Quick Reference: Solutions

The qbpp::Sol class represents a solution to a QUBO problem. It stores variable assignments (as a packed bit array) along with the energy value and time-to-solution.

Creating a Sol

Expression Description
Sol(expr) Create a solution for expression expr with every variable set to 0 (a native integer variable is set to its lower bound)

Evaluating Variables, Terms, and Expressions

Given a solution sol, variable values and expression results can be obtained using two equivalent calling conventions:

Expression Equivalent Return Type Description
sol(x) x(sol) energy_t Evaluate Var x (returns 0 or 1)
sol.get(x) sol(x) coeff_t Get the value (0 or 1) of Var x; for a native integer variable, its integer value (energy_t)
sol(vi) vi(sol) energy_t Evaluate an integer variable vi (integer value)
sol(t) t(sol) energy_t Evaluate Term t
sol(f) f(sol) energy_t Evaluate Expr f
sol(c) c(sol) energy_t Evaluate the penalty of a constraint expression c
sol(arr) arr(sol) Array<coeff_t> Evaluate an array of variables (Var), of native integer variables, or of expressions (Term, Expr) (same shape)

Both sol(x) and x(sol) produce the same result. The x(sol) form is convenient for use in array contexts: for a Var array x of size $n \times m$, x(sol) returns an Array<coeff_t> of size $n \times m$ containing the assigned values (0 or 1). An array of expressions (Array<Dim, Expr>) is also evaluated by sol(arr) / arr(sol), which returns an Array<coeff_t> of the same shape (a sub-array arr[i] works too). An element whose value does not fit coeff_t throws std::overflow_error; evaluate such elements one at a time with arr[i](sol) (energy_t).

Reading a variable that is not in the solution’s model (the expression given to the solver), or evaluating an expression that contains one, is an error (a std::runtime_error exception). A variable whose value was fixed with replace is not in the solution of the fixed expression: restore a solution of the original expression f with Sol(f).set(sol, ml) before reading it. sol.has(x) reports whether a variable is in the model.

Setting Variable Values

Expression Description
sol.set(x, value) Set variable x to value (0 or 1, or an integer for a native integer variable)
sol.flip(x) Flip the value of binary variable x (0 ↔ 1); the stored energy becomes invalid. Not available for native integer variables
sol.set(other) Copy all variable values from another Sol
sol.set(ml) Set variable values from a MapList
sol.set(other, ml) Copy from other, then apply MapList ml

A MapList is a list of (Var, value) or (Expr, value) pairs (where the Expr is an integer variable):

qbpp::MapList ml = {{x[0], 1}, {x[1], 0}, {vi, 5}};
sol.set(ml);

set() returns Sol&. To build a solution of the original expression f from a solution sol of an expression whose variables were fixed with replace, use set(sol, ml):

auto full_sol = qbpp::Sol(f).set(sol, ml);

The values in ml may also be expressions such as ~x[1]; they are evaluated with the variable values of sol. In set(ml) without a solution, the values must be constants: an expression throws std::invalid_argument.

Energy and Evaluation

Expression Return Type Description
sol.energy() energy_t Return the stored energy value
sol.comp_energy() energy_t Return the energy (recomputed from current variable values only when invalid)
sol.tts() double Time-to-solution (seconds)

sol.energy() returns the energy value that was stored when the solver found the solution. It does not recompute the energy. After calling sol.set() or sol.flip() to modify variable values, the stored energy becomes invalid. Calling sol.energy() in this state throws an error. Call sol.comp_energy() to recompute and update the energy before accessing it. If the energy is already valid, comp_energy() returns the cached value without recomputation.

Extracting Integers from Solutions

onehot_to_int() — One-Hot Decoding

Expression Return Type Description
onehot_to_int(sol(x)) Array<coeff_t> Decode one-hot along last axis (default)
onehot_to_int(sol(x), k) Array<coeff_t> Decode one-hot along axis $k$

Decodes along the specified axis and returns an array with one fewer dimension. For a one-dimensional array, it returns a coeff_t scalar (the position of the 1) instead of an array (no axis is given). The output shape is the input shape with axis $k$ removed, and each element is the index of the 1 along that axis. Negative indices are supported (e.g., -1 = last axis). Returns $-1$ for slices that are not valid one-hot vectors (i.e., do not contain exactly one 1).

For more details and examples, see Extracting Integers.

Solver Info

The solver result classes (EasySolverSol, ExhaustiveSolverSol, ABS3SolverSol) inherit from Sol and provide additional information via info().

Expression Return Type Description
sol.info() const KeyValueVector& Key-value pairs of solver information
sol.info().get(key) const char* Value for a key (nullptr if absent)
sol.sols const std::vector<Sol>& All collected solutions
sol.size() size_t Number of collected solutions
sol.sols[i] const Sol& Access the $i$-th solution

The info() object contains solver metadata as string key-value pairs. Representative keys include:

Key Description
"flip_count" Total number of variable flips performed ("total_flip_count" for ABS3Solver)
"var_count" Number of variables in the model
"term_count" Number of terms in the model
"version" QUBO++ version
"cpu_name" CPU model name
"hostname" Machine hostname

The keys returned differ by solver. For example, ExhaustiveSolver does not return flip_count, version, cpu_name, or hostname (nor var_count and term_count for a model with native integer variables). info().get(key) returns nullptr for an absent key. You can iterate over all entries or access by key:

for (const auto& kv : sol.info()) {
  std::cout << kv.key << " = " << kv.value << std::endl;
}
std::cout << "var_count = " << sol.info().get("var_count") << std::endl;

Printing

Sol objects can be printed with std::cout:

std::cout << sol << std::endl;

This outputs a human-readable representation of the solution including variable assignments. For examples, see Quick Reference: Printing Objects.


Back to top

Page last modified: 2026.10.15.

© 2026 Koji Nakano, Hiroshima University