R-NP: Nondeterministic Polynomial Computation on R

• R-NP is defined via polynomial-time verification on R-machines, yielding equivalent characterizations through satisfiability and existential second-order metafinite logic.
• Under weak conditions on R such as bipointedness, finite type, and all constants, the canonical complete problem SAT(R) anchors the theory of R-NP.
• In the real-number setting, NP_R includes the quadratic-feasibility problem whose exponential deterministic complexity separates P_R from NP_R.

R-NP denotes nondeterministic polynomial-time computation relative to a structure $R$, and in the real setting it appears as $\text{NP}_{\mathbb R}$ in the Blum–Cucker–Shub–Smale model. In the abstract formulation, $\text{NP}(R)$ is defined by polynomial-time verification on $R$-machines and, under weak conditions on $R$, admits equivalent characterizations by satisfiability and by existential second-order metafinite logic. In the real-number setting, the class $\text{NP}_{\mathbb R}$ contains the quadratic-feasibility problem, and a scheme-theoretic lower bound yields exponential deterministic complexity for that problem, giving $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$ (Kirn et al., 7 Oct 2025, Çivril, 2021).

1. Machine models and the verifier definition

In the abstract framework, computation is parameterized by a fixed first-order structure

$R=(R,C,F,P),$

where $C$ is a set of constants, $F$ a set of function symbols, and $\text{NP}_{\mathbb R}$0 a set of relation symbols. An $\text{NP}_{\mathbb R}$1-machine is a RAM-style device with an unbounded array of $\text{NP}_{\mathbb R}$2-registers $\text{NP}_{\mathbb R}$3 and a bounded array of index-registers $\text{NP}_{\mathbb R}$4. Its instruction set is induced by the vocabulary of $\text{NP}_{\mathbb R}$5: assignment of constants, evaluation of function symbols, branching on equality and predicates over $\text{NP}_{\mathbb R}$6, indirect copy, increment and monus on index registers, branching over $\text{NP}_{\mathbb R}$7, and optionally oracle tests. A machine is polynomial-time if its running time satisfies $\text{NP}_{\mathbb R}$8 for some $\text{NP}_{\mathbb R}$9 (Kirn et al., 7 Oct 2025).

The class $\text{NP}(R)$0 is defined by verification. A language $\text{NP}(R)$1 lies in $\text{NP}(R)$2 if there exist a polynomial $\text{NP}(R)$3 and a polynomial-time $\text{NP}(R)$4-machine $\text{NP}(R)$5 such that for every input $\text{NP}(R)$6,

$\text{NP}(R)$7

Here $\text{NP}(R)$8 is a verifier and $\text{NP}(R)$9 is a certificate. The same source states that $R$0 is equivalently the class $R$1 of the $R$2-machine hierarchy, with $R$3 (Kirn et al., 7 Oct 2025).

A closely related real-number model is the BCSS model over $R$4. There, a real-number machine may store real registers, apply in one step any of the basic field operations $R$5, and test any register-wise inequality $R$6. A decision problem over $R$7 is in $R$8 if a deterministic BCSS machine decides it in polynomial time, and in $R$9 if it is decidable by a nondeterministic BCSS machine, equivalently by a polynomial-time verifier with polynomial-size real certificate (Çivril, 2021).

2. Structural conditions on $R$0 and the complete problem $R$1

The abstract theory isolates three weak conditions on $R$2 that support a Cook–Levin analogue: bipointedness, finite type, and all constants. Bipointed means that $R$3 has two distinct constants $R$4; finite type means $R$5; all constants means that for each $R$6 there is a constant symbol $R$7 with $R$8. The roles of these assumptions are explicit: bipointedness is needed to encode binary choices, finite type gives a uniform finite encoding of atomic operations, and all constants allow the hardness proof to hard-code the input $R$9 into a formula $\text{NP}_{\mathbb R}$0 via constants (Kirn et al., 7 Oct 2025).

Condition	Formal content	Stated role
Bipointed	$\text{NP}_{\mathbb R}$1 in $\text{NP}_{\mathbb R}$2	Encode binary choices
Finite type	$\text{NP}_{\mathbb R}$3	Uniform finite encoding of operations
All constants	$\text{NP}_{\mathbb R}$4 for each $\text{NP}_{\mathbb R}$5	Hard-code inputs into formulas

Under these assumptions, the canonical complete problem is $\text{NP}_{\mathbb R}$6, the satisfiability of a quantifier-free Boolean combination of atomic $\text{NP}_{\mathbb R}$7-formulas. An input is a finite encoding of a formula $\text{NP}_{\mathbb R}$8 built from symbols in $\text{NP}_{\mathbb R}$9 using $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$0, and $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$1 is satisfiable in $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$2 if there exist $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$3 such that $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$4. The theorem stated in the source is that if $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$5 is bipointed, finite-type, and all-constants, then $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$6 and $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$7 is $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$8-hard, hence $\text{P}_{\mathbb R}\neq \text{NP}_{\mathbb R}$9-complete (Kirn et al., 7 Oct 2025).

The same work also emphasizes a limitation: there are infinite-vocabulary structures for which $R=(R,C,F,P),$0 does not have a complete problem. Even in these cases, however, $R=(R,C,F,P),$1 still has a characterization in terms of existential second-order metafinite logic. The paper states that this suggests descriptive complexity theory is well suited to working with infinite-vocabulary structures, such as real vector spaces (Kirn et al., 7 Oct 2025).

3. Descriptive complexity, metafinite logic, and higher hierarchies

The descriptive-complexity characterization is formulated on $R=(R,C,F,P),$2-metafinite structures. A metafinite vocabulary over $R=(R,C,F,P),$3 is $R=(R,C,F,P),$4, where $R=(R,C,F,P),$5 is a finite relational vocabulary interpreted on a finite set $R=(R,C,F,P),$6, $R=(R,C,F,P),$7 is the secondary infinite structure, and $R=(R,C,F,P),$8 is a finite family of weight functions. An $R=(R,C,F,P),$9-metafinite structure is $C$0. Existential second-order metafinite logic over $C$1, denoted $C$2, then serves as the logical counterpart of nondeterministic computation (Kirn et al., 7 Oct 2025).

The stated Fagin analogue is that for any bipointed $C$3, $C$4 on $C$5-metafinite structures captures $C$6. One direction is data-complexity containment: given a fixed sentence $C$7, an $C$8-machine can evaluate $C$9 in polynomial time. The converse direction encodes accepting computations by existentially quantified second-order relations that represent the computation table. This establishes the three-way correspondence between verification, satisfiability, and existential second-order metafinite logic under the appropriate assumptions (Kirn et al., 7 Oct 2025).

The same framework extends to the full polynomial hierarchy over $F$0. A language $F$1 is in $F$2 if there exist a polynomial $F$3 and a polynomial-time $F$4-machine $F$5 such that

$F$6

with $F$7 and alternating quantifier blocks. The classes $F$8, $F$9, the constant-free Boolean part $\text{NP}_{\mathbb R}$00, and the Boolean hierarchy $\text{NP}_{\mathbb R}$01 are defined analogously. The cited results include higher Cook–Levin analogues and oracle characterizations, notably

$\text{NP}_{\mathbb R}$02

Analogous statements are given for the Boolean hierarchy and constant-free machines (Kirn et al., 7 Oct 2025).

4. The real case: $\text{NP}_{\mathbb R}$03 and $\text{NP}_{\mathbb R}$04-completeness

Within the BCSS model over the reals, the quadratic-feasibility problem $\text{NP}_{\mathbb R}$05 is defined as follows. An instance consists of $\text{NP}_{\mathbb R}$06 quadratic polynomials

$\text{NP}_{\mathbb R}$07

with real coefficients, encoded by their coefficients, and the task is to decide whether there exists $\text{NP}_{\mathbb R}$08 satisfying all equations. The source states that it is known from Blum–Shub–Smale (1997) that $\text{NP}_{\mathbb R}$09 is $\text{NP}_{\mathbb R}$10-complete (Çivril, 2021).

The lower-bound argument is formulated in terms of two quantities. The deterministic complexity $\text{NP}_{\mathbb R}$11 is the minimum number of machine steps needed in the worst case. The quantity $\text{NP}_{\mathbb R}$12 is the maximum number of instances in any prime homogeneous simple sub-problem. A simple sub-problem is a collection of instances all having the same constant Hilbert polynomial; homogeneous means that all instances use exactly the same $\text{NP}_{\mathbb R}$13 variables; prime means that no nontrivial unit operation can map one instance to another in more than one way. The fundamental lemma, adjusted to $\text{NP}_{\mathbb R}$14, is

$\text{NP}_{\mathbb R}$15

This reduces the complexity lower bound to the construction of a large prime homogeneous simple sub-problem (Çivril, 2021).

5. Construction of the exponential lower bound over $\text{NP}_{\mathbb R}$16

The base gadget is a pair of quadratic systems in three variables $\text{NP}_{\mathbb R}$17, each with four equations. The first system is

$\text{NP}_{\mathbb R}$18

and it has the unique solution $\text{NP}_{\mathbb R}$19. The second system uses the same first three equations but replaces the fourth by

$\text{NP}_{\mathbb R}$20

and it has the unique solution $\text{NP}_{\mathbb R}$21. Since both have Hilbert polynomial $\text{NP}_{\mathbb R}$22, they form a homogeneous simple sub-problem of size $\text{NP}_{\mathbb R}$23 (Çivril, 2021).

Inductively, if one has constructed $\text{NP}_{\mathbb R}$24 instances on $\text{NP}_{\mathbb R}$25 variables and $\text{NP}_{\mathbb R}$26 equations forming a homogeneous simple sub-problem, one passes to $\text{NP}_{\mathbb R}$27 by introducing three new variables $\text{NP}_{\mathbb R}$28 and appending one of the two base gadgets to each existing instance. This yields $\text{NP}_{\mathbb R}$29 total instances, all still sharing Hilbert polynomial $\text{NP}_{\mathbb R}$30 because each block has an isolated solution and the blocks are independent. To enforce primeness, one adds a cycle of mixed quadratic constraints: in each block, one of the four equations is replaced by a quadratic equation involving variables of the next block, with a pattern depending on whether that next block is of type 1 or type 2. According to the source, cycling around all $\text{NP}_{\mathbb R}$31 blocks forces any unit-instance operation that moves from one global instance to another to modify exactly one mixed equation in a pairwise distinct way, thereby enforcing primeness (Çivril, 2021).

After introducing the mixed constraints, not all of the $\text{NP}_{\mathbb R}$32 instances retain the same Hilbert polynomial. One therefore restricts to the balanced instances, those with exactly $\text{NP}_{\mathbb R}$33 blocks of type 1 and $\text{NP}_{\mathbb R}$34 blocks of type 2. These balanced-block instances share an identical Hilbert polynomial, and their number is

$\text{NP}_{\mathbb R}$35

By Stirling’s formula, for every $\text{NP}_{\mathbb R}$36 and large $\text{NP}_{\mathbb R}$37,

$\text{NP}_{\mathbb R}$38

Since each instance uses $\text{NP}_{\mathbb R}$39 variables, the paper derives

$\text{NP}_{\mathbb R}$40

Combining this with the fundamental lemma yields

$\text{NP}_{\mathbb R}$41

hence

$\text{NP}_{\mathbb R}$42

The stated conclusion is that any deterministic real machine requires exponential time to decide quadratic feasibility (Çivril, 2021).

6. Separation results, scope, and conceptual significance

Because $\text{NP}_{\mathbb R}$43 is $\text{NP}_{\mathbb R}$44-complete, an exponential lower bound for deterministic algorithms deciding it implies

$\text{NP}_{\mathbb R}$45

The cited paper states this consequence explicitly for the BCSS model over the reals. Its abstract also states that the feasibility of quadratic systems over $\text{NP}_{\mathbb R}$46, $\text{NP}_{\mathbb R}$47, and $\text{NP}_{\mathbb R}$48 requires exponential time, separating $\text{NP}_{\mathbb R}$49 and $\text{NP}_{\mathbb R}$50 over these fields and rings in the BCSS model (Çivril, 2021).

The broader abstract framework places this real-number result within a larger theory of nondeterminism over structures. Under weak assumptions on $\text{NP}_{\mathbb R}$51, $\text{NP}_{\mathbb R}$52 can be characterized in three equivalent ways: by polynomial-time verification algorithms implemented on $\text{NP}_{\mathbb R}$53-machines, by the $\text{NP}_{\mathbb R}$54-complete problem $\text{NP}_{\mathbb R}$55, and by existential second-order metafinite logic over $\text{NP}_{\mathbb R}$56. The same work extends analogous results to $\text{NP}_{\mathbb R}$57, to the entire polynomial hierarchy over $\text{NP}_{\mathbb R}$58, to its constant-free Boolean counterpart, and to oracle $\text{NP}_{\mathbb R}$59-machines (Kirn et al., 7 Oct 2025).

Two recurrent misconceptions are directly addressed by these results. First, $\text{NP}_{\mathbb R}$60 does not automatically have a complete problem for every structure $\text{NP}_{\mathbb R}$61; infinite-vocabulary structures can fail to admit one. Second, the absence of a complete problem in such cases does not eliminate logical characterization, because existential second-order metafinite logic still captures $\text{NP}_{\mathbb R}$62. A plausible implication is that “R-NP” is best understood not as a single isolated class, but as a family of nondeterministic polynomial-time notions whose machine-theoretic, logical, and complete-problem characterizations depend in a controlled way on the ambient structure $\text{NP}_{\mathbb R}$63 (Kirn et al., 7 Oct 2025).