs: the bilinear form alone, or the form together with an orientation or sign condition.^[1]^[2]

Structural–Framed Character¶

Formal structure: the carrier, mixed-signature form, and preservation equation give an exact membership rule. Evaluative weight: the rule is mathematical, with no built-in judgment that one group action is desirable. Human-practice dependence: people choose examples and coordinate conventions, but membership does not depend on a social practice. Institutional origin: the term belongs to mathematics, yet the form-preserving structure is not a product of an institution's rules.^[1]

Vocabulary travel: “orthogonal” and “indefinite” retain technical meanings tied to real forms; casual use for any mixed or uncertain system loses the equation. Import versus recognition: in a new mathematical example one can recognize the group by proving the equation; applying the label to an unrelated process merely imports the terminology. Its character: structural-dominant domain-specific. The structure is exact and portable across its mathematical instances, while the named family still depends on real linear algebra and mixed-signature bilinear forms.

Structural Core vs. Domain Accent¶

The skeletal relation is a transformation preserving a declared invariant. The live Prime Symmetry captures this much more general idea, but does not require a real vector space, a nondegenerate symmetric form, or mixed signature. A future Prime built only from O(p,q)'s preservation equation would duplicate that broad symmetry identity; a proposed more specialized prime would need independently evidenced cross-domain residue beyond this mathematical family.

The domain-bound mechanism is exact: finite-dimensional real linear transformations preserve a nondegenerate symmetric form with p,q > 0. Minkowski time and space coordinates, sublattice ordering, determinant weights, and component labels are accents of particular uses. They do not alter the membership equation, but they determine which extra claims can be made in a given use.^[1]^[2]

The named entry does not clear the Prime bar. Its distinguishing content is the mathematical type of carrier and invariant. Generalizing away those conditions leaves the existing broad notion of symmetry, while retaining them leaves a domain-specific family. No new substrate-independent mechanism is established by the two examples.

This entry is a kind of Classical group.

O(p,q) realizes form preservation and therefore is related to Symmetry, but an edge must use the live Prime's exact identity rather than the ordinary-language meaning of symmetry. The independently accepted strict subsumption edge to Classical group records a kind-of relation: every O(p,q) with p,q > 0 is one of the real orthogonal matrix-group families preserving a nondegenerate form. The broader Classical group also includes definite orthogonal, unitary, and symplectic families, so it exists without this child. This is a taxonomic parent, not a component contained inside each matrix.^[1]

The indefinite inner-product or pseudo-Euclidean space is the carrier whose form is preserved; it is not the group of transformations itself. A Lie group describes smooth group structure, a property the indefinite orthogonal groups have, but it does not replace the mixed-form membership test.^[1]

Relationships to Other Abstractions¶

Current abstraction Indefinite orthogonal group Domain-specific

Parents (1) — more general patterns this builds on

Indefinite orthogonal group is a kind of Classical group Domain-specific

Every real mixed-signature orthogonal group is a classical form-preserving matrix group.

Hierarchy path (1) — routes to 1 parentless root

Indefinite orthogonal group → Classical group → Classification

Neighborhood in Abstraction Space¶

Indefinite orthogonal group sits in a moderately populated region (54^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Vector Spaces & Linear Structures (17 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Do not identify O(p,q) by a Lorentzian example alone, by a familiar boost matrix, or by the appearance of positive and negative entries in an arbitrary matrix. The signature belongs to the form G, and membership belongs to transformations satisfying AᵀGA = G. Distinguish the full group from its special or identity components, and distinguish the linear group from affine spacetime translations.^[1]^[2]

References¶

[1] Y. Feng and Katherine E. Stange, The spin homomorphism SL2(C) → SO1,3(R): A summary from multiple sources, University of Colorado mathematical notes, §§1.4.1–1.4.2, PDF pp. 2–4. https://math.colorado.edu/~kstange/papers/notes-Spin.pdf registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w

[2] Lei Wang et al., Split orthogonal group: A guiding principle for sign-problem-free fermionic simulations, Physical Review Letters 115, 250601 (2015), DOI: 10.1103/PhysRevLett.115.250601, Eq. (2), Theorem (3), Corollary (4), and spinless t–V example. https://arxiv.org/html/1506.05349v3 registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l