Heat release rate is the rate at which a process releases energy as heat, commonly used to characterize combustion. It has the dimensions of power and is measured in watts, with $1\mathrm{W}=1\mathrm{J}{\mathrm{s}}^{-1}$. Using $Q_{\mathrm{rel}}(t)$ for cumulative released energy,The heat release rate is the height of the curve; total released energy is its area.

For a synthetic fire whose heat release rate rises linearly from zero to $100\mathrm{kW}$ in $60\mathrm{s}$ and falls linearly to zero over the next $60\mathrm{s}$, the area is a triangle:A second fire with a smaller peak but longer duration can release more total energy. A peak value alone therefore cannot describe the full event.

A fuel-based estimate multiplies fuel mass flow rate by the effective released energy per unit mass, under stated assumptions about composition and combustion. For a deliberately stipulated effective value $20\mathrm{MJ}\mathrm{k}{\mathrm{g}}^{-1}$ and fuel rate $0.02\mathrm{kg}{\mathrm{s}}^{-1}$, the estimate is $400\mathrm{kW}$. In oxygen consumption calorimetry, measured gas flow and oxygen depletion instead provide an estimate of heat release rate. NIST uses independently metered reference burners to compare these two inference routes.^[ref]

A fire's heat release rate differs from the heat transfer received by one nearby surface or from the temperature of its exhaust. Released energy can leave by several paths, while a receiving object can store internal energy as well as lose heat. Even a gas-based calorimeter needs a defined measurement boundary, capture conditions and time response. The NIST National Fire Research Laboratory provides an example of using exhaust collection to quantify large-fire heat release rate.^[ref]

A torus-fixed point of a toric variety is a point left unchanged by every element of its acting algebraic torus. In a normal split toric variety $X_{\Sigma }$ of dimension $n$, the orbit-cone correspondence givesConsequently fixed points correspond exactly to full-dimensional cones. Maximality under containment is not enough if a maximal cone has smaller dimension.

For ${\mathbb{A}}^2$ with coordinatewise scaling, the origin is fixed. Points on a punctured coordinate axis have one-dimensional orbits, while points with both coordinates nonzero lie in the dense two-dimensional orbit. For the positive-dimensional torus itself, multiplication has no fixed point.

The fan consisting only of the two positive coordinate rays and zero describes ${\mathbb{A}}^2\setminus \{0\}$. Both rays are maximal cones, but neither is full-dimensional, so there are no fixed points. This example isolates the dimension condition from the informal idea of a maximal face.^[ref]

A time-changed filtration records the information available after a new clock has reached an old time. Let $({\mathcal{F}}_t)$ be a filtration and let $({\tau }_s)$ be a nondecreasing family of finite stopping times. DefineThe second expression is the stopping-time sigma-algebra. If $s\le u$ and $A\in {\mathcal{F}}_{{\tau }_s}$, thenThus ${\mathcal{G}}_s\subseteq {\mathcal{G}}_u$, as required for a filtration.

For the deterministic clock ${\tau }_s=2s$, the new information is ${\mathcal{F}}_{2s}$, not ${\mathcal{F}}_s$. A continuous adapted process sampled at these stopping times is adapted to the new filtration. Preservation of the local martingale property requires further hypotheses such as those in the optional time-change theorem; merely substituting arbitrary random times does not establish it.

A momentum buffer stores an accumulated history of gradients for an optimization algorithm. In one common convention, a parameter vector ${\theta }_k$, supplied gradient $g_k$, learning rate ${\eta }_k$ and constant coefficient $0\le \mu <1$ obey the recurrence relationThe buffer belongs to the optimizer state: it remembers earlier inputs even when the current parameters are unchanged. This is the undampened momentum convention used by PyTorch stochastic gradient descent with weight decay and the Nesterov accelerated gradient method disabled; its documented first buffer is the first gradient, agreeing with this recurrence from $b_0=0$. Other conventions absorb the learning rate into the stored buffer, so a buffer value needs its update convention to be interpreted.^[ref]

Repeated substitution gives the exponentially weighted gradient historyTo verify this identity, substitute its expression for $b_k$ into $b_{k+1}=\mu b_k+g_k$: multiplication by $\mu$ raises every old exponent by one, and the added $g_k$ is precisely the new exponent-zero term. The formula holds at $k=0$ because its sum is empty, completing the mathematical induction.

For an original scalar example with $b_0=0$ and a constant supplied gradient $g_j=1$, the geometric series yields $b_k=(1-{\mu }^k)/(1-\mu )$. At $\mu =0.9$, $b_{10}\approx 6.5132$ and the limit is 10. The buffer is an unnormalized sum: multiplying it by $1-\mu$ changes its scale, and would require a corresponding learning rate adjustment to preserve these parameter updates.

The same memory explains why momentum can temporarily oppose the current gradient. If $b_k=10$, $\mu =0.9$ and the new scalar gradient is $g_k=-1$, then $b_{k+1}=8$. With a positive learning rate, the parameter still decreases, although the present negative gradient alone would call for an increase. Persistence can carry motion through rapidly alternating gradients, but this example also shows why the rule need not decrease the objective function at every step.

An optimization checkpoint is a saved description of an ongoing optimization algorithm from which its optimizer updates can resume. It combines the current parameters with the relevant optimizer state. The practical criterion is optimizer restart equivalence: after restoring the saved information and supplying the same subsequent inputs, the chosen update rule should produce the same subsequent states.

For a deterministic scalar example, minimize the quadratic function $f(\theta )={\theta }^2/2$ with gradient $g_k={\theta }_k$, learning rate $\eta =0.1$ and momentum buffer recurrence $b_{k+1}=0.9b_k+g_k$, followed by ${\theta }_{k+1}={\theta }_k-0.1b_{k+1}$. Start with $({\theta }_0,b_0)=(1,0)$. After one step the saved pair is $({\theta }_1,b_1)=(0.9,1)$. Restoring both gives $b_2=1.8$ and ${\theta }_2=0.72$. Restoring only ${\theta }_1$ and initializing a new momentum buffer to zero gives $b_2=0.9$ and ${\theta }_2=0.81$. Equal saved parameters can therefore lead to unequal continuations. These numbers are an original calculation for the declared recurrence.

In PyTorch, a general training checkpoint commonly serializes separate model and optimizer dictionaries, together with the training position. The model dictionary contains learned parameters and registered model buffers; the optimizer dictionary contains its own changing state and group settings.^[ref] Restoring the optimizer state-to-parameter association also matters: the optimizer loader pairs stored identifiers with parameters in group order, without independently establishing that they denote the same mathematical parameters.^[ref]

The required checkpoint boundary depends on the computation being resumed. A changing learning rate schedule, the next position in the training data, and the state of a pseudorandom number generator can influence future updates. Matching them is part of computational reproducibility; a saved parameter vector alone does not specify the whole training process.

Optical contrast in decibels expresses a specified ratio logarithmically. For peak-to-local radiant power contrast $C=P_{\max }/P(t)>0$,The inverse convention $r=P(t)/P_{\max }$ gives $10{\log }_{10}r=-L$. State the convention and the time of the weak feature. A $60\mathrm{dB}$ peak-to-feature contrast means one million to one in power, not in field amplitude.

An original readout example uses a linear photodetector whose baseline-corrected output is proportional to optical power. A main reading of $0.80\mathrm{V}$ and weak-feature reading of $80\mu \mathrm{V}$ imply $C=10^4$ and $L=40\mathrm{dB}$. Applying $20{\log }_{10}$ to those readings would incorrectly report $80\mathrm{dB}$: the electrical signal represents power, not optical wave amplitude.

If instead a measurement is proportional to an optical field amplitude $A$, with the same impedance and spatial normalization for both fields, power is proportional to $|A{|}^2$. ThenA field-amplitude ratio of $100$ and a power ratio of $10^4$ therefore express the same $40\mathrm{dB}$ contrast. The correct multiplier follows the measurement model, not whether the readout unit happens to be volts.

Contrast does not specify pedestal energy fraction, optical prepulse duration or delay coverage. The Sequoia primary sheet specifies dynamic range, scan range and optical temporal resolution separately.^[ref] A signal beneath a known measurement floor supports a contrast bound within that window; subtracting an uncertain background does not establish unlimited contrast.

In online learning, cumulative regret compares the total loss of sequential choices $a_t$ with the best fixed action that could have been chosen in hindsight:The comparator uses one action for every round; comparing with a different best action at each round defines a stronger benchmark. For two actions whose losses over three rounds are $(0,1,0)$ and $(1,0,1)$, always choosing the second has loss two, while the best fixed action has loss one, giving regret one.

A bound $R_T=o(T)$ means average regret tends to zero. It does not promise zero loss or identify a true data-generating probability distribution; even adversarial loss sequences can be studied this way. In a multi-armed bandit, only the chosen action's reward is observed, making information gathering part of the decision problem.

Finite-time ambiguity between stiffness and slow filling means that a small volume response at one time can come from a stiff compartment or from a more compliant compartment that has not finished filling. In a passive single-compartment pressure-step model,One observation supplies one equation for equilibrium lung compliance $C$ and time constant $\tau$. It does not generally identify both.

An original family makes the ambiguity exact. For an observed $A(T)=A>0$, choose any $C>A$ and setEvery choice reproduces the observation. At $T=1$ s and stipulated $A=0.10\mathrm{L}(\mathrm{cm}{\mathrm{H}}_2\mathrm{O}{)}^{-1}$, equilibrium compliances 0.12, 0.20 and 1.00 correspond to time constants about 0.558, 1.443 and 9.491 s. Their agreement at one time does not imply agreement about equilibrium volume response.

Later measurements distinguish these particular models. At $2T$, the predicted response isThe three examples predict approximately 0.1167, 0.1500 and 0.1900 in compliance units. Those differences motivate observing more than one stage of the transient. They do not prove that a one-compartment model is adequate for an actual lung.

Early-time resistance dominance of a filling response explains the limiting behavior. Expanding the exponential for $T\ll RC$ givesThe leading response depends on resistance rather than equilibrium compliance. As $C$ grows in the indistinguishable family, $\tau \sim TC/A$ and $R=\tau /C$ tends to $T/A$. Much larger equilibrium compliance can therefore be hidden behind similar early filling.

The primary respiratory pilot study discusses estimates based on expiratory time constants and why deflation need not be uniexponential.^[ref] An experiment should constrain its pressure step, observation duration and model class. A fitted finite-time response is not automatically an identified equilibrium material property.

The effective mass in a normal-contact penalty mode is the inertia seen by one specified relative normal displacement, rather than automatically the total mass of either body. A contact away from a center of mass can excite rotation as well as translation, changing that inertia and the normal contact penalty time scale.

Use independent generalized velocities $v$, a positive-definite mass matrix $M$, and a normal gap $g$ whose local velocity is $\dot{g}=Jv$. A scalar normal force $\lambda$ produces generalized force $J^{\mathsf{T}}\lambda$. The primary multibody formulation gives this contact-Jacobian velocity/force pairing explicitly.^[ref] At a frozen configuration, the part of gap acceleration caused by that force isIf the scalar $a=J{M}^{-1}{J}^{\mathsf{T}}$ is positive, defineThe definition uses a physical normal gap measured in metres and a matching normal force measured in newtons. Arbitrarily rescaling the gap coordinate changes the associated inertia. Constraints must first be accounted for by independent coordinates or the appropriate constrained response.

A short normal impulse $P$ gives $\Delta v=M^{-1}{J}^{\mathsf{T}}P$, hence $\Delta \dot{g}=P/m_{\mathrm{eff}}$. This supplies a direct interpretation: less effective mass means a larger relative normal-velocity change for the same impulse. At a fixed local configuration, a compressed linear penalty spring of stiffness $k$ gives the small-motion estimateIt omits changing contact geometry, other driving forces and damping; the source separately explains why stiff compliant contact challenges numerical integration.^[ref]

For two freely translating particles, the relative gap sees $a=1/m_1+1/m_2$, so $m_{\mathrm{eff}}=m_1{m}_2/(m_1+m_2)$. This is the reduced mass. Equal $2\mathrm{kg}$ particles give $m_{\mathrm{eff}}=1\mathrm{kg}$; replacing one by a fixed wall gives $2\mathrm{kg}$ instead.

For a planar rigid body above a fixed plane, let its mass be $m$, rotational moment of inertia about its center be $I$, and the contact normal be vertical. A contact a horizontal distance $r$ from the center hasFor $m=2\mathrm{kg}$, $I=0.020\mathrm{kg}{\mathrm{m}}^2$ and $r=0.10\mathrm{m}$, $m_{\mathrm{eff}}=1\mathrm{kg}$. At the center it is $2\mathrm{kg}$. With $k=10^{5}N/m$, the off-center angular frequency is $316.2rad/s$, compared with $223.6rad/s$ at the center. Using the full mass in both cases misses this rotational response.

Several simultaneous contacts require more than separate scalar replacements. Their matrix $A=J{M}^{-1}{J}^{\mathsf{T}}$ generally has off-diagonal terms: a force at one contact changes another contact’s acceleration. With frozen geometry and positive diagonal stiffness matrix $K$, compressed gap perturbations satisfy $\ddot{\xi }+AK\xi =0$; its nonzero modal squared frequencies are those of the symmetric matrix $K^{1/2}A{K}^{1/2}$. Coupled normal-contact penalty modes can therefore be faster than estimates treating each contact independently. A proposed verification applies a small known impulse or displacement perturbation at a stated configuration and checks relative contact motion, followed by a timestep convergence study.

For two configurations $\eta ,\xi \in \{0,1{\}}^V$, the weighted Hamming distance for the noisy voter model is $D(\eta ,\xi )={\sum }_v\pi (v)1_{\{\eta (v)\ne \xi (v)\}}$, where $\pi$ is the stationary distribution of the copying walk. On a connected undirected graph, $\pi (v)=\mathrm{deg}(v)/(2|E|)$. Positive weights make this a metric, with values between zero and one.

Couple two chains using the same uniformly selected vertex, refresh decision, fresh bit and copied neighbor. If the refresh probability is $\epsilon$, stationarity $\pi P=\pi$ gives $\mathbb{E}[D_{t+1}\mid {\eta }_t,{\xi }_t]=(1-\epsilon /|V|)D_t$. Copying redistributes disagreements without increasing their stationary-weighted total in expectation; refreshes remove them. The factor $|V|$ belongs to the single-vertex update convention, not a simultaneous update of every vertex.^[ref]

Correlation of neighboring periodogram powers describes dependence between squared spectral magnitudes estimated from the same random record. Their dependence is not determined by the sign of amplitude covariance. A negative correlation between complex Fourier amplitudes can produce a positive correlation between their powers, changing the precision obtained by averaging nearby spectral values.

Let $Z_k$ be zero-mean jointly Gaussian complex Fourier estimates. Define their complex covariance and nonconjugated second moment byand put $I_k=|Z_k{|}^2$. Applying Isserlis's theorem to the four factors gives three pairings:Subtracting the product of the means therefore yieldsIn particular, Gaussian power covariance is nonnegative even when a complex amplitude covariance is negative. Also $\mathrm{Var}(I_k)=C_{kk}^2+|P_{kk}{|}^2$; ignoring the pseudo-covariance of a complex random variable changes the answer when that term is present.

Consider independent sample noise with independent Gaussian real and imaginary components of equal variance. Its nonconjugated moments vanish, so $P=0$ after any deterministic complex linear transform. If the normalized bin variances are all $V$, thenFor the periodic Hann window with $N\ge 5$ on the original $N$-point Fourier grid, noise covariance between windowed Fourier bins gives amplitude covariance ratios $-2/3$ at separation one and $1/6$ at separation two. The corresponding power correlations are $4/9$ and $1/36$. More distant nonconjugate bins have zero covariance in that particular model.

An original calculation averages $M$ consecutive powers without wrapping around the Fourier grid. Assume $N\ge M+2$ so that the block does not acquire additional cyclic neighbor pairs. Counting the $M-1$ adjacent pairs and $\max (M-2,0)$ second-neighbor pairs givesFor $M=2$, the variance is $13V^2/18$, rather than the independent-value result $V^2/2$. For $M=10$, it is $0.18444V^2$, rather than $0.10V^2$. A variance-equivalent independent-power count is the reciprocal of this normalized variance: about $5.42$ for those ten bins. For $M\ge 2$, the bracket simplifies to $35M/18-1$, approaching roughly $18M/35$ independent powers for large blocks in sufficiently long records.

Real-valued Gaussian input has conjugate Fourier symmetry and can have nonzero $P_{kl}$, especially for mirrored frequencies and endpoint bins. A deterministic spectral component also invalidates the zero-mean calculation unless its additional mean-dependent terms are included. Colored noise, taper convention and detrending change $C$ and $P$; the formulas above do not make all measured spectra follow one correlation pattern.

A periodogram may be expressed as a squared spectrum or a power spectral density. SciPy documents their window-dependent conversion, one-sided folding and optional detrending explicitly.^[ref] A common positive scaling does not change the correlation coefficient, but combining unequal scalings or folding conjugate values requires carrying those transformations through the covariance.

A proposed check generates independent records under the stated noise model, calculates bin powers with the same window, and compares empirical pair covariances and block-average variance with the predictions. Independent records provide a different averaging resource from additional bins or zero-padded frequency points in one record. Report which values were averaged and their covariance before assigning an independent-sample precision gain.

Long short-term memory, usually abbreviated LSTM, is a recurrent neural network architecture that carries an explicit memory state through a sequence and learns when to write, retain and expose its contents. It addresses the vanishing gradient problem: in ordinary recurrence, repeated multiplication by small derivatives can make a distant input have almost no effect on the gradient used for learning. Sepp Hochreiter and Jürgen Schmidhuber's 1997 architecture provided a memory-cell path designed to preserve error flow, with gates controlling access.^[ref]

A common simplified modern form, omitting peephole connections, isHere $x_t$ is the input, $c_t$ the memory, $h_t$ the exposed state, $\sigma$ the logistic sigmoid function, $\tanh$ the hyperbolic tangent, and $\odot$ componentwise multiplication. The learned matrices and biases determine an input gate in long short-term memory $i_t$, a forget gate in long short-term memory $f_t$, and an output gate in long short-term memory $o_t$. LSTM: A Search Space Odyssey by Klaus Greff and others gives a more general baseline with peephole connections; the equations above deliberately show the simpler variant.^[ref]

For a scalar memory with writing switched off in the idealized limit and constant retention $f$, $c_n=f^n{c}_0$. Over 100 steps, $f=0.99$ retains about $0.366c_0$, while $f=0.999$ retains about $0.905c_0$. This illustrates how a learned gate can select a retention timescale. Exactly zero or one are limiting idealizations for a finite-input logistic sigmoid function. The product of forget gates describes the direct memory path; it is not the entire recurrent derivative, because gates also depend on earlier states.

The forget gate in long short-term memory was introduced after the original architecture, in Learning to Forget: Continual Prediction with LSTM by Felix A. Gers, Jürgen Schmidhuber and Fred Cummins, to allow old contents to be reset during continuing input streams. Training still uses methods such as backpropagation through time. Artificial delayed-dependency tasks in the original paper demonstrated learning across more than 1,000 time steps in particular experiments; this does not imply unlimited memory or guaranteed success on every long sequence.^[ref][ref]

A neural network logit is a real-valued class score produced before a probability transformation such as the softmax function. The vector of scores determines the class probabilities jointly; an isolated score is not itself a probability. Negative scores and scores larger than one are permitted.^[ref]

For scores $z_1,\dots ,z_K$, the softmax function givesAdding one common scalar $a$ to every coordinate leaves all probabilities unchanged:Consequently the absolute offset of a logit vector cannot be recovered from these probabilities.

For example, scores $(2,0)$ and $(7,5)$ both give first-class probability $e^2/(e^2+1)\simeq 0.8808$. The difference between the two scores, rather than the first coordinate alone, determines this binary probability. Indeed,This connects the binary score difference with the logit link function. It does not make every coordinate of a multiclass score vector a separately defined binary log-odds.

A common shift also differs from rescaling. Dividing $(2,0)$ by two gives $(1,0)$ and first-class probability $e/(e+1)\simeq 0.7311$. This is the transformation used in temperature scaling with temperature two. Positive scaling preserves the selected class within an example while changing confidence.^[ref]

When interpreting a neural network logit, state its score vector and probability transformation. Comparing raw coordinates from independently shifted parameterizations can otherwise suggest a confidence change where the actual probability vector is identical.

A Mycielski graph commonly means a member of the family obtained by repeatedly applying the Mycielski construction to $K_2$. Writing $M_2=K_2$ and $M_{k+1}=\mu (M_k)$, its chromatic number is $k$, although it has no triangle. More broadly, the term is sometimes used for a graph produced by that construction from another starting graph; the chosen family convention should be stated.

The first examples are $M_2=K_2$, $M_3=C_5$, and the eleven-vertex $M_4$. Each construction adds a shadow for every old vertex and one apex, so the vertex counts satisfy $n_{k+1}=2n_k+1$. Hence $n_k=3\cdot 2^{k-2}-1$.

These graphs demonstrate that forbidding triangles in a graph does not bound the number of colors needed. A large clique is only one obstruction to coloring: constraints spread through a triangle-free graph can force arbitrarily large chromatic number.

The polynomial model of a regularity structure realizes abstract monomials as ordinary coordinate differences around each base point. It is the example that turns a regularity structure into familiar Taylor polynomials. In $d$ dimensions, use a polynomial ring in symbols $X_1,\dots ,X_d$, graded by total degree. For a multi-index $k=(k_1,\dots ,k_d)$,Re-expansion uses ${\Gamma }_{xy}p(X)=p(X+x-y)$. Substitution immediately gives ${\Pi }_x{\Gamma }_{xy}={\Pi }_y$ and ${\Gamma }_{xy}{\Gamma }_{yz}={\Gamma }_{xz}$, the compatibility identities of a model of a regularity structure.^[ref]

For example, $f(z_1,z_2)=z_1{z}_2$ has local abstract expansion $a_x=2+2X_1+X_2+X_1{X}_2$ at $x=(1,2)$. At $y=(3,4)$ it has $a_y=12+4X_1+3X_2+X_1{X}_2$. Transporting the latter substitutes $X_1-2,X_2-2$:The coefficients change because the expansion point changes; the represented function is still $z_1{z}_2$.

The analytic scaling also follows by calculation. For a smooth test function and ${\phi }_x^{\lambda }(z)={\lambda }^{-d}\phi ((z-x)/\lambda )$, changing variables givesThus the assigned degree predicts exactly the power of $\lambda$ in the model bound. Anisotropic scaling can assign different weights to different coordinates instead; ordinary total degree is the isotropic choice. General regularity structures retain this organization of local expansions while admitting symbols that are not polynomials.

Batch size counts the examples combined in a mini-batch. In the simple stochastic gradient descent loop where each group produces one parameter update, it controls how many examples contribute to that update. It differs from a training epoch, which counts a pass over the available training data.^[ref]

For an original counting example, partition ten examples with nominal batch size $b=4$ and retain the remainder. The actual sizes are $4,4,2$, giving $\lceil 10/4\rceil =3$ updates per training epoch. Discarding the remainder instead gives two updates and uses only eight examples. More generally the retained-remainder update count is $\lceil n/b\rceil$ for $n$ examples, provided there is one update per group.

Batch size also changes gradient sampling noise. If independently sampled component gradients at a fixed parameter have scalar variance ${\sigma }^2$, their sample mean hasFor an original exact example, suppose one component gradient is equally likely to be 0 or 2. Its mean is 1 and variance is 1. A two-example mini-batch gradient takes values $0,1,2$ with probabilities $1/4,1/2,1/4$, giving variance $1/2$. This calculation holds parameters fixed and assumes independent sampling; sampling without replacement has a finite population correction.

With gradient accumulation, several groups may contribute before an update. Two full four-example groups then involve eight examples per update if their gradients are normalized consistently. State whether a reported size counts the processed group or all examples contributing to an update. Lower sampling variance alone does not establish faster training for a fixed computation budget.

An air-displacement pipette uses a piston to change the pressure and volume of an air cushion above a liquid sample. On piston withdrawal, reduced pressure allows ambient pressure to drive liquid into the tip; subsequent piston motion expels it. The piston itself remains separated from the sample.^[ref]

The air cushion explains why a mechanical setting alone does not establish the delivered volume. Its response can depend on temperature and sample properties, including volatility and density. A positive-displacement pipette uses direct piston contact instead of that intervening gas volume.^[ref]

For an original repeated-delivery example, compare sets of $100\mu \mathrm{L}$ settings with measured mean deliveries of $98\mu \mathrm{L}$ and $100\mu \mathrm{L}$. Knowing those means alone does not identify which set has less scatter. Both the mean error and a dispersion measure such as standard deviation are needed to describe performance. A microlitre is an exact unit relationship; an individual micropipette delivery is an experimental measurement.

A positive-displacement pipette transfers liquid with a piston that contacts the sample directly, avoiding the intermediate air cushion of an air-displacement pipette. In the handheld designs described by the manufacturer, the capillary and piston form a disposable assembly. The mechanism can reduce effects associated with the expansion or contraction of trapped air.^[ref]

This distinction is useful when handling liquids whose viscosity, density or volatility differs substantially from water. The instrument still measures a volume: a selected $25\mu \mathrm{L}$ does not imply a fixed mass for every liquid. For an illustrative density of $1.20mg/\mu \mathrm{L}$, the corresponding mass is $30\mathrm{mg}$ by $m=\rho V$.

A gravimetric check therefore needs the appropriate density conversion. Direct displacement also does not make every delivery perfect: instrument condition, the fitted assembly and the test procedure remain relevant. Repeated measurements distinguish a systematic displacement of the mean from scatter in individual deliveries.^[ref]

A microbial cryoprotectant is a substance added to help viable microorganisms withstand freezing and subsequent recovery during microbial cryopreservation. Protection is assessed in the preserved material; the label does not mean that a substance is harmless at every concentration or suitable for every strain.

Freezing changes both the amount of liquid water and the concentration of dissolved substances. Intracellular ice can injure structures, while water movement by osmosis can shrink a cell. ATCC identifies glycerol and dimethyl sulfoxide as common protective agents and stresses that useful formulations depend on the material. Protecting a frozen preparation involves more than simply lowering its nominal temperature.^[ref]

An original comparison measures recovery rather than assuming protection from composition. Split a benign yeast preparation into two matched portions. Suppose each has1000 initial colony-forming units per original-volume equivalent, while recovery gives300 units without a candidate additive and700 with it. The recovered fractions are0.30 and0.70: a difference of0.40, or40 percentage points, and a ratio $700/300\approx 2.33$. These invented measurements would support improved recovery in that comparison, not a universal survival rate.

Equal original-volume equivalents matter because adding a solution also dilutes the cells. A matched nonfrozen comparison can help separate freezing injury from an additive's effect on the growth readout. Recovery of colony-forming units, unchanged genotype and preservation of a chosen phenotype are separate questions.

The practical motivation is to retain a reproducible biological starting point between experiments. A microbial cryoprotectant is one part of that preservation system, alongside the physical freezing history, storage and subsequent authentication.^[ref]

The Metropolis algorithm is the symmetric-proposal case of the Metropolis–Hastings algorithm. At state $x$ with positive target density $\pi (x)$, propose $y$ from $q(x,\cdot )$ with $q(x,y)=q(y,x)$, and accept with probabilityOtherwise remain at $x$. A normalizing constant in $\pi$ cancels, which makes the method useful when a target distribution is known only up to scale.

The accepted probability flow is $\pi (x)q(x,y)\alpha (x,y)=q(x,y)\min \{\pi (x),\pi (y)\}$, symmetric in $x,y$. This proves detailed balance and hence invariance of the target under the usual kernel assumptions.

For two states with target masses $1/4,3/4$ and an always-flip proposal, moves from the first state are always accepted, while reverse moves are accepted with probability $1/3$. The rejection probability creates holding at the second state. Target invariance alone does not prove convergence from every start; irreducibility and appropriate periodicity conditions matter. An asymmetric proposal requires the Hastings proposal-density correction.